UNIVERSIDADE FEDERAL DE MATO GROSSO
FACULDADE DE AGRONOMIA E MEDICINA VETERINÁRIA Programa de Pós-Graduação em Agricultura Tropical
PADRÕES ESPAÇO-TEMPORAIS DE COMUNIDADES DE PLANTAS NO PANTANAL MATO-GROSSENSE:
GEOESTATÍSTICA E MODELAGEM BASEADA EM PROCESSOS DE EFEITOS AMBIENTAIS E INTERAÇÃO
ESPACIAL
JULIA ARIEIRA
CUIABÁ MATO GROSSO – BRASIL
2010
2
2
UNIVERSIDADE FEDERAL DE MATO GROSSO
FACULDADE DE AGRONOMIA E MEDICINA VETERINÁRIA Programa de Pós-Graduação em Agricultura Tropical
PADRÕES ESPACO-TEMPORAIS DE COMUNIDADES DE
PLANTAS NO PANTANAL MATOGROSSENSE:
GEOESTATÍSTICA E MODELAGEM BASEADA EM
PROCESSOS DE EFEITOS AMBIENTAIS E INTERAÇÃO
ESPACIAL
JULIA ARIEIRA
Bióloga
Orientador: Prof. Dr. EDUARDO GUIMARÃES COUTO
Tese apresentada à Universidade Federal de Mato Grosso, como parte das
exigências do Programa de Pós- Graduação em Agricultura Tropical, para obtenção
do título de Doutor em Agricultura Tropical
CUIABÁ MATO GROSSO – BRASIL
2010
3
3
FICHA CATALOGRÁFICA A698p Arieira, Julia Padrões espaço-temporais de comunidades de plantas no
Pantanal mato-grossense: geoestatística e modelagem baseada em processos de efeitos ambientais e interação espacial / Julia Arieira. – 2010.
15 f. : il. ; color. ; 30 cm. Orientador: Prof. Dr. Eduardo Guimarães Couto.
Co-orientadora: Profª. Drª. Cátia Nunes da Cunha. Co-orientador: Prof. Dr. Derek Karssenberg. Tese (doutorado) – Universidade Federal de Mato Grosso, Faculdade de Agronomia e Medicina Veterinária, Pós-gradua-ção em Agricultura Tropical, 2009. Bibliografia: f. 137-157 Prof. Dr. Derek Karssenberg. 1. Ecologia da paisagem – Pantanal mato-grossense. 2. Plantas – Pantanal – Geoestatística. 3. Vegetação – Pantanal – Padrões espaço-temporais. 4. Flora – Botânica. 5. Plantas – Pantanal mato-grossense. I. Título.
CDU – 504.54(817.2:252.6) Ficha elaborada por: Rosângela Aparecida Vicente Söhn – CRB-1/931
4
4
UNIVERSIDADE FEDERAL DE MATO GROSSO
FACULDADE DE AGRONOMIA E MEDICINA VETERINÁRIA PROGRAMA DE PÓS-GRADUAÇÃO EM AGRICULTURA TROPICAL
CERTIFICADO DE APROVAÇÃO
Título: PADRÕES ESPAÇO-TEMPORAIS DE COMUNIDADES DE PLANTAS NO PANTANAL MATOGROSSENSE: GEOESTATÍSTICA E MODELAGEM BASEADA EM PROCESSOS DE EFEITOS AMBIENTAIS E INTERAÇÃO ESPACIAL
Autora: JULIA ARIEIRA Orientador: Dr. EDUARDO GUIMARÃES COUTO
Avaliada em 26 de fevereiro de 2010.
Comissão Examinadora:
Profª. Cátia Nunes da Cunha. Prof. Derek Karssenberg (Co-Orientadora) (Co-Orientador)
Prof. Wolfgang J. Junk Prof. Peter Zeilhofer
Prof. Eduardo Guimarães Couto (Orientador)
5
5
DEDICATÓRIA
Às Arieira, Zilah e Eliane,
formadoras da minha educação, por seu estímulo constante ao meu crescimento.
Ao Eduardo Barcellos,
meu companheiro de vida e fiel admirador, por seu amor e incentivos.
6
6
AGRADECIMENTOS
Um trabalho de doutorado não se constrói em solidão;
ele resulta da colaboração, direta ou indireta, de muitas pessoas e instituições.
Por isso agradeço:
aos meus orientadores, Eduardo, Cátia e Derek,
pelo carinho e atenção e pelas boas discussões científicas que me proporcionaram;
ao programa de pós-graduação em Agricultura Tropical
da Universidade Federal de Mato Grosso pela oportunidade;
aos pesquisadores da Universidade de Utrecht,
Steven de Jong, Elisabeth Addink e Jon Skøien,
pela enorme colaboração ao desenvolvimento desta tese;
aos integrantes da banca examinadora
pelos comentários construtivos sobre os resultados deste trabalho;
aos técnicos, Hélio, Zezinho, Libério, Joaquim, Rodrigo, Antônio e Nequinho,
e aos alunos da UFMT, Orleans, Joseane, Abílio, Eduardo, dentre outros,
pela ajuda, em tempos felizes e árduos, no trabalho de campo;
às organizações de fomento científico, CAPES e CNPq,
que tornaram possível o desenvolvimento desta tese,
acreditando e incentivando a formação de qualidade de alunos brasileiros;
e ao SESC Pantanal, por autorizar e apoiar o trabalho dentro de seus domínios.
7
7
LISTA DE FIGURAS
1.1 Área de estudo. (A) localização do sitio de estudo na Reserva Particular do
Patrimônio Natural (RPPN) SESC Pantanal, Pantanal Matogrossense,
Mato Grosso; Brasil; (B) imagem multiespectral IKONOS-2 de 4-m de
resolução, cor verdadeira, do sitio estudado adquirida em Outubro de
2003; (C) dados da flutuação media anual do nível de água do rio Cuiabá
registrados em régua fluviométrica localizada à margem do rio, e
precipitação média próximo à Cuiabá entre 1963-2000, Pantanal. Dados
de precipitação do INMET, dado do nível do rio do
DNAEE.....................................................................................................23
2.1 Study site. (A) Natural Reserve SESC Pantanal located at the Pantanal
Matogrossense, Mato Grosso; Brazil; (B) Mean annual water depth
fluctuation of the River Cuiabá (1963-2000) and mean precipitation near
Cuiabá, northern Pantanal. Rainfall data from INMET (National Institute of
Meteorology of Brazil), river level data from DNAEE (National Department
of Waters and Electric Energy of Brazil); (C) Four meter resolution
multispectral IKONOS-2 image, of the study site acquired in October 2003,
true color. White circles are the sampling locations; (D) 90-m Resolution
SRTM (NASA Shuttle Radar Topographic Mission,
http://ww2.jpl.nasa.gov/srtm/) Digital Elevation Model of the study
area.…………………………………………………………..…………….…40
2.2 Flow diagram describing the procedural steps in the analysis of the data.
…………………………………………………………………………………..42
8
8
2.3 Sampling scheme modified from the RAPELD method (see Magnusson et
al., 2005). (A) 23 trails are regularly spaced over the site; (B) Each trail is
placed at the same elevation level and divided in five sampling trails each
of 50 meter length. (C) Sampling along the trail. Herbaceous species: point
samples at regular interval along the trail centre line. Shrubs, large-sized
trees and medium-sized trees: exhaustive sampling in the indicated zone.
The tree size category is based on diameter of the trunk at breast height
(DBH)………………….…………………………………………………………46
2.4 Factor analysis biplots of the axes 1 to 4 on vegetation variables obtained
from 115 sampling locations. Seven clusters (symbols) represent the
vegetation communities found in the studied floodplain. Factor 1 describes
the gradient of tree biomass found in the study site; Factor 2 of herb cover
and richness; Factor 3 of tree dominant species; and Factor 4 of shrub
biomass.………………………………………………………………………..51
2.5 Maps of the kriged estimates of factor scores and the semi-variograms of
the residuals of the regression between factor axes and remotely sensed
and ancillary data; Fitted variogram models: Mat: Matheron family, Exp:
Exponential. Values between brackets are nugget effect, structured
variance and variogram range, respectively. (A) Factor 1; (B) Factor 2; (C)
Factor 3; (D) Factor 4.…………………………………….………………….60
2.6 Maps of the standard deviation of the predicted error resulting from
universal kriging; (A) Factor 1; (B) Factor 2; (C) Factor 3; (D) Factor
4………………………………………………………………………..….. …..62
9
9
2.7 (A) Predicted distribution of the plant communities identified at the study
site. (B) Results of leave-one-sample out cross-validation. Percentage of
predicted classes at sampling locations. Each bar shows the results for
sampling locations with a certain observed class (indicated by the colors at
the bottom of the C panel). (C) Idem, leave-five-out cross-validation. N =
115.………………………………………………….…………………………64
2.8 Results of Monte Carlo simulation, using 1000 random simulations. (A) The
colors on the map indicate the vegetation class with the highest probability
of occurrence at a cell. A color gradient is used to show the value of this
highest probability; (B) Maps of two single random realizations, color scale
is identical to Figure 8…………………..……………………………………68
2.9 (A) Map with average number of days flooded per year at the study site,
calculated over the period 1969-2007; (B) Relationship between flood
duration (days yr-1) and elevation of the soil surface (meter a.s.l) observed
at the 23 study trails; (C) water level fluctuation in the River Cuiabá
between 1969 and 2007. Vertical dotted lines indicate the occurrence of
drier and wetter years.…….....………………………………………………71
2.10 Fraction of occupied sites by the seven identified communities along the
flood duration gradient. Flooding gradient is divided in five flood classes
representing number of flooded months……………………………………72
3.1 Study site. (A) Natural Reserve SESC Pantanal located at the Pantanal
Matogrossense, Mato Grosso; Brazil; (B) Mean annual water depth
fluctuation of the River Cuiabá (1963-2000) and mean precipitation near
Cuiabá, northern Pantanal. Rainfall data from INMET (National Institute of
10
10
Meteorology of Brazil), river level data from DNAEE (National Department
of Waters and Electric Energy of Brazil); (C) Four meter resolution
multispectral IKONOS-2 image, of the study site acquired in October 2003,
true color. White circles are the sampling locations; (D) 90-m Resolution
SRTM (NASA Shuttle Radar Topographic Mission,
http://ww2.jpl.nasa.gov/srtm/) Digital Elevation Model of the study
area.………………………………….………………………………………….83
3.2. (A) Predicted distribution of the plant communities and (B) spatial pattern of
flood duration on the study floodplain, identified at the study by Arieira et
al. (in preparation)……………………………………………………………89
3.3 Conceptual model of vegetation dynamics on Aquatic-Terrestrial
Transitional Zones in the Pantanal Mato-grossense. Successional changes
(solid arrows) occur from an initial herb dominated stage toward tree
dominated stages. Disturbance, such as fire and exceptional flood events
may set back succession to previous stages (arrows in dotted lines).
Transition probabilities among vegetation states ( ( )jip lk ,→ ) and waiting
times before transitions ( kw , in years) vary according with the vegetation
position on wetter (right side values) or drier parts (left side values) of the
flood duration gradient. The strength of the neighboring effect on transition
probabilities is determined by a neighborhood effect parameter term ( m =
18)………………………………………………………………………………93
3.4. (A) Spatial pattern of community distribution identified by Arieira et al. (in
preparation) and (B) resulted from our model. Uncertainty in vegetation
11
11
classification of the map in A is shown in (C), as two maps resulted from
Monte Carlo simulation (see Arieira et al. in preparation)………………..103
3.5. Model calibration. Comparison between original (bar) and modeled (line)
fraction of occupied area (log scale) by each vegetation states at different
classes of flood duration (monthly intervals)……………………………….104
3.6. Comparison of spatial patterns between the original and modeled
distribution of community states. A) total occupied area; B)mean patch
size……………………………………………………………………………105
3.7. Spatio-temporal model behaviour. Frequency of transitions among
vegetation states, each year, over 5000 years (dots) and frequency
distribution of ‘number of neighbors’ of 500 grid cells (bars), in four classes
of flood duration. Flood duration classes were derived from the map in
Figure 3.2B: class 1: 0 to 2 months; class 2: greater than 2 to 4 months;
class 3: greater than 4 to 6 months; class4: greater than 6 months. Mean
frequency of changes among vegetation states is highest at intermediary
flood sites……………………………………………………………………108
3.8. Scenarios illustrating spatial patterns of flood duration on the study site
found in a historical dry year (A; 1971) and in a historical wet year (B;
2006). (C) Water level fluctuation in the River Cuiabá between 1969 and
2007 is provided by Brazilian National Water Agency (ANA;
(http://hidroweb.ana.gov.br).................................................................110
3.9. Vegetation response to shifts in the hydrologic regime (namely, duration of
inundation) in the Pantanal; base realizations. A) spatially homogeneous
12
12
dry scenario (DZ), B) spatially homogeneous wet scenario (WZ).
Hydrological changes begin after 500 yr timesteps…………………….113
3.10. Vegetation response to shifts in the hydrologic regime (namely, duration of
inundation) in the Pantanal; base realizations: A) historical dry scenario
(DY; 1971); (B) historical wet scenario (WY; 2006); base realizations:
Hydrological changes are simulated after 500 yr timesteps……………115
13
13
LISTA DE TABELAS
2.1 Summary statistics for the factor analysis. Numbers in bold highlight the
highest correlation with the factor axis…………….………………………..48
2.2 Structural and floristic characteristics of plant communities, given as mean
and standard deviation.………………………….…………………………...54
2.3 Pearson’s correlation coefficients between factor axes and image variables:
four spectral bands, Normalized Difference Vegetation Index (NDVI),
Principal Component transformation to the IKONOS-2 image (PC), and
canopy topography derived from DEM-SRTM (DEM). * P ≤
0.05…………………………………………………………………….………..57
2.4 Multiple linear regression models relating factor axes scores (F1-4) to
imagery derived variables: four spectral bands (blue, green, red and infra-
red), Normalized Difference Vegetation Index (NDVI), Principal
Component transformation to the IKONOS-2 image (PC), and canopy
topography derived from DEM-SRTM (DEM). R2 is the coefficient of
determination showing the strength of these
relationships.……………..……………………………………………………..58
3.1 Historical life traits of dominant species of the seven successional states
found at the experimental area…………………………………………..….. 95
14
14
SUMÁRIO
1. INTRODUÇÃO GERAL...............................................................................16
1.1. Contexto, definição do problema e objetivos...................................16
1.2. Área de estudo.................................................................................21
1.3. Resumo da tese...............................................................................25
1.4. Thesis summary…………………………………………………………29
2. INTEGRATING FIELD SAMPLING, SPATIAL STATISTICS AND REMOTE SENSING TO MAP FLOODPLAIN VEGETATION IN THE PANTANAL, BRAZIL……………………………………………………………………………33
2.1. Introduction………………………………………………………………35
2.2. Study area…………………………………………………………….....38
2.3. Outline of the approach……………………………………………..….41
2.4. Field data…………………………………………………………..…….43
2.5. Identifying plant communities…………………………………….……47
2.6. Mapping plant communities……………………………………….…...52
2.7. Flood duration-vegetation relationship…………………………….….69
2.8. Discussion……………….……………………………………………....72
2.9. Acknowledgements……………………………………………………..76
3. MODELLING WETLAND VEGETATION DYNAMICS BASED ON SPATIO-TEMPORAL INTERACTION AND FLOODING TOLERANCE OF PLANT COMMUNITIES IN THE PANTANAL MATOGROSSENSE (BRAZIL)…………………………………………………………………….……77
3.1. Introduction……………………………………………………….……..79
3.2. Study area………………………………………………………………83
15
15
3.3. Spatio-temporal Markov chains……………………………………..86
3.4. Model calibration……………………………………………………...101
3.5. Model spatiotemporal behaviour……………………………………106
3.6. Scenarios……………………………………………………………...109
3.7. Discussion ……………………………………………......................116
3.8. Acknowledgements…………………………………………………...120
4. SÍNTESE ………………………………………………………………….......121 4.1. Definição do problema……………………………………….....……121
4.2. Integração de amostragem de campo, sensoriamento remoto e
estatística espacial para predizer distribuição de comunidades de plantas
no Pantanal..........................................................................................123
4.3. Modelando dinâmica de vegetação de área úmida baseada em
interação espaço-temporal e tolerância à inundação ..........................130
4.4. Considerações finais – impactos da pesquisa em futuros estudos e na
conservação dos recursos naturais do Pantanal.................................135
5. LITERATURA CITADA............................................................................137
Curriculum Vitae.........................................................................................158
16
16
CAPÍTULO 1
INTRODUÇÃO GERAL
1.1. Contexto, definição do problema e objetivos Áreas úmidas estão entre as paisagens mais ameaçadas por mudanças
climáticas ou quaisquer mudanças ambientais que alterem o regime hidrológico
do sistema (Keogh et al. 1999, Junk 2002). Consideradas habitats
transacionais entre sistemas aquáticos e terrestres, áreas úmidas englobam
diferentes tipos de habitats como brejos, mangues, florestas ripárias e planícies
de inundação (Mitsch e Gosselink 2000). Diversidade de habitats é um fator
chave da dinâmica destas áreas (Junk et al 2006a), determinando o grau de
conectividade entre organismos e o recurso disponível (Ward e Tockner 2001).
Comunidades de plantas, por sua vez, ditam fortemente esta diversidade,
indicando a existência de certas condições ambientais e interações biológicas
(Tilman 1988). Diferentes tipos de comunidades de planta formam paisagens
sob forma de mosaico, cujas estrutura e composição podem variar
amplamente, no espaço e no tempo (Forman e Godron 1986). Em função da
natureza dinâmica de paisagens, identificar e quantificar padrões espaciais,
assim como ligá-los a processos ecológicos, é uma tarefa imprescindível às
funções de monitoramento, planejamento e conservação ambiental (Metzger
2004). Isto se aplica, principalmente, a áreas úmidas situadas nos trópicos,
onde critérios de classificação e monitoramento são ainda deficientes (Junk e
Piedade 2004).
A relação entre padrões espaciais e processos ecológicos é tema central
em estudos de ecologia de paisagem (Turner 1989). O estabelecimento das
17
17
relações entre a distribuição de espécies, ou grupo de espécies, com
determinada característica espacial, como tamanho e a forma de habitat, torna
factível a criação de modelos preditivos sobre mudanças da paisagem (Metzger
2004). No entanto, a criação de tais modelos é ainda um tema desafiador, já
que complexos e variados padrões podem emergir de interações entre
espécies, e entre estas e seu ambiente multidimensional (Austin e Smith 1989).
Qual o nível de detalhe e quais as informações necessárias para definir
padrões espaciais de uma paisagem? E como estes padrões são influenciados
pelas características físicas e biológicas do ecossistema? As respostas para
estas indagações dependem de estudos, em um nível de detalhe, que
contemplem tanto a heterogeneidade da paisagem, como a homogeneidade
das manchas que a formam. Neste contexto, a escala espacial relacionada ao
detalhe e à amplitude do fenômeno estudado deve estar claramente definida,
na medida em que exerce influência sobre os padrões observados (Leps 1990,
Turner et al. 2001).
Entender os mecanismos ecológicos que determinam padrões de
vegetação tem sido importante objetivo em ecologia, por décadas (e.g.,
Whittaker 1967, Tilman 1988, Connell e Slatyer 1977, Grime1994, Svenning et
al 2004, Gardner e Engelhardt 2008).
Características abióticas da paisagem podem levar espécies de plantas
a se estabelecerem ou evitarem certos habitats, criando paisagens compostas
por certos tipos de comunidades de plantas. Em áreas úmidas, diferenças
locais em nível d’água e duração da inundação costumam determinar quais
espécies germinarão e se desenvolverão num habitat (van der Valk 1981),
18
18
gerando variações espaciais em composição e estrutura de comunidades
vegetais (Casey e Ewel 2006). Embora a inundação seja considerada um dos
principais fatores seletivos ao estabelecimento e desenvolvimento de espécies
de plantas, em áreas úmidas (van der Valk 1981, Parolin et al. 2002, Junk et al.
2006a), muitos outros fatores ambientais devem servir como filtro seletivo.
Atributos do solo (e.g. textura, conteúdo de matéria orgânica), por exemplo, são
capazes de determinar a distribuição espacial de comunidade de plantas
(Burke 2003), apesar de muitos destes atributos serem colineares às variantes
da inundação (Mitsch e Gosselink 2000).
Por outro lado, padrões de distribuição da vegetação podem resultar de
efeitos espaciais relacionados a processos biológicos (Tilman 1994, Tilman e
Kareiva 1997). Interações entre plantas, associadas às diferenças em suas
estratégias de vida, i.e., arbustos, árvores, ervas, são capazes de gerar
heterogeneidade espacial (Greig-Smith 1979). Dependência espacial em
mecanismos de dispersão costuma criar padrão de distribuição agrupado
(Tilman 1994), resultando em diminuição de competição interespecífica e
promovendo coexistência e persistência em longo prazo (Gardner e Engelhardt
2008).
Em paisagens onde características abióticas e interações espaciais em
processos biológicos atuam, conjuntamente, gradientes um tanto complexos
são gerados. Tentativas de entender o papel destas diferentes forças sobre o
padrão de distribuição da vegetação e entender como estas conduzem
comunidades de plantas a diferentes estados e transições são questões
19
19
relevantes para o entendimento da dinâmica da vegetação (Gardner e
Engelhardt 2008).
O presente trabalho foi desenvolvido em uma das maiores áreas úmidas
de planície de inundação do mundo - o Pantanal Mato-grossense (Brasil). Com
cerca de 150.000 km2, o Pantanal está localizada na região central da América
do Sul, estando sua maior área em território Brasileiro, estendendo-se ainda
pelo Paraguai e pela Bolívia. Sujeito a anuais alagamentos de planícies laterais
aos rios que cruzam seu território, o Pantanal apresenta grande diversidade de
habitats e riqueza de espécies (Junk et al. 2002). É considerado um dos
ecossistemas brasileiros em melhor estado de conservação, em conseqüência
da sua localização regional interiorana, da atividade econômica de baixo
impacto (i.e. pecuária extensiva) e das periódicas inundações. No entanto, a
integridade do Pantanal tem sido ameaçada, nas últimas décadas, devido a
interesses econômicos no desenvolvimento regional, através da intensificação
da pecuária e construção de diques de navegação em rios (Da Silva e Girard
2004, Junk et al. 2006a). Tais recentes pressões indicam a necessidade
imediata de criação de diretrizes ecológicas para subsidiar políticas públicas.
Muitos estudos no Pantanal têm-se focado na distribuição de
comunidades de plantas, ao longo de gradientes ambientais (e.g. inundação,
solo) (Prance e Schaller 1982, Zeilhofer e Schessl 1999, Damasceno-Junior et
al. 2005, Arieira e Nunes da Cunha 2006, Nunes da Cunha e Leitão-Filho 2007)
e alguns outros, na dinâmica temporal da vegetação (cf. Silva et al. 1988,
Mauro et al. 1998; Pott 2007). Baseados no conceito de gradientes ou
continuum, tais estudos aplicam-se a espaços ambientais abstratos, não
20
20
provendo informação necessária à construção de modelos espacialmente
explícitos e baseados em processos. Tais modelos são construídos com base
na dependência entre processos ecológicos e padrões espaciais da paisagem
(Austin e Smith 1989).
Recentes progressos, em ecologia de paisagem e em estudo de
comunidades, tornaram possível o desenvolvimento de novas ferramentas para
estudo da vegetação em escalas amplas. Representar sistemas ecológicos
complexos, resultantes de uma rede de processos e atuando em múltiplas
escalas (Turner et al. 2001, Aumann 2007), através de modelos espacialmente
explícitos, envolve a integração de grupos de dados regionalizados (Bascompte
e Solé 1998, Guisan e Zimmermann 2000). A sobreposição destas diferentes
fontes de dados e a maneira como os processos ecológicos são avaliados e
relacionados a padrões espaciais variam amplamente e contam com uma série
de ferramentas de análise e manipulação de dados, tais como sensoriamento
remoto, sistemas de informação geográfica e geoestatística (Burrough e
McDonnell 1998). Tais ferramentas, além de gerarem dados para modelos
quantitativos, ajudam na identificação de características relevantes e
classificadores adequados para mapear diferentes tipos de cobertura do solo e
monitorar mudanças na vegetação (Burrough e McDonnell 1998). O presente
trabalho incorpora essas tendências, no sentido de contribuir para o
entendimento dos padrões espaço-temporais de vegetação de área úmida e na
busca por novas abordagens para se estudar e representar tais padrões.
A presente tese teve quatro principais objetivos:
21
21
1) descrição, classificação e mapeamento de comunidades de plantas, em
uma paisagem no Pantanal Mato-grossense, usando uma abordagem
estatística e integrando dados de vegetação derivados de amostragem
de campo e de imagens de sensoriamento remoto;
2) avaliação da influência da duração da inundação sobre o padrão de
distribuição espacial das comunidades mapeadas;
3) descrição e modelagem da dinâmica da vegetação terrestre de zonas de
transição aquático-terrestres no Pantanal, destacando o papel da
duração da inundação e interação espacial entre comunidades vizinhas,
sobre estados e transições da vegetação;
4) avaliação da influência de cenários de inundação/seca para o Pantanal
sobre os padrões espaço-temporais da vegetação.
1.2. Área de estudo
Caracterização abiótica do Pantanal
O Pantanal Mato-grossense é uma depressão aluvial, localizada no Alto
da Bacia do Paraguai (Ab`Saber 1988), cobrindo cerca de 150.000 km2 da
parte centro-oeste do Brasil e parte da Bolívia e Paraguai (Fig.1.1). De acordo
com o sistema de classificação de Köppen (1948), o clima atual desta região é
Aw, que corresponde a invernos secos e verões chuvosos, com precipitação
anual entre 1000 e 1500 mm. O Pantanal é uma planície de inundação com
altitudes, que variam de 80 a 180 m a.n.m, e declividade do terreno, variando
entre 2-3 cm km-1 N-S e 5-25 cm km-1 O-L (Alvarenga et al. 1984). Solos são
geralmente mal drenados, mostrando variação em conteúdos de argila e areia,
em diferentes posições topográficas no leque aluvial (Assine and Soares 2004).
22
22
Baixa declividade, alta precipitação concentrada durante o verão e solos mal
drenados resultam em inundação temporária da planície de inundação (Junk
1993, BRASIL 1997). Padrões de inundação são, em grande parte,
determinados pelas feições topográficas (Wantzen et al. 2005), criando um
gradiente em duração de inundação que nos permite classificar zonas
aquáticas, terrestres e de transição aquático-terrestre (ATTZ) (Junk et al 1989).
Estas últimas zonas (ATTZ) são importantes partes deste gradiente,
correspondendo a áreas que experimentam estados de seca, na estação seca,
e estados inundados, na estação úmida (Wantzen et al. 2005). Padrões de
precipitação causam esta flutuação anual, resultando em um padrão de
inundação previsível, unimodal e de pequeno alcance (Hamilton et al. 1996).
Apesar dos eventos de inundação serem previsíveis, duração e nível de
inundação podem apresentar padrões pluri-anuais mais úmidos ou mais secos,
como resultado de oscilações climáticas (Collischonn et al. 2001, Junk et al.
2006a).
23
23
Figura 1.1 Área de estudo. (A) localização do sitio de estudo na Reserva
Particular do Patrimônio Natural (RPPN) SESC Pantanal, Pantanal Mato-
grossense, Mato Grosso; Brasil; (B) imagem multiespectral IKONOS-2 de 4-m
de resolução, cor verdadeira, do sitio estudado adquirida em Outubro de 2003;
(C) dados da flutuação media anual do nível de água do rio Cuiabá, registrados
em régua fluviométrica localizada à margem do rio, e precipitação média
BRASIL
Pantanal
N
RPPN SESC Pantanal
Rio São
Lour
enço
ESCALA
Rio
Cui
abá
0 105 Km
Pantanal
MT
MS
16o
21o
55o58o
100200300400500
0N D J F M A M J J A S O
Precipitação, mm
Nivel, cm
Fonte: AN
A/ G
EF/ PNU
MA
/ OEA
A C
B
24
24
próximo à Cuiabá entre 1963-2000, norte do Pantanal. Dados de precipitação
do INMET, dado do nível do rio do DNAEE.
Vegetação do Pantanal
A vegetação do Pantanal possui elementos florísticos de três
importantes domínios morfoclimáticos e fitogeográficos: cerrado, amazônia e
chaco (Ab`Saber 1988). Diferentes formações de cerrado são fitofisionomias
dominantes no Pantanal (67%), mas não são as únicas: floresta semidecidual,
floresta de galeria, brejos, chaco e formações pioneiras, tais como floresta
monodominante de Vochysia divergens Pohl. (Silva et al. 2000) fazem parte do
mosaico de vegetação. A variabilidade em profundidade e duração da
inundação, além de conexões e desconexões estabelecidas entre elementos
da paisagem, através da água da inundação, são as causas preponderantes da
alta diversidade de comunidades biológicas no Pantanal (Junk et al 1989,
Wantzen et al. 2005). Estas causas ditam onde e quando espécies de plantas,
com diferentes estratégias de vida e tolerância à inundação, aparecerão (Junk
et al. 2006a). A influência do uso da terra sobre a vegetação ‘natural’ do
Pantanal ainda é freqüentemente discutida. A pecuária extensiva em campos
nativos do Pantanal, iniciada há cerca de 250 anos, parece não ter afetado,
substancialmente, padrões de distribuição da vegetação no Pantanal (Pott e
Pott 2004, Junk e Nunes da Cunha 2005). No entanto, pressões
governamentais, para intensificação desta atividade nas últimas décadas,
foram responsáveis pela conversão de cerca de 4,5 % da vegetação ‘natural’
25
25
do Pantanal, principalmente florestas, em áreas de pastagem (Silva et al.
1998).
Área experimental
Um sítio com 60 km2 (5 km x 12 km), localizado na região norte do
Pantanal (16o 30’ – 16o 44’S and 56o 20’– 56o 30’W), dentro da Reserva
Particular do Patrimônio Natural (RPPN) SESC Pantanal - Barão de Melgaço,
Mato Grosso, Brasil - foi selecionado através de imagem IKONOS-2, 4 m de
resolução, como área experimental deste trabalho (Fig. 1.1B). O sítio está
localizado em uma ATTZ, sob influência do transbordamento periódico do rio
Cuiabá. Formações vegetais encontradas em diferentes partes do Pantanal,
como floresta aluvial, campos, arbustais e florestas monodominantes,
compõem o mosaico de vegetação, o que torna o sítio selecionado relevante
para estudos da vegetação do Pantanal. Desde a criação da RPPN, em 1998,
esta área tem sido usada com interesses científicos, o que nos permitiu
investigar as características estruturais e funcionais do ecossistema, na
ausência de intensa e freqüente atividade humana. Pesquisas Ecológicas de
Longa Duração (PELD) têm sido conduzidas nesta reserva, há quase uma
década, resultando em acúmulo de informações sobre estados da vegetação e
influências abióticas sobre ela. Estas informações foram de suma importância
para o desenvolvimento da presente tese.
1.3. Resumo da tese Padrões de distribuição da vegetação, no tempo e espaço, são
causados por diferentes forças. Forças externas, como aquelas determinadas
26
26
por distúrbio, e internas, como aquelas determinadas por interações espaciais
entre indivíduos que partilham o mesmo espaço e recurso, influenciam de
maneiras distintas estes padrões. Em áreas úmidas, como o Pantanal Mato-
grossense, estrutura, composição e dinâmica da vegetação são consideradas
fortemente influenciadas pelo regime de inundação. No entanto, apesar de
haver um número significante de estudos, discutindo como a distribuição de
comunidades de plantas do Pantanal está relacionada à inundação, poucos
consideram a influência de processos espaciais e temporais da vegetação,
sobre a estrutura da paisagem e biodiversidade existente. A proteção de
ecossistemas de áreas úmidas, como o Pantanal, necessita da compreensão
dos elementos estruturantes da paisagem e da identificação de métodos
eficientes, para descrevê-los e monitorá-los. Esta tese visou contribuir para o
entendimento de padrões espaço-temporais da vegetação do Pantanal Mato-
grossense e na busca por novas abordagens, para se estudar e representar
tais padrões. O estudo foi realizado dentro de uma área inundável pelo rio
Cuiabá de 60km2 situada na Reserva Particular do Patrimônio Natural do
Serviço Social do Comércio (RPPN SESC Pantanal), município de Barão de
Melgaço, Mato Grosso. A tese é apresentada em quatro capítulos. O primeiro
capítulo apresenta uma introdução geral do trabalho desenvolvido, definindo os
problemas científicos abordados e situando este trabalho no âmbito da ecologia
de plantas e ecologia da paisagem. No segundo capítulo, modelagem preditiva
da vegetação, baseada em técnicas estatísticas sofisticadas, técnicas de
interpolação e propagação de erros, é usada para identificar e mapear
comunidades vegetais na área estudada. Como resultado de análise de fatores
27
27
e técnica de agrupamento com dados de estrutura e composição da vegetação,
sete comunidades de plantas foram identificadas. A relação entre variáveis
derivadas de imagem de sensoriamento remoto IKONOS-2 e de um modelo de
elevação digital SRTM, com os quatro primeiro eixos de fatores, foram
formalizadas matematicamente, usando modelos de regressão linear múltipla.
Os modelos explicaram 70%, 66%, 31% e 26% dos padrões de vegetação
representados pelos quatro eixos de fatores, respectivamente, e foram usados
num procedimento de krigagem universal, para reduzir a incerteza nas
comunidades mapeadas. Procedimentos de cross-validation e simulações de
Monte Carlo quantificaram incertezas no mapa de vegetação produzido por
modelagem. A porcentagem de amostras preditas na classe correta em cross-
validation variou entre 49% a 52%, indicando que densidade de amostras afeta
a acurácea das predições espaciais e, conseqüentemente, do mapa final
produzido. Resultados das simulações de Monte Carlo mostraram que o
padrão espacial geral de distribuição das comunidades, sobre a planície de
inundação estudada, foi predito acuradamente. Comparação entre mapa de
vegetação e de duração da inundação mostrou que há uma preferência das
diferentes comunidades mapeadas a ocupar certas partes do gradiente de
duração de inundação. O mapeamento de comunidade de plantas em extensas
áreas, usando modelagem preditiva da vegetação, como visto neste capítulo,
mostrou-se uma abordagem promissora para conservação e monitoramento
ecológico de longa-duração, no Pantanal, devido à detalhada informação
biológica de amostragem de campo, integrada a dados sensoriados
remotamente, em predições no espaço e no tempo. No capitulo 3 desta tese,
28
28
um modelo de sucessão vegetal para ATTZ foi desenvolvido e testado, através
de Modelo de Cadeia de Markov Espaço-Temporal. O modelo foi construído
com base no papel que duração da inundação e interação espacial, entre
comunidades vizinhas, têm sobre a dinâmica da vegetação. Informações sobre
requerimentos ecológicos e traços de história de vida, em espécies
características das comunidades estudadas, adquiridas da literatura ou
informadas por especialistas, foram utilizadas para determinar os estados de
vegetação e as probabilidades de transição entre estes. Tempos de espera
antes de transições foram incluídos no modelo, simulando o tempo de
desenvolvimento entre um estado de vegetação e outro. A calibração do
modelo foi realizada, através de modelagem inversa, baseada em
comparações entre padrões espaciais da vegetação, observados e simulados,
usando índices da paisagem. O comportamento espaço-temporal do modelo foi
examinado, através da relação entre freqüência de mudanças, em estados de
vegetação ao longo de 5000 anos (interações do modelo), e a distribuição em
freqüência de número de vizinhos, em diferentes posições do gradiente de
inundação. Esta análise mostrou que dinâmica da vegetação e diversidade de
vizinhos variam em função da duração de inundação. O aumento da
diversidade de vizinhos, em posições intermediárias de inundação, refletiu em
mudanças mais freqüentes entre estados de vegetação. Mudanças nos
padrões espaço-temporais da vegetação foram preditas sob influência de
quatro cenários de inundação. Rápidas e substanciais mudanças no padrão da
vegetação se apresentaram como resultado de um cenário, ilustrando uma
paisagem espacialmente homogênea e com período de inundação bastante
29
29
reduzido. Um atraso na resposta das comunidades às mudanças ambientais
refletiu a inércia da vegetação, ligada à sua capacidade de absorver distúrbio.
Modelos de simulação espacialmente explícitos, como o aqui desenvolvido,
configuram-se como ferramentas importantes para interpretação de dinâmicas
complexas, como a da vegetação do Pantanal, e na formulações de novas
hipóteses. O capítulo final da tese (CAPITULO 4) apresenta uma síntese do
trabalho e suas conclusões gerais. As abordagens de modelagem da
vegetação, apresentadas aqui, mostraram-se importantes técnicas de
identificação e descrição de padrões espaço-temporais da vegetação de áreas
úmidas. No entanto, incertezas contidas nos mapas e modelo produzidos
sugerem que informações empíricas, sobre processos ecológicos e padrões
espaciais, continuem sendo adquiridas e monitoradas, a fim de tornar possíveis
predições espaciais mais acuradas e previsões temporais mais robustas, sobre
a vegetação do Pantanal.
1.4. Thesis summary
Patterns of vegetation distribution, in time and space, are caused by different
forces. External forces, such as those determined by disturbance, and internal
forces, such as those determined by spatial interaction among individuals that
share a similar space and resource, influence in different ways these patterns.
In wetlands, like the Pantanal, vegetation structure, composition and dynamics
are considered strongly influenced by flooding. However, in spite of there is a
significant number of studies discussing how distribution of plant communities of
the Pantanal is related to flooding, few studies consider the impact of spatial
and temporal processes on landscape structure and biodiversity. The protection
30
30
of wetland ecosystems, like the Pantanal, needs to count on the understanding
on the structuring features of the landscape, as well as, the identification of
efficient methods to describe and monitor them. This thesis aimed at
contributing for the understanding of space-time vegetation patterns in the
Pantanal and the search for new approaches to study and represent these
patterns. The study was carried out in a 60km2 area at the Cuiabá river
floodplain, situated in the Natural Reserve SESC Pantanal, Barão de Melgaço,
Mato Grosso. The thesis is presented in four chapters. The first chapter gives a
general introduction of the work, defining the scientific problems and positioning
this study in the scope of plant and landscape ecology. In the second chapter,
predictive vegetation modeling, based on sophisticate statistic techniques,
interpolation techniques and error propagation, is used to identify and map
vegetation communities in the studied area. As result of factor analysis and
clustering technique in structural and compositional vegetation data, seven
communities were identified. The relation between variables derived from
remote sensing IKONOS-2 images and a digital elevation model-SRTM, with
the four factor axes were formalized mathematically using multiple linear
regression models. The models explained 70%, 66%, 31% and 26% of the
vegetation patterns represented by the first four factor axes, respectively, and
were used in a proceeding of universal kriging to reduce the uncertainty in the
mapped communities. Cross-validation and Monte Carlo simulation quantified
uncertainties in the produced vegetation map. The percentage of samples
assigned to the correct class in cross-validation was between 49% and 52%,
indicating that sampling density affects spatial prediction accuracy. The
31
31
evaluation of the model using Monte Carlo simulation showed that the overall
spatial pattern of community distribution over the floodplain is predicted
accurately. By comparing the plant community map with a flood duration map, it
was shown that there is a preference of communities to occupy certain positions
at the flooding gradient. Mapping of plant communities across extensive areas
using predictive modeling, as shown in this chapter, is a promising approach for
conservation and long-term ecological monitoring in the Pantanal wetland, due
to the detailed biological information that it is integrated with remotely sensed
data producing a fine scale representation of vegetation spatial patterns over
large areas. In the third chapter of this thesis, a succession vegetation model for
aquatic-terrestrial transition zones of the Pantanal was developed and tested
using Spatiotemporal Markov Chain model. The successional model was
created based on the effect that flood duration and spatial interaction among
neighboring communities have on vegetation dynamics. Information on
ecological requirement and life history traits of characteristic species, acquired
from literature and expert knowledge, were used to determine the vegetation
states and transition probabilities. Waiting times before transitions were also
included in the model, simulating the time spent for one state to develop into
another. The model calibration was performed using inverse modeling, by
comparing observed and simulated spatial patterns using landscape indices.
The space-time model behaviour was examined by relating frequency of
changes among vegetation states over 5000-yr (model iterations) and
frequency distribution of number of neighbors at different position on the
flooding gradient. This analysis showed that vegetation dynamics and neighbor
32
32
diversity varied according with the flood duration condition. The increase in
neighbor diversity at intermediate flood positions reflected more frequent
changes among vegetation states. Changes in space-time vegetation patterns
under four flooding scenarios were forecasted. Quick and substantial changes
in vegetation patterns resulted from a scenario simulating a spatially
homogeneous landscape and with reduced flooding. A delay in community
response to the environmental shifts reflected vegetation inertia, linked to
species capability to absorb disturbance. Spatially explicit simulation models, as
the developed here, help to interpret complex dynamics, such as of the
vegetation of the Pantanal, and formulate new hypothesis. The Chapter 4
shows a synthesis of the work and provides some general conclusions. The
modeling approaches presented in this thesis consist of important techniques to
identify and describe space-time patterns of wetland vegetation. However,
uncertainties in the mapped communities and model outputs suggest that
empirical information on ecological processes and spatial patterns to be still
acquired and monitored in order to generate more accurate spatial and
temporal predictions on the vegetation of the Pantanal.
33
33
CAPÍTULO 2
INTEGRATING FIELD SAMPLING, SPATIAL STATISTICS AND
REMOTE SENSING TO MAP FLOODPLAIN VEGETATION IN THE
PANTANAL, BRAZIL
Com contribuição de: Derek Karssenberg, Steven M. De Jong, Elisabeth A.
Addink, Jon O. Skøien e Cátia Nunes da Cunha
Abstract Wetland ecosystems belong to a type of habitats that are highly
threatened by changes in precipitation and evapotranspiration due to the
preponderant influence of hydrology on structural and functional characteristics
of the ecosystem. New methods and tools to describe, understand, model and
monitor patterns and processes of wetland vegetation are urgently needed. In
this paper, we describe a mapping procedure based on statistical and
geostatistical techniques aiming at identifying and mapping plant communities
in a 60 km2 floodplain landscape in the Pantanal (Brazil), and at investigating
the influence of flooding duration on the community distribution. Seven plant
communities were identified using factor analysis and clustering techniques
performed on vegetation structure and composition data. The relation between
IKONOS-2 remote sensing images and SRTM-digital elevation model and the
first four factors in a factor analysis of vegetation patterns was formalized
mathematically in multiple linear regression models and used in a universal
kriging procedure to reduce the uncertainty in mapped communities. Image
derived variables explained 70%, 66%, 31% and 26% of the first four factors in
34
34
a factor analysis of vegetation patterns, respectively. Cross validation
procedures and Monte Carlo simulations were used to quantify the uncertainty
in the resulting map. The percentage of samples assigned to the correct class in
cross-validation was between 49 % and 52 %, indicating that sampling density
affects spatial prediction accuracy. The evaluation of the model using Monte
Carlo simulation showed that the overall spatial pattern of community
distribution over the floodplain is predicted accurately. By comparing the
resulting plant community map with a flood duration map, we showed a
significant relationship between plant community distribution and flooding
duration. Mapping of plant communities across extensive areas using predictive
modeling, as shown in this study, is a promising approach for conservation
assessment and long-term ecological monitoring in the Pantanal wetland, due
to the detailed biological information that it is integrated with high spatial
resolution remotely sensed data producing a fine scale representation of
vegetation spatial patterns over large areas.
Key-words: 1.– mapping – 2. spatial autocorrelation – 3. life form – 4.
uncertainty evaluation – 5. plant community – 6. digital images
35
35
2.1. Introduction
Wetland ecosystems are among the habitats most threatened by climatic
change, due to their high sensitivity to the hydrological regime (Junk 2002).
They form transitional habitats between aquatic and terrestrial systems and
embody different kinds of habitats such as mangroves, peatlands, freshwater
swamps and marshes (Mitsch et al. 2009). The ecological importance of these
habitats has been recognized worldwide as well as the urgent need to preserve
them, as stressed in the Cuiabá Declaration on Wetland elaborated during the
8o International Wetlands Conference of INTECOL, Brazil. However, lack of
knowledge about the complex natural dynamics of wetlands may lead to
arbitrary management decisions (Junk et al. 2006b). To improve the protection
of wetlands, it is imperative to have a thorough understanding of the structuring
elements and of the identification of efficient methods to describe and monitor
them.
Vegetation communities have distinct spatial and temporal patterns.
Understanding the mechanisms that determine these patterns has been an
important issue in ecology for decades (e.g., Connell and Slatyer 1977,
Svenning et al 2004). Two factors play a key role: spatial interactions in
ecological processes (e.g. competition), and environmental factors (e.g. flooding
duration) (Tilman 1988). Ecological processes include interactions between
individuals, which may cause particular spatial patterns in the distribution of
plants. Spatial variation in environmental factors causes spatial patterns in
vegetation communities due to the differences of species requirements. These
36
36
two factors do not usually operate independently but act together at different
spatio-temporal scales (Turner et al. 1989, Svenning et al. 2004). This multi-
scale interaction may lead to complex spatial patterns that are continuously
changing (Wagner and Fortin 2005). Consequently, the ability of distinguishing
plant communities that arise from multi-scale ecological processes requires an
understanding of the processes and parameters causing the heterogeneity
(Turner et al. 1989).
Classical methods describing vegetation distribution patterns along
environmental gradients are based on sampling field plots, often along transects
(McIntosh 1958, Whittaker 1967). Such an approach yields detailed insights into
the vegetation occurrence and vegetation assemblages but does not provide
spatially continuous information required to study mechanistic processes and
spatial patterns of the landscape (Austin and Smith 1989). To retrieve such
spatially continuous information requires techniques that consider space
explicitly (Gardner & Engelhardt 2008). One of these techniques is remote
sensing (Gluck and Rempel 1996, Ozesmi and Bauer 2002, Zeilhofer 2006,
Martinez and Toan 2007). By using the spectral signature of different vegetation
species and states, remote sensing enables us to describe spatial and temporal
patterns of vegetation in a spatially continuous way (Jensen 2007). A restriction
of this approach is the limited level of detail in attribute information that can be
mapped by remote sensing, hindering the detection and identification of many
ecologically important properties of vegetation communities, such as floral
composition (Chambers et al. 2007).
37
37
Whereas field plots and remote sensing data each have their limitations
as a source for continuous vegetation maps, is it possible to combine them
through a statistical approach (Guisan and Zimmerman 2000, Ferrier et al.
2002, Pfeffer et al. 2003, Miller et al. 2007). Point-data from field plots and
spatially continuous information from remote sensing are here incorporated by
means of statistical methods, such as ordination analysis (Jongman et al. 1995)
and spatial interpolation techniques such as kriging. In this way, we can make
maps representing the spatial distribution of vegetation across large areas that
incorporate detailed information on floral composition (Pfeffer et al. 2003). This
approach has become increasingly important in ecological studies as it
recognizes the influence of spatial correlation in vegetation patterns
(Bascompte and Sole, 1996, Turner et al. 2001). In addition, these techniques
allow quantifying the uncertainty in mapped vegetation, which is valuable when
vegetation maps are used for further quantitative analysis or for calibration and
evaluation of mechanistic vegetation models (e.g., Brzeziecki et al 1993, Guisan
and Zimmerman 2000, Chong et al 2001). Here, we will use mapped vegetation
(and its uncertainty) to study the effect of flood duration on plant community
patterns.
In this study, we integrate field data and remote sensing data through
geostatistical methods for a case study in the Pantanal, a 150,000 km2
floodplain in the center-west part of Brazil. The variability in water depth and
flood duration are considered to be the preponderant causes of the high
diversity of biological communities and plant zonation patterns found in the area
(Junk et al 1989, Wantzen et al. 2005). In this extensive and pristine wetland
38
38
floodplain, long-term conservation depends on habitat diversity maintenance
(Junk et al. 2006a).
The aims of this paper are: 1) to indentify plant communities of the
Pantanal on key structural and compositional attributes of plant life forms,
based on a new data set collected in a field survey; 2) to present a spatial
statistical approach based on the integration of field data and remotely sensed
data to make accurate predictive maps of vegetation distribution; 3) to evaluate
the uncertainties in vegetation classification on the basis of this novel statistical
approach; and 4) to investigate relation between flood duration and vegetation
zonation.
2.2. Study area
The Pantanal contains a large variety of alluvial ecosystems with different
drainage patterns, flooding characteristics, geomorphologic aspects and
vegetation types covering about 150,000 km2 of the upper Paraguay basin (Fig.
2.1A) (Assine and Soares 2004). The climate of this region is tropical humid
with marked seasonality between winter and summer periods (Köppen 1948).
The summer from November to April is characterized by high temperatures
(average day temperature 34oC) and it is the season with the largest amount of
precipitation (Fig. 2.1B). The precipitation decreases in winter, causing this
season to be very dry (de Musis et al. 1997). The water level in the rivers of the
Pantanal follows the seasonal trend in the precipitation. Due to the poor surface
and subsurface drainage and the smooth, low topography relative to the river
level (Alvarenga et al. 1984, Assine and Soares 2004), large areas of the
39
39
Pantanal are flooded every summer (Junk 1993, Hamilton et al. 1997). Climate
oscillations have been shown to be the main cause of the observed multi-year
period of cyclic variation in flooding (Junk et al. 2006a).
The Pantanal vegetation presents floristic elements of three important
morphoclimatic and phytogeographic domains, i.e., Cerrado (Brazilian
savanna), Amazonia and Chaco (Ab`Saber 1988). Savanna vegetation types
are dominant physiognomies in the Pantanal (67%), but are not the only one:
semideciduous forest, gallery forest, swamp, Chaco, pioneer formations such
as monodominant forest of Vochysia divergens Pohl (Silva et al. 2000) are the
remaining components of the vegetation mosaic. The variability in water depth
and flooding duration and the temporal connections and disconnection
established between different elements of the landscape by means of the flood
pulse (Junk et al 1989) are considered the preponderant causes of the high
diversity of biological communities in the Pantanal (Wantzen et al. 2005),
dictating where and when plant species with different life strategies and flooding
tolerance will appear (Junk et al. 2006a). Our study site covers 60 km2 and is
located within a nature reserve in North Pantanal (16 o 30’ – 16o 44’S and 56 o
20’– 56o 30’W) (Fig. 2.1A). The site is representative of a large part of the
Pantanal regarding vegetation and environmental conditions. The fluctuation in
water level of the river Cuiabá, which crosses the north part of the studied area,
is the main cause of the flooding over the studied area.
40
40
Figure 2.1 Study site. (A) Natural Reserve SESC Pantanal located at the
Pantanal Matogrossense, Mato Grosso; Brazil; (B) Mean annual water depth
fluctuation of the River Cuiabá (1963-2000) and mean precipitation near
Cuiabá, northern Pantanal. Rainfall data from INMET (National Institute of
Meteorology of Brazil), river level data from DNAEE (National Department of
41
41
Waters and Electric Energy of Brazil); (C) Four meter resolution multispectral
IKONOS-2 image, of the study site acquired in October 2003, true color. White
circles are the sampling locations; (D) 90-m Resolution SRTM (NASA Shuttle
Radar Topographic Mission, http://ww2.jpl.nasa.gov/srtm/) Digital Elevation
Model of the study area.
2.3. Outline of the approach
Figure 2.2 shows a diagram with the procedural steps followed to identify
vegetation communities, to determine their spatial distribution, and to study their
relationship with flooding duration. The first part of the paper addresses the
extraction of vegetation communities from high resolution field sampling using
factor analyses and clustering (Fig 2.2, top-right). Spatially continuous variables
were obtained from remotely sensed imagery and a digital elevation model
providing spatial information necessary for vegetation mapping (Fig 2.2, top-
left). These remote sensing and elevation data are related to vegetation field
data using regression analysis (Fig 2.2, centre). After fitting variograms
describing the spatial correlation in the residuals of these regressions, universal
kriging is performed to combine the field point-data and spatially continuous
information from remote sensing to map vegetation communities (Fig 2.2,
centre). The second part of the paper describes an extensive uncertainty
analysis on this mapping procedure by cross-validation and random simulations
to quantify the quality of the vegetation community maps (Fig. 2.2, bottom-left).
Finally, the vegetation map is used to study the vegetation-environment
42
42
relations by comparing spatial patterns of plant community distribution with
spatial patterns of observed flooding duration (Fig 2.2, bottom right).
Figure 2.2 Flow diagram describing the procedural steps in the analysis of the
data.
43
43
2.4. Field data
Vegetation Data Classification and characterization of plant communities were made
based on field sampling of key structural and compositional attributes of the five
following plant life forms as defined by Michin (1989): herbaceous species
(including gramineous plants), vines, shrubs, and two size classes of trees.
Here, we use the term life form with its wider connotation of functional groups
based on ‘group of plants that are similar in a set of traits and their association
to certain variables’ (Pillar and Sosinski 2003). Shrubs were considered
individuals with the trunk bifurcated at the ground level and maximum canopy
height of three meters. Palm species were considered either as a shrub life form
or as a tree, depending on species morphology. Due to possible phenotypic
plasticity found in species living under different micro-environmental conditions,
life form of a species was defined according to the predominant morphologic
form found in our sampling. Two reasons motivated us to focus on on life forms
instead of individual species when describing vegetation. First, this approach
reduces the data dimensionality (Colosanti et al. 2007), and second, life form
and ecology of plants are associated (Grime 1979), which guarantees that each
life form is an ecological unit. Dominant species within each plot, that is, the
woody species with the highest biomass or the vine and herbaceous species
with the highest coverage degree, were identified and included in vegetation
observations and analyses to ensure discrimination between structurally similar
but floristically distinct communities.
44
44
Sampling scheme and data collection
Field sampling of vegetation was done in 2006 and 2007. A sampling
scheme modified from the RAPELD method (c.f., Magnusson et al., 2005) was
used here (Fig. 2.3). The adjusted RAPELD method comprised the
establishment of 23 trails of 250 m length distributed over the study site (Fig.
2.3A). In order to study the effect of flooding duration and soil properties on
vegetation composition, each trail was positioned at a different topographical
elevation. Each trail was thus placed along an elevation contour, defined using
a tripod-mounted telescope. In order to capture variation in vegetation over
short distances, the trails were divided into sampling trails of 50 m length (Fig.
2.3B), producing a total number of 115 sampling units.
Measurement acquisition and sample dimensions of a sampling unit
varied according to life form (Fig. 2.3C). Herbaceous and vine species were
sampled according to the point quadrat method, which is based on point-
intercept frequency measurements by plants (Bullock 1996). Presence or
absence of species was recorded at 25 points spaced at 2 m intervals along the
sampling trail. The coverage value for a sampling trail was calculated as the
proportion of these points being intercepted by the plant. For woody life forms,
plots were positioned along the trail (Fig. 2.3C). The plots have a length equal
to the length of the sampling unit along the trail (50 m), and a width depending
on the life form size as suggested by the RAPELD method (Fig. 2.3C). Shrub
measurements were taken in plots of 200 m2 (50 x 4 m); medium-sized tree
measurements in plots of 1000 m2 (50 x 20 m); and large-sized tree
measurements in plots of 2000 m2 (50 x 40). All species found in the plot were
45
45
identified and trunk diameter and species height were measured for trees and
shrubs. For shrub species, diameters were measured for each individual at 5
cm above the soil surface and for tree species at breast height. These data
were also used to calculate some variables describing the vegetation structure.
Aboveground biomass of woody individuals was estimated by two different
allometric equations for shrubs and trees, respectively. Aboveground woody
biomass (Bs, kg) of shrubs was calculated using the allometric model
developed by Barbosa and Ferreira (2004):
Bs = exp(-3.9041+ 0.4658ln(Cb2H) + 0.0458(ln(Cb2H))2) (1)
with, Cb is circumference at the ground height (cm) and H the species
height (m).
Biomass (Bt, Kg) of a tree species was estimated following Chave et al.
(2005):
Bt = 0.112·(ρ·H·d²)0.916, (2)
with, ρ (g cm-3) the wood specific density, H (m) the species height, and d
(cm) the species diameter at breast height. Information on species densities
was obtained from Schöngart et al. (2008). Canopy height (CH) was considered
the average height of the eight highest individuals in a plot.
46
46
Figure 2.3 Sampling scheme modified from the RAPELD method (see
Magnusson et al., 2005). (A) 23 trails are regularly spaced over the site; (B)
Each trail is placed at the same elevation level and divided in five sampling
trails each of 50 meter length. (C) Sampling along the trail. Herbaceous
species: point samples at regular interval along the trail centre line. Shrubs,
large-sized trees and medium-sized trees: exhaustive sampling in the indicated
zone. The tree size category is based on diameter of the trunk at breast height
(DBH).
DBH)
DBH)
47
47
2.5. Identifying vegetation communities
Velloso et al. (1991) developed a classification system of Brazilian
vegetation, which was adapted by Nunes da Cunha et al. (2006), providing a
detailed description of the plant communities in the Pantanal. Here, we aim at
mapping the communities described by Nunes da Cunha et al. (2006),
considering that these can be identified by means of quantification of structure
and composition (i.e. only dominant species) of different vegetation layers.
Communities are represented at a broad level as vegetation formation types
rather than plant associations. We used factor analysis (Bray and Curtis 1957)
where the factor scores summarize the structural and compositional
characteristics of different vegetation samples. These factors were found in a
principal component analysis of the correlation matrix, generating a small
number of orthogonal factors explaining the correlation among the vegetation
variables (Legendre and Legendre 1998). The different factor scores are plotted
against each other in Fig. 2.4, and the proximity among point-samples and our
field background about the vegetation classes found in these points were used
to classify them in vegetation classes/clusters. Finally, cluster centers were
calculated by averaging factor scores corresponding to the community/cluster
and were used in the final part of the mapping procedure below.
Ecological Interpretation of Ordination Space
The first four factor axes explain 46% of the total variance (Table 2.1).
We assumed that the strongest correlations with each axis reflect the main
vegetation gradients captured by it. Factor 1 explains a relatively large
48
48
proportion (22%) of the total variance. It mainly distinguishes communities
dominated by a tall and rich tree layer (negative loadings) and those dominated
by vine, shrub or herbaceous life forms (positive loadings). Although explaining
considerably smaller proportions of the total variance, the remaining factors are
still useful for identifying the vegetation classes. Factor 2 separates plant
communities by their degree of coverage and richness of herbaceous species.
Factor 3 mainly represents variation in biomass of two trees, Brosimum
latescens and Mouriri guianensis, and one shrub, Psychotria capitata. Factor 4
mainly represents variation in the biomass of shrubs and of two species, the
medium-sized tree Sapium obovatum and the shrub Ruprechtia brachycepala.
Table 2.1 Summary statistics for the factor analysis. Numbers in bold highlight
the highest correlation with the factor axis.
Variable Factor 1 Factor 2 Factor 3 Factor 4 Richness tree -0.76 -0.25 0.17 -0.28 Richness shrub 0.28 0.07 0.20 -0.35 Richness herb 0.31 0.58 0.43 -0.11 Canopy height -0.86 -0.05 -0.33 0.09 Cover %herb 0.04 0.79 0.16 0.06 Cover % vine 0.82 -0.15 -0.16 0.16 Richness vine 0.73 -0.09 -0.14 0.16 Biomass tree (total) -0.88 0.00 -0.26 0.18 Biomass shrub 0.54 -0.42 0.07 -0.58 Biomass (DBH 10 cm > 30 cm)
Vochysia divergens -0.31 -0.05 0.05 0.03 Sapium obovatum -0.03 -0.23 -0.09 -0.56 Licania parvifolia -0.43 -0.03 -0.17 0.11 Brosimum latescens -0.29 -0.34 0.51 0.22 Trichilia catigua -0.30 -0.27 0.45 0.06 Duroia duckei -0.59 0.09 -0.35 0.19 Cecropia pachystachya -0.31 0.23 -0.20 -0.02 Mouriri guianensis -0.51 -0.35 0.36 0.22
Biomass (DBH > 30 cm) Vochysia divergens -0.72 0.17 -0.47 0.21
49
49
Variable Factor 1 Factor 2 Factor 3 Factor 4 Mouriri guianensis -0.37 -0.42 0.55 0.23
Biomass (shrub) Albizia polycephala 0.62 -0.11 0.02 0.12 Ruprechtia brachycepala -0.02 -0.25 0.05 -0.50 Peritassa dulcis 0.18 -0.52 -0.46 -0.06 Melochia villosa 0.41 0.16 0.18 0.24 Byrsonima cydoniifolia -0.23 -0.40 0.43 0.15 Psychotria capitata -0.48 -0.49 0.51 0.20 Bauhinia rufa 0.17 0.04 0.09 -0.32 Mimosa pellita 0.57 0.05 0.02 0.35 Laetia americana 0.74 -0.20 -0.09 0.14 Solanum
pseudoauriculatum 0.26 0.19 0.15 0.22
Eugenia florida 0.03 -0.29 -0.32 0.07 Alchornia discolor -0.22 0.37 0.03 -0.37 Mabea paniculata -0.30 0.12 0.17 -0.25 Byrsonima orbygniana 0.00 0.19 0.25 -0.49
Cover % herbaceous species Paspalum hydrophilum 0.34 0.37 0.24 0.05 Panicum guianense -0.17 0.06 -0.18 -0.35 Scleria bracteata -0.57 0.24 -0.27 0.16
Cover % vine Cissus spinosa 0.67 -0.31 -0.20 -0.001 Aniseia cernua 0.54 0.14 0.09 0.34 Paullinia pinata 0.52 -0.31 -0.27 0.09 Dolliocarpus dentatus 0.23 -0.57 -0.27 -0.01 Ipomea rubens 0.36 0.16 0.17 0.22
% Variance 22 9 8 7
Defining vegetation communities through ecological interpretation of clusters
The seven clusters are indicated in a scatter plot of the different factors (Fig.
2.4) and are identified as: Monodominant forest, Shrubland, Alluvial seasonal
semideciduous forest (Alluvial forest), Alluvial seasonal semideciduous low
forest (Alluvial low forest), Seasonally flooded grass-woody savanna
(Grassland), Low open tree and shrub savanna (Open savanna) and Low dense
tree and shrub savanna (Dense savanna). The number of samples in each
cluster and their distribution over the ordination space express the structural
50
50
and floristic variability found within the community and which communities have
dominated the floodplain landscape. Table 2.2 provides a statistical summary of
structural and floristic characteristics of communities.
A number of communities show overlapping ranges of scores on some of
the factor axes, while other factor axes provide clear boundaries between these
communities. For instance, the transitions between Alluvial forest and
Monodominant forest are smooth (Fig. 2.4A-C). These two communities are
mainly separated through the tree biomass and coverage of herbaceous
species in Monodominant forest (Fig. 2.4A) and the dominance of Brosimum
latescens and Mouriri guianensis in Alluvial forest (Fig. 2.4B). Dense savanna
lies between Open savanna and Monodominant forest. Richness and coverage
of herbaceous life form distinguish these communities (Fig. 2.4A). Dense
savanna, Grassland, Open savanna and Monodominant forest have similar
correlation values with Factor 2, indicating that there may be a small variation in
coverage of herbaceous species between these communities. The low tree
biomass in Shrubland is responsible for its positive scores on the first factor.
Alluvial low forest is distinguished from other forests based on Factor 4. Its high
shrub coverage compared to Shrubland and the dominance of Sapium
obovatum and Ruprechtia brachycepala generates scores on Factor 4 in
intermediate position between Shrubland and Alluvial forest.
51
51
Figure 2.4 Factor analysis biplots of the axes 1 to 4 on vegetation variables
obtained from 115 sampling locations. Seven clusters (symbols) represent the
vegetation communities found in the studied floodplain. Factor 1 describes the
52
52
gradient of tree biomass found in the study site; Factor 2 of herb cover and
richness; Factor 3 of tree dominant species; and Factor 4 of shrub biomass.
2.6. Mapping plant communities
Remote sensing and ancillary data
Remotely sensed imagery and ancillary data are frequently used in spatial
vegetation modeling due to their capability of providing accurate environmental
information related to vegetation patterns (Guisan and Zimmerman 2000,
Pfeffer et al. 2003, Miller et al. 2007). An IKONOS-2 image and a Digital
Elevation Model (DEM) (Fig. 2.1C, D) were used in this study to derive variables
related to vegetation patterns. The acquisition date of the IKONOS-2 image is
October 1st, 2003 corresponding to the dry season in the Pantanal and
representing an optimal time for detecting spectral signatures of terrestrial
vegetation on the floodplain, due to the availability of cloud free images and
nonflooded soil conditions. The IKONOS-2 image consists of four spectral
bands: three bands in the visible part of the spectrum located at blue (450-
520nm), green (520-600 nm) and red (630-690 nm) and one band in Near
Infrared (760-900 nm). The pixel size is approximately 4 by 4 m. The registered
radiance values by the IKONOS-2 sensor were converted to reflectance values
using the calibration information provided by Bowen (2002). A Normalized
Difference Vegetation Index (NDVI) was computed from the spectral bands by
taking a ratio of the difference of the near infrared and red spectral bands and
the sum of the near infrared and red band (Tucker 1979). Such an NDVI image
shows stronger contrast between vegetation and soil and water surfaces while
53
53
reducing noise in the image. Furthermore, we applied a principal component
(PC) transformation to the IKONOS-2 image to reduce inter-band correlation
and extract new spectral information that arises from this transformation. The
four original bands, the NDVI image and the four principal component images
were used for further data analysis as described in the next section. Ancillary
data, such as soil and topography maps, in combination with multi-spectral
bands have been used to improve classification of wetlands (Ozesmi and Bauer
2002). The 90-m resolution DEM of the study area was obtained from the
SRTM (NASA Shuttle Radar Topography Mapping Mission) and used to provide
continuous information of canopy height rather than soil surface (Jacobsen
2006) (Fig. 2.1D).
Re-scaling and extracting image derived data
The original geodata with cell sizes of 4 m (IKONOS-2) and 90 m (DEM)
were re-sampled to the support of the field data, i.e., to the plot size used to
take measurements of large tree species (50 x 40 m). The resampling
technique applied consists of: 1) delineating the irregular plot boundaries in the
IKONOS-2 image using ARC/INFO GIS software (version 9.0; ESRI, 2006); 2)
calculating average remote sensing and elevation values for exactly these
digitized plots; and 3) extracting variables from the IKONOS-2 derived images
and SRTM DEM for the 115 plots to be used in the analysis. The two last steps
were done with PCRaster (PCRaster 2002; Wesseling et al. 1996).
54
54
Table 2.2 Structural and floristic characteristics of plant communities, given as mean and
standard deviation
Monodominant forest Shrubland Alluvial
forest Alluvial Low
Forest Grassland Open savanna
Dense savanna
Vochysia divergens Pohl.
Laetia americana L.
Byrsonima cydoniifolia A. Juss.
Ruprechtia brachysepala Meisn.
Paspalum hydrophilum Henrard
Paspalum hydrophilum Henrard
Byrsonima orbignyana A. Juss.
Duroia duckei Huber
Mimosa pellita Humb. & Bonpl. ex Willd.
Psychotria capitata Ruiz & Pav.
Crataeva tapia L.
Panicum guianense Hitchc.
Hybiscus furcellatus Desr.
Bauhinia rufa (Bong.) Steud.
Licania parvifolia Huber
Peritassa dulcis (Benth.) Miers
Trichilia catigua A. Juss.
Banara arguta Briq. Laetia
americana L. Alchornea discolor Poepp.
Scleria bracteata Cav.
Albizia polycephala (Benth.) Killip
Mouriri guianensis Aubl.
Sapium obovatum Klotzsch ex Müll. Arg.
Cissus spinosa Cambess.
Brosimum lactescens (S. Moore) C.C. Berg
Cecropia pachystachya Trécul
Aniseia cernua Moric.
Paullinia pinnata L.
Characteristic species
Ipomea rubens Chousy
Richness of herbs (no of sp. per sample)
1.98 ± 1.92 2.54 ± 1.75 2 ± 0.82 2.4 ± 1.52 3.67 ± 1.56 4.6 ± 1.34 4.2 ± 1.92
Richness of vines (no of sp. per sample)
2.96 ± 1.52 7.71 ± 1.72 1.71 ± 1.25 4.6 ± 1.82 4.42 ± 1.31 3.8 ± 1.1 1.4 ± 1.52
Richness of shrubs (no of sp. per sample)
7.85 ± 3.58 9.89 ± 2.36 6.71 ± 4.61 6.4 ± 1.95 8.58 ± 2.91 8.6 ± 2.3 14.6 ± 3.6
Richness of medium sized trees (no of sp. per sample)
4.91 ± 1.64 0.32 ± 0.61 8 ± 2.65 6.2 ± 2.17 0.17 ± 0.39 1 ± 0.45 5.8 ± 1.64
Richness of large trees (no of sp. per sample)
2.77 ± 0.71 0.14 ± 0.45 4.71 ± 1.11 1.6 ± 1.34 0.17 ± 0.39 0.6 ± 0.55 2 ± 0.7
Biomass of shrubs (Mg ha-1) 3.25 ± 6.71 10.11 ± 3.79 3.82 ± 3.01 13.54 ± 8.18 4.03 ± 2.08 2.63 ± 0.78 9.96 ± 6.71
Biomass of middle sized trees (Mg ha-
1) 109.58 ± 17.29 2.38 ± 7.58 91.54 ± 29.87 28.50 ± 21.78 4.57 ± 9.56 29.5 ± 31.01 26.75 ± 17.29
Biomass of large trees (Mg ha-1) 84.94 ± 11.36 1.87 ± 6.80 57.15 ± 18.55 9.47 ± 8.41 1.81 ± 4.29 24.8 ± 28.06 9.40 ± 11.36
Canopy height (m) 20.19 ± 1.30 2.41 ± 0.84 15.2 ± 3.19 5.72 ± 1.74 1.76 ± 0.21 2 ± 0.1 3.14 ± 1.3
Cover % herbs 37.06 ± 9.55 27 ± 24.26 16 ± 7.66 18.4 ± 11.52 61.33 ± 21.46 97.6 ± 3.58 56.8 ± 9.55
Cover % vines 21.74 ± 6.07 83 ± 11.71 8.57 ± 6.70 47.2 ± 21.05 43.33 ± 19.66 27.2 ± 7.16 5.6 ± 6.07
55
55
Correlating field data and image derived data
We first examined the relationship between image and DEM derived
variables and the vegetation patterns captured in the four factorial axes. The
functional relationships between each factor axis and image and DEM derived
variable were found by Pearson’s correlation analysis, to facilitate the ecological
interpretation of the variables (James and McCulloch 1990). Next, the
relationship between the image derived variables and the factor axes was found
using the following multiple linear regression model:
ipipiii xaxaxaaY ε+++++= L22110 (3)
where Yi is the score value, paaa ,,, 10 L are the model parameters, x1i,
x2i, …, xpi are the values of the image derived variables and iε are uncorrelated
residuals. The analyses were done with log transformed reflectance values to
ensure that the statistical distribution of the data is close to Gaussian (Draper
and Smith 1998).
Before performing the multiple regression analysis, image derived
variables were selected to be included in the multiple regression models using
the best-subset regression method (Hofmann et al. 2007). In this method, all
combinations of explanatory variables in regressions are tested, and Mallow’s
C-p statistic (Mallows 1973) is used as eliminatory criterion of variables (Draper
and Smith 1998). We consider the best regression equation for each factor that
one combining lowest C-p value and lowest number of explanatory variables.
56
56
Vegetation patterns captured by digital images
Table 2.3 shows correlations between explanatory variables and the
factor axes. Except for NDVI, all image derived data present significant
correlation with Factor 1. The strongest correlations with this first axis are found
with blue, green and red bands, PC1 and canopy topography. Lower reflectance
values in the three spectral bands and lower score values in the PC1-3 images
are linked to areas occupied by communities with high stored tree biomass such
as Monodominant forest and Alluvial forest (Table 2.3). Lower score values in
the PC4 reflects communities with lower tree biomass values even though this
axis explains the noise from the spectral band transformation. In spite of its
weak correlation with Factor 1, NDVI shows an expected spectral behaviour:
the values decrease toward areas with lower tree biomass, such as those areas
covered by Grassland, Open savanna, Shrubland and Dense savanna. The
strong negative correlation between canopy topography and Factor 1 shows
that the boundaries between communities dominated by trees and those
dominated by shrubs, lianas and herbs are detected by differences in canopy
height.
57
57
Table 2.3 Pearson’s correlation coefficients between factor axes and image
variables: four spectral bands, Normalized Difference Vegetation Index
(NDVI), Principal Component transformation to the IKONOS-2 image (PC),
and canopy topography derived from DEM-SRTM (DEM). * P ≤ 0.05
The variability in cover degree and richness of herbaceous life forms
expressed by the second axis is best described by the PC2 image (61%; Table
2.3). Communities with higher and richer coverage of herbaceous species such
as Grassland, Open savanna and Dense savanna are associated with higher
reflectance values in blue, green and red bands and higher score values in the
PC2 image. The negative correlations between Factor 2 and infra-red band and
NDVI show that communities dominated by herbaceous species present weaker
spectral response to these two images.
As observed earlier, Factor 3 mostly justifies the spatial distribution
pattern of three tree species that dominate in Alluvial forest. Relatively to
Monodominant forest, the lower biomass content and canopy height of Alluvial
forest might be the cause for the negative correlations between Factor 3 and
NDVI and Factor 3 and canopy topography.
Variable Factor 1 Factor 2 Factor 3 Factor 4blue band 0.71* 0.30* 0.38* 0.008green band 0.71* 0.23* 0.38* -0.028red band 0.67* 0.36* 0.37* -0.009infra-red band 0.43* -0.34* 0.023 -0.001NDVI -0.124 -0.47* -0.30* 0.012PC1 0.69* -0.126 0.25* -0.028PC2 0.29* 0.61* 0.30* 0.035PC3 0.26* -0.076 0.025 0.095PC4 -0.33* 0.129 -0.097 -0.1Canopy topography (DEM) -0.72* 0.153 -0.39* -0.163
58
58
The spatial variability of Factor 4 represents vegetation patterns that are
mainly explained by canopy topography (i.e. DEM) showing that areas with
higher biomass of shrubs are associated with lower canopy height.
The equations found in the multiple regression analysis are shown in
table 2.4. The regression models significantly explain 70.4%, 66.3%, 31.3% and
25.6% of the variance in factors 1 to 4, respectively.
Table 2.4. Multiple linear regression models relating factor axes scores (F1-4) to
imagery derived variables: four spectral bands (blue, green, red and infra-red),
Normalized Difference Vegetation Index (NDVI), Principal Component
transformation to the IKONOS-2 image (PC), and canopy topography derived
from DEM-SRTM (DEM). R2 is the coefficient of determination showing the
strength of these relationships.
Variogram analysis
We applied variogram analysis on the residuals of the multiple linear
regression to derive information on their spatial structure (Wagner and Fortin
2005). This information was used for two reasons: 1) to investigate the spatial
autocorrelation associated with the observed vegetation patterns; 2) to use this
information when making spatial predictions (Miller et al 2007). Sample
Equation R 2
F1 = 30.6 + 9.49 blue - 0.024DEM - 34.7 PC1 - 46.4 PC2 - 33.7 PC3 + 8.7 PC4 + 2.15 NDVI 70.4
F2 = - 9.43 + 5.15 NDVI + 0.0715 DEM - 3.38 green + 3.52 red + 55.1 PC2 - 140 PC3 66.3
F3 = 15.2 + 419 PC4 - 9.98 NDVI + 15.9 blue - 15.2 red + 4.90 infra-red 31.3
F4 = 2.4 -27.9 PC1 + 53.9 PC2 + 129 PC3 - 49 PC4 - 0.9 NDVI - 0.15 DEM - 20.9 blue + 2.2 green + 9.8 red+ 3.6 infra-red 25.6
59
59
variograms were estimated and variogram models fit using the function
autofitVariogram from the library automap (Hiemstra et al., 2008) in the
statistical environment R (R Development Core Team 2009).
The results indicate that the vegetation gradients represented by the
residuals of each factor (Factor 1-4) vary on different spatial scales (Fig. 2.5).
Variograms of the Matheron family, a family of semivariogram models where the
degree of smoothness of the random field is controlled through a shape
parameter (kappa) (Pardo-Iguzquiza and Chica-Olmo 2008), were fit for Factors
1, 2 and 4, Fig. 2.5A,B,D), whereas an exponential variogram (special case of
the Matheron family) was fit for Factor 3 (Fig. 2.5C). The first and third factors
show large-scale patterns as revealed by their ranges of spatial dependence.
The variogram of Factor 1 has a range of 3,380 m., whereas the variogram of
Factor 3 is monotonically increasing within the extent of the sample variogram
and consequently has a larger range. The variograms of Factors 2 and 4 show
short ranges of spatial dependence (close to a pure nugget effect) suggesting
that processes governing their spatial patterns show small scale variability.
Universal Kriging
Universal kriging is a spatial interpolation technique that can incorporate
environmental data and spatial dependence in the modeled error to predict at
locations without observations and generate accurate vegetation distribution
maps (Pfeffer et al. 2003, Pebesma and Wesseling 1998). Universal kriging was
done on the regression residuals and the interpolated residuals were added to a
trend surface to predict factor scores at unobserved locations. This trend
surface was based on the regression equation in equation 3 (Pfeffer et al.,
60
60
2003). The predicted scores were used to create four factor score maps. In
addition, the universal kriging approach was used to estimate the prediction
error (standard deviation), which is typically increasing as a function of the
distance to observation locations (Stein and Corsten 1991).
Figure 2.5 Maps of the kriged estimates of factor scores and the semi-
variograms of the residuals of the regression between factor axes and remotely
sensed and ancillary data; Fitted variogram models: Mat: Matheron family, Exp:
61
61
Exponential. Values between brackets are nugget effect, structured variance
and variogram range, respectively. (A) Factor 1; (B) Factor 2; (C) Factor 3; (D)
Factor 4.
Continuum representation of vegetation spatial patterns
The score maps in figure 2.6 show the vegetation spatial patterns predicted by
universal kriging. The score maps of the first and third factor axes (Fig. 2.5A
and C) show mainly large-scale variability. These axes, as mentioned earlier,
mostly represent spatial variation of tree life forms. Contrarily, the score maps of
the second and fourth axes (Fig. 2.5B and D) representing the occurrence of
herbaceous and shrub layers, respectively, show small-scale spatial variability.
Examining the pattern of the prediction errors of the scores for each factor axis
(Fig. 2.6), one can infer to which extent sample data and image data contribute
to predictions. When the range of the semivariogram is large, as seen in the
semivariograms of Factor 1 and 3 (Fig 2.5A,C), the prediction errors increase
slowly with the distance away from samples. On the other hand, a short range
in the variogram results in prediction errors increasing rapidly with distance
away from samples, as is the case with Factor 2 and 4. Image data will in this
case have greater impact on predictions. Nevertheless, the quality of the factor
score maps is not only related to differences between small-scale and large-
scale spatial variation but rather reflects the explanatory strength of the
relationship between factor axes and image derived variables as shown by the
mean error in the score maps. According to these averages, Factor 1
62
62
represents the most accurate map (mean SD = 0.51) followed by Factor 2
(mean SD = 0.64), Factor 3 (mean SD = 0.69) and Factor 4 (mean SD = 0.87).
Figure 2.6 Maps of the standard deviation of the predicted error resulting from
universal kriging; (A) Factor 1; (B) Factor 2; (C) Factor 3; (D) Factor 4.
63
63
Spatial distribution of plant communities across the floodplain
In the final part of this procedure, we combined results from spatial and
non-spatial analyses generated as described in the former sections to create
the final map of plant communities. The clusters/communities centers,
calculated in the section ‘IDENTIFYING VEGETATION COMMUNITIES’ were
used to assign each location on the map to a community class. This was done
by calculating Euclidean distances between centers and predicted scores
values. Each location was then assigned to the community whose center was
nearest to the predicted score values at that location.
The map of plant communities (Fig. 2.7A) resulting from this classification
method shows the predicted spatial distribution of the seven identified plant
communities on the floodplain. Grassland (16% of coverage), Shrubland (30%
of coverage) and Monodominant forest (32% of coverage) sum up to 78% of the
coverage at the studied site. These communities mostly appear as large and
contiguous patches across the site. Alluvial forest and Alluvial low forest, as
expected, appear as strips covering exclusively places close to water bodies:
along rivers, channels and surrounding baías, i.e. temporary or permanent
lakes seasonally connected to the river. These two communities cover just 4%
(2% each) of the studied floodplain. The greatest portion of the 10% of Open
savanna that covers the study area is located towards the Northern boundary.
The 8% of Dense savanna is found as small patches generally surrounded by
Open savanna and as a big patch beside Monodominant forest.
64
64
Figure 2.7 (A) Predicted distribution of the plant communities identified at the
study site. (B) Results of leave-one-sample out cross-validation. Percentage of
predicted classes at sampling locations. Each bar shows the results for
65
65
sampling locations with a certain observed class (indicated by the colors at the
bottom of the C panel). (C) Idem, leave-five-out cross-validation. N = 115.
Evaluating uncertainty
Vegetation mapping using statistical approaches carries different sources
of uncertainties related to sampling scheme, interpolation errors, sampling
support, data quality, lack of data and others, which may compromise the
model’s capability of accurately predicting vegetation patterns (Guisan and
Zimmerman 2000, Pfeffer et al. 2003, Miller et al. 2007). The predictive success
of our mapping approach was evaluated using cross-validation (Efron and
Tibshirani 1986) and random-simulations (Bourennane et al. 2007), both
performed in R (R Development Core Team 2009).
Cross-validation
We have used cross-validation to investigate the sensitivity of vegetation
predictions performed by universal kriging as a result of sampling variability
(Pfeffer et al. 2003). Two resampling techniques were applied: leave-one-out
cross-validation (LOOCV) and leave-five-out cross-validation (LFOCV). The first
technique is the standard procedure (Efron and Tibshirani 1986) which consists
of omitting one sample at a time from the data set and based on the remaining
observed values make predictions at this location using the interpolation
technique, i.e., universal kriging. Because samples/plots within the same trail
are considerably closer to other observations than the typical distance between
prediction locations and observations locations (Miller et al. 2007), LFOCV was
used to test the prediction quality of the model when the whole trail, that is, five
66
66
plots, is left out to make predictions. Vegetation classes were assigned from the
predicted scores and compared with the observed vegetation classes at the 115
sample plots.
Overall agreement between predicted and observed classes does not
differ substantially between the two resampling techniques: leave-one-out
results in 52.2 % agreement and leave-five-out in 48.7 % agreement. Both
techniques show that accuracy in classification varies according to the
community type (Fig. 2.7B and C). Communities which have been observed on
a large number of plots and occupy large portions of the vegetation map, such
as Monodominant forest and Shrubland, are less sensitive to sampling density
than those communities which occur in smaller and few patches, such as
Alluvial forest and Alluvial low forest. Consequently, communities observed in
few of the plots are wrongly classified also for LOOCV (Fig. 2.7B). Other
possible causes of uncertainty in classification from our mapping approach
derives from the similarity between community types having a small distance
between cluster centers in the ordination space (Fig. 2.4). Communities such as
Alluvial forest and Dense savanna are frequently predicted to be their
neighboring communities, namely, Monodominant forest; and Alluvial low forest
are frequently predicted to be Shrubland (Fig. 2.7B and C).
Simulation
A Monte Carlo approach was applied to examine the uncertainty of our
method (Legendre and Legendre 1998). In this approach, we performed the
same universal kriging, however creating random realizations of score maps
conditioned to the observations instead of predicted values as was done in the
67
67
original procedure. This was done by simulating 1000 realizations of score
maps for each factor with Gstat (Pebesma, 2004), based on the scores at the
observation locations and the fit variograms. These realizations reflect the
prediction uncertainty at the prediction locations; all realizations are equally
probable. For each realization, we calculated the vegetation pattern, using the
same Euclidean distance algorithm applied in the original mapping procedure.
This was repeated for all 1000 realizations, resulting in 1000 realizations of
vegetation community maps. Two realizations are shown in Fig. 2.8 B, C. From
these 1000 realizations, we created a map showing the probability, from 0 to 1,
that a certain community is found in a 40 m grid cell (Miller and Franklin 2006)
(Fig. 2.8). On this map, a value 1 indicates zero prediction uncertainty.
The result shows that the quality of classification varies spatially, even
though the proportion and arrangement of communities observed in the original
map is preserved to a great extent. The central zone of a community patch is
more likely to be classified correctly than border areas, as shown by the
increasing probabilities towards the center of patches of communities (Fig.
2.8A). This might be related to intrinsic uncertainties in classification of natural
ecotones reflected in the overlapping of score values of very close communities
in the factor space. The quality of classification also varied between
communities. Classification of Dense savanna and Open savanna, for instance,
exhibit lower probabilities of being in the correct class as indicated by their more
random distribution across the landscape (Fig. 2.8B and C). Here, sampling
configuration and distance between clusters in factor space are an important
source of errors.
68
68
Figure 2.8 Results of Monte Carlo simulation, using 1000 random simulations.
(A) The colors on the map indicate the vegetation class with the highest
probability of occurrence at a cell. A color gradient is used to show the value of
this highest probability; (B) Maps of two single random realizations, color scale
is identical to Figure 2.8.
69
69
2.7. Flood duration-vegetation relationship
The relationship between vegetation distribution and flooding was
assessed by comparing the plant community map with a flood duration map as
in direct gradient analysis. The flood duration map (Fig. 2.9) was created from a
digital elevation map and 38 years of daily recordings of the water level in the
River Cuiabá (Fig. 2.1B) provided by the Brazilian National Water Agency (ANA;
(http://hidroweb.ana.gov.br). The 40-m resolution digital elevation map was
created with universal kriging from 81 GPS elevation measurements at the site
and using SRTM DEM as an auxiliary variable (Valeriano and Abdon 2007). A
base station was installed for increased precision of the GPS measurements.
Flood duration and flood depth data were also monitored by direct reading of
staff gauges for two years (2007-2008) at the 23 sampling trails. The
relationship between flooding and elevation data was tested with Pearson’s
correlation coefficient. Statistically significant and strong correlations were found
among them (r > 70%; P ≤ 0.05) indicating the possibility of calculating flooding
duration values over the floodplain through the indirect relationship between
river water depth and elevation. Flood duration of a cell was calculated by
comparing the water level in the river and the topographical elevation of the cell
for each day as follows: if the elevation value at a cell was lower than the water
level in the river on a certain day, the cell was considered flooded that day. This
approach ignores spatial variation in water level associated with downstream
gradients in water level, local depressions containing water that is only partially
connected with the main river, and surface water fed by groundwater. However,
70
70
the effect of these processes is relatively small as indicated by additional field
sampling with the staff gauges.
The flood duration map (Fig. 2.9) shows the number of flooded days per
year in the study area. Flood duration data extracted from this map were
classified into monthly intervals and the distribution of the plant communities
found in the vegetation map along this flooding gradient was plotted in Fig. 2.10.
Fig. 2.10 shows that the zonation of plant communities along the floodplain is
clearly related to the duration of inundation. Alluvial forest and Dense savanna
occur in areas with a flooding duration of less than two months. Monodominant
forest, although occupying a high proportion of the highest areas, has the
highest occurrence at intermediary flooding conditions, with a flooding duration
between two and four months. Open savanna is mostly found where flooding
lasts for four to six months per year. Grassland is found under almost the whole
range of flooding durations, however with peaks of occurrence in areas with a
flooding duration below two months and between four and six months of
inundation. Alluvial low forest is mostly situated at locations with a flooding
duration between 6-8 months. Shrubland dominates the areas with the highest
flooding duration. Above eight months of flood duration, there is no suitable
condition for tree species establishment and the landscape is occupied mostly
by Shrubland, Open savanna and Grassland. The occurrence of monodominant
forest in this last flood duration class might be associated with the coarse
representation of spatial variation in flood duration, illustrated in the map (Fig.
2.9).
71
71
Figure 2.9 (A) Map with average number of days flooded per year at the study
site, calculated over the period 1969-2007; (B) Relationship between flood
duration (days yr-1) and elevation of the soil surface (meter a.s.l) observed at
the 23 study trails; (C) water level fluctuation in the River Cuiabá between 1969
and 2007. Vertical dotted lines indicate the occurrence of drier and wetter years.
72
72
Figure 2.10 Fraction of occupied sites by the seven identified communities
along the flood duration gradient. Flooding gradient is divided in five flood
classes representing number of flooded months.
2.8. Discussion
We showed that it is possible to classify vegetation at locations in the
studied floodplain by measuring structural and floristic attributes of different
vegetation layers (herbaceous, tree, shrub and vines), and combining these
data with remote-sensing imagery and DEM data. The plant communities
described in an existing classification could be clearly identified as clusters in
73
73
the ordination space, thanks to the floristic properties included in the analyses
that differentiated structurally similar but floristically different plant communities.
However, sometimes clusters showed overlap on a number of factor axes and
boundaries between clusters were not always accurate. Such overlap probably
indicates the existence of gradual changes in vegetation (Brzeziecki et al 1993),
which is not represented in our model with sharp boundaries between
vegetation communities. Thus, the vegetation community studied deviates
slightly from our crisp plant community model. This had two implications for our
analysis. One is the subjective determination of cluster boundaries in the
ordination space, particularly in cases where boundaries were not crisp. The
other is related to the interpretation of the uncertainty analysis. One of the
causes of uncertainty of the mapped vegetation is the uncertainty in the
assignment of an interpolated point to a cluster in the ordination space. Overlap
of clusters in the ordination space may actually represent transition zones
between plant communities, and are related to intrinsic uncertainty in
classification (see also, Fortin et al. 2000, Hernandez-Stefanoni and Dupuy
2007). Potential misclassification in these zones appears as uncertainty on the
interpolated crisp map, particular in areas close to mapped boundaries between
plant communities. However, the assumption of crisp plant communities as a
spatial concept is in most cases sufficient to interpret vegetation patterns, as in
our study (see also Austin and Smith 1989, Brzeziecki et al. 1993, Pfeffer et al.
2003).
The statistical approach described here shows the value of integrating field
observations and high resolution remote sensing. The field observations provide
74
74
enough information to identify vegetation communities, while remote sensing
reduces the interpolation error because derivatives of the remote sensing image
explain a significant part of the spatial variation in vegetation. The use of
universal kriging is valuable here because parts of the remaining variability can
be modeled as a spatially correlated residual. These findings confirm that the
use of remote sensing and a spatial interpolation method reduce the uncertainty
in mapped vegetation, as was also shown in other studies (e.g., Guisan and
Zimmerman 2000, Ferrier et al. 2002, Pfeffer et al. 2003, Miller et al. 2007).
The different techniques used to evaluate the behavior of the statistical
model used for mapping vegetation, e.g. cross-validation and Monte Carlo
simulation, allowed us to identify possible causes of misclassification and
determine spatial prediction uncertainty (Congalton and Green 1999, Guisan
and Zimmerman 2000, Pfeffer et al. 2003). The accuracy levels of the
vegetation map derived from the mapping procedure described here and
assessed by cross-validation (e.g. 49 of 52%) were of the same magnitude as
to those found by Pfeffer et al. (2003) in their maps of Alpine vegetation (e.g. 50
to 65%). The uncertainties in vegetation classification that resulted from the
sampling density and configuration suggest that the map quality may be
improved when samples are collected at a higher density (c.f., Guisan and
Zimmerman 2000, Pfeffer et al., 2003, Miller et al. 2007). In geostatistical
approaches for vegetation mapping as used in this paper, large distances
between the observations directly affect the estimated accuracy of the
predictions (Miller and Franklin 2006, Miller et al. 2007). Our study showed that
the gap of vegetation information due to large-distance separated sampling
75
75
points can be filled with information contained in the remote sensing images. As
a result of the sampling scheme used here, the systematical sampling produced
an uneven number of samples per community type and possible omission of
communities limited to small and scattered patches over the floodplain, as was
the case of Savanna forest (personal observation). Consequently, the level of
uncertainty in predictions varied among communities and in space.
Uncertainty assessment and its cartographic representation is an important
tool for management and research, indicating zones of high and low
classification confidence and helping to find strategies for mapping
improvement (Chong et al 2001, Guisan and Zimmerman 2000, Pfeffer et al.
2003, Scheller and Mladenoff 2007). Many strategies can be used in such a
statistical approach to improve vegetation map quality. Increasing the number of
samples and better sample designs are the most obvious ways to improve
classification accuracy (Guisan and Zimmermann 2000, Pfeffer et al. 2003,
Rempel and Kushneriuk, 2003). There are also other image derived predictors
that could have been included in this study, such as digital maps of soil
properties (e.g. soil texture) or flooding attributes.
Our analysis of the causes of vegetation zonation on the floodplain
indicated that flood duration is an important determinant of plant community
distribution in space, influencing spatial transitions between different plant
communities (Zeihofer and Schessl 2000, Damasceno-Junior et al. 2005).
Different mechanisms of tolerance to prolonged flooding evolved by species of
a community (c.f. Parolin 2009) might be related to vegetation zonation by
controlling expansion of different set of plants (Damasceno-Junior et al. 2005).
76
76
On the other hand, non-linear response of communities to the flood duration
gradient, as was the case of Grassland, indicate that interaction with
neighboring communities might have a strong influence on the vegetation
distribution of these communities (Fig. 2.10) (Austin 2002). Based on these
findings, we conclude that vegetation zonation in the studied floodplain might be
influenced not just by physiographic limits from flood duration, as stressed in
most of the studies in the Pantanal (Junk 1989, Nunes da Cunha and Junk
1999, 2000, Zeihofer and Schessl 2000), but also by biological constraints
related to competition between neighbors (Tilman 1994).
The significant advantage of the mapping approach described in this
paper is that detailed biological information present in field observations can be
integrated with high spatial resolution remotely sensed data producing accurate
vegetation maps. Different from ‘classical’ approaches to vegetation class
mapping, our modeling carries quantitative information on vegetation variability
allowing future application in modeling concerned with the effects of
environmental shifts on biological patterns and processes (Arieira et al. in
preparation; Brzeziecki et al 1993). We believe that mapping of plant
communities by integrating field observations and high-resolution imagery is a
promising approach for conservation assessment and long-term ecological
monitoring in extensive wetland areas.
2.9. Acknowledgments
The authors are grateful to the Brazilian governmental agencies, CAPES
and CNPq, for the financial support. Helpful comments and assistance were
provided by P. Girard, Peter Zeihofer, Arnildo Pott, Vali J. Pott, Sandra Santos
77
77
and José F. M. Valls. We also thanks to the Social Service of the Commerce
(SESC) and technicians and students of the Federal University of Mato Grosso,
for the technical support in field work.
CAPÍTULO 3
MODELLING WETLAND VEGETATION DYNAMICS BASED ON
SPATIO-TEMPORAL INTERACTION AND FLOODING
TOLERANCE OF PLANT COMMUNITIES IN THE PANTANAL
MATOGROSSENSE (BRAZIL)
Com contribuição de: Derek Karssenberg e Cátia Nunes da Cunha
Abstract Predict whether and how vegetation will change in response to
environment shifts is a current issue for ecologists due to strong alterations in
habitat quality by human activities. Vegetation dynamics in wetland ecosystems
are considered associated to hydrological dynamics. However, spatial
interaction among neighboring species can also drive vegetation to different
states and transitions because of the complex nature of ecosystem dynamics.
Understanding the mechanisms involved in vegetation dynamics usually needs
to account to long-term and detail vegetation records that are not always
available. Modelling approaches have been developed to help us to describe
and interpret complex ecosystem behaviours. Spatiotemporal Markov Chain
model is used to implement and test our conceptual vegetation successional
model for the Pantanal wetland vegetation. STMC aggregates the main aspects
78
78
of two other models used to study vegetation dynamics: cellular automaton (CA)
and Markov Chain models (MC). The conceptual successional model is
developed based on literature and expert knowledge about physiological plant
limits, life history traits and ecosystem functioning. The mathematical
formalization of the conceptual model is done by giving transition probabilities
among vegetation states based on spatial effect among neighboring
communities and environmental effect over vegetation development. We
calibrate the model by comparing observed and simulated spatial patterns using
landscape indices. Space-time vegetation patterns are assessed and insights
are gained about the underlying mechanism of vegetation dynamics. We
observed that the transitions among vegetation states are controlled by the
flooding condition and the diversity of neighbors. Intermediary flood conditions
and high neighbor diversity are associated to less stable vegetation dynamics.
Changes in space-time vegetation patterns under four flooding change
scenarios are forecasted. We found that extreme drought scenario exerts the
strongest impact on the vegetation of the studied landscape, modifying
substantially the spatial pattern of vegetation distribution. The model developed
in this study allowed us to explore some hypotheses about vegetation change
into the Pantanal and compare our findings with those found in other wetlands.
Key words: 1.- cellular automata; 2.- scenarios; 3.- vegetation change; 4.-
floodplain; 5.- dynamic modelling; 6.- Markovian processes; 7.- climate
changes, 8.- flooding, 9.- succession
79
79
3.1. Introduction
The sensitivity of wetland ecosystems to climate oscillations is a key
element for understanding vegetation dynamics (Junk et al. 2006b). Because
the dynamics of wetland ecosystems are intrinsically connected to hydrological
regime, changes in precipitation and evapotranspiration patterns, that affect the
water balance in the system (Junk 2002), will very likely change the realized
environmental space (Jackson and Overpack 2000) (e.g. flood spatial patterns)
and, consequently, may modify the spatial pattern of the vegetation. However,
we are not sure when and how these changes will affect vegetation community
distribution. This will not only depend on the magnitude and rates of
environmental change, but, also, on species characteristics, such as dispersion
ability and tolerance to new environmental states (Jackson and Overpeck 2000,
Walther et al. 2002, Davis et al. 2005).
Vegetation communities are not static entities, they rather change
frequently over time and space, adjusting to a fluctuating environment. These
changes may be caused by either, plant interactions through mechanisms of
facilitation, tolerance or inhibition (i.e. autogenic succession), or external forces,
such as flood events (i.e. allogenic succession) (Connell and Slatyer 1977). The
strength, frequency and direction of environmental shifts may determine the
permanence (i.e. length of time before a change occurs) of communities and
the successional direction of the change, sometimes resulting in continuous and
progressive change, others, setting back certain characteristics formerly
acquired by the community, such as diversity and productivity (Whittaker 1967,
Connell and Slatyer 1977). The result of such a dynamic system is that, under a
80
80
general expected trend in vegetation shifts, such as those governed by broad-
scale climate conditions, the complex patterns that emerge from species
interactions may cause uncertainty in vegetation response to abiotic changes
(Jackson and Overpack 2000, Ives et al. 2007).
Mathematical formalization of descriptive successional models has been
developed by ecologists, over the last decades, aiming at testing hypotheses
and making predictions about future states and distribution of vegetation under
different environmental scenarios (Hulst 1980, Usher 1981, Baptist et al. 2006).
Different approaches are used to model vegetation dynamics (Hulst 1979,
Acevedo et al. 1996, Baltzer et al. 1998, Caswell and Etter 1999, Reynolds et
al. 2001, Yemshanov and Perera 2002, Baptist et al 2006, Scheller and
Mladenoff 2007). Gap models, for instance, focus on plant individuals, requiring
detail characteristics on establishment, growth and mortality to describe
vegetation dynamics in small patch sizes (Reynolds et al. 2001). Contrarily,
transition or Markovian models focus on the orderly of succession processes,
using transition probabilities among discrete vegetation entities (Usher 1981).
Individual-based models, as is the case of gap models, despite bringing an
opportunity to make fine-scale and, maybe, more realistic predictions about
changes (van der Valk 1981, Reynolds et al. 2001), may result in high
computational costs (Scheller and Mladenoff 2007); beside, because it requires
large number and high quality data, may eventually become too complex to
have practical applications (Usher 1981). Contrarily, Markovian models,
although using simple representations of processes, have shown to be a
valuable approach to interpret and compare mechanisms of succession (Hulst
81
81
1980, Usher 1981) and predict the impact of different scenarios of external
forcing, namely, climate change, fire events or management activities, on
vegetation dynamics (Iverson and Prasad 2001, Mouillot et al. 2002, Hamann
and Wang 2006). Hybrid approaches that combine the strongest aspects of
different model types have also been developed (Acevedo et al. 1996, Baltzer
et al. 1998, Reynolds et al. 2001). Spatio-Temporal Markov Chain model
(STMC) is an example of that (Baltzer et al. 1998). By including spatial aspect
of cellular automaton theory in the Markovian succession framework, STMC is
able to mimic complex system behaviours (Wolfram 1984, Baltzer et al. 1998).
The present study is the first attempt to describe and understand spatio-
temporal dynamics in vegetation communities in the Pantanal Mato-grossense
(Mato Grosso, Brazil) using a spatially explicit model. This will be done through:
(1) the creation of a conceptual successional model, considering the effects of
flood duration and spatial interactions between neighboring communities on
vegetation changes; (2) the mathematical formalization of this theoretical
framework using spatio-temporal Markov chain model; (3) the calibration of
model parameters by comparing simulated and observed vegetation patterns;
(4) the comparison between ecological concepts and model behaviour, and (5)
the investigation of impacts of different flood scenarios on current vegetation
patterns.
The Pantanal Mato-grossense, one of the largest floodplain wetland in
the world, is internationally recognized as a region of overwhelming value and
high conservation priority (Ramsar Information, Bureau 1998; Junk and Nunes
da Cunha 2005). This region encompasses a great variety of aquatic, terrestrial
82
82
and transitional aquatic-terrestrial ecosystems (ATTZ) (Junk et al. 1989). In
wetlands, such as the Pantanal, flood conditions are important environmental
filters to plant establishment (Keddy et al. 2009), resulting in successive or
cyclical changes among vegetation states, that are associated with dispersion
strategies, growth rates and tolerance to stress (van der Valk 1981, Casanova
and Brock 2000). Frequent climate fluctuations occurred in the Pantanal during
the Quaternary, resulting in alternation between more humid and dry
environments (Assine and Soares 2004). Current climate changes, however,
are predicted to affect low latitudes by increasing year-to-year variations in
precipitation and causing heavy drought and flood events (Junk 2002). The
short and long-term consequences of these events on composition and spatial
distribution of vegetation are not fully understood. Understanding how sensitive
vegetation patterns are to different flood scenarios is one of the objectives of
this work.
3.2. Study area
Abiotic characterization
The Pantanal contains a large variety of alluvial ecosystems with different
drainage patterns, flooding characteristics, geomorphologic aspects and
vegetation types covering about 150,000 km2 of the upper Paraguay basin (Fig.
3.1) (Assine and Soares 2004). The climate of this region is tropical humid with
marked seasonality between winter and summer periods (Köppen 1948). The
summer from November to April is characterized by high temperatures (average
day temperature 34oC) and it is the season with the largest amount of
83
83
precipitation (Fig. 3.1). The precipitation decreases in winter, causing this
season to be very dry (de Musis et al. 1997). The water level in the rivers of the
Pantanal follows the seasonal trend in the precipitation. Due to the poor surface
and subsurface drainage and the smooth, low topography relative to the river
level (Alvarenga et al. 1984, Assine and Soares 2004), large areas of the
Pantanal are flooded every summer (Junk 1993, Hamilton et al. 1997). Climate
oscillations have been shown to be the main cause of the observed multi-year
period of cyclic variation in flooding (Junk et al. 2006a).
Figure 3.1 Study site. Natural Reserve SESC Pantanal located at the Pantanal
Mato-grossense, Mato Grosso, Brazil. The water depth in the rivers of the
BRAZIL
Pantanal
N
Nature Reserve
River S
ão Lo
urenço
SCALE
Rive
r Cui
abá
0 105 Km
Pantanal
MT
MS
16o
21o
55o58o
16o
15o
100200300400500
0N D J F M A M J J A S O
Precipitation, mm
1.436 mm
Depth, cm
84
84
Pantanal follows the seasonal trend in the precipitation, resulting in river water
overflow. Mean annual water depth fluctuation of the River Cuiabá (1963-2000)
and mean precipitation near Cuiabá are provided by INMET (National Institute
of Meteorology of Brazil) and river level data by DNAEE (National Department
of Waters and Electric Energy of Brazil).
Vegetation
The Pantanal vegetation presents floristic elements of three important
morphoclimatic and phytogeographic domains, i.e., Cerrado (Brazilian
savanna), Amazonia and Chaco (Ab`Saber 1988). Savanna vegetation types
are dominant physiognomies in the Pantanal (67%), but are not the only one:
semideciduous forest, gallery forest, swamp, Chaco, pioneer formations such
as monodominant forest of Vochysia divergens Pohl (Silva et al. 2000) are the
remaining components of the vegetation mosaic. The variability in water depth
and flooding duration and the temporal connections and disconnection
established between different elements of the landscape by means of the flood
pulse (Junk et al 1989) are considered the preponderant causes of the high
diversity of biological communities in the Pantanal (Wantzen et al. 2005),
dictating where and when plant species with different life strategies and flooding
tolerance will appear (Junk et al. 2006a).
The impact of historical land use on the current landscape pattern of the
Pantanal is not fully understood. However, ecosystem functionality seems to
have maintained untouched over the last Centuries. There are three main
reasons that ensured the Pantanal conservation: extensive cattle ranching as
85
85
the dominant economic activity over the last 300 years; its countryside location;
and, last, the existence of annual flood events (Pott and Pott 2004, Junk and
Nunes da Cunha 2005).
Studied site We have taken a 60 km2 floodplain located at a private nature reserve
(RPPN SESC Pantanal) in the Northern Pantanal (16 o 30’ – 16o 44’S and 56 o
20’– 56o 30’W), Mato Grosso, Brazil (Fig. 3.1), as study area to describe and
test a conceptual model on vegetation succession in ATTZs. Since its creation,
in 1998, this reserve is mainly used for scientific purposes. This site is
representative of a large part of the Pantanal, regarding vegetation and
environmental conditions. The fluctuation in annual water level in the river
Cuiabá is the main cause of periodic flooding on the floodplain.
The vegetation and flooding duration maps of the studied site were
provided by Arieira et al. (in preparation) (Fig. 3.2). This study uses
sophisticated statistical classification, interpolation and error propagation
techniques in order to describe the spatial patterns in wetland vegetation
communities and flooding duration. The study includes the description and
analysis of field sampling and remotely sensed data, the classification of
vegetation communities, the universal kriging procedure to map the
communities, and an evaluation of relations between mapped vegetation and
flooding duration. In addition, it describes an uncertainty analysis to describe
the reliability of the findings.
86
86
Figure 3.2. (A) Predicted distribution of the plant communities and (B) spatial
pattern of flood duration on the study floodplain, identified at the study by Arieira
et al. (in preparation).
A
16 46’So
56 18’Wo 56 23’Wo
16 57’So
0 500 1000 meter
Grassland
Secondary forest
Alluvial semideciduous forest
Shrubylands
Low tree and shrub savanna
Monodominant forest of PohlVochysisa divergens
Savanna forest
%
%
%
%
%
%
%
LEGEND365
0
flood duration
B
87
87
3.3. Spatio-temporal Markov chains
The conceptual framework of vegetation succession was implemented
and tested using Spatio-Temporal Markov Chain (STMC) (Baltzer et al. 1998).
STMC aggregates the main aspects of two other models frequently used to
model vegetation dynamic: cellular automaton (CA) and Markov Chain models
(MC). The capability of CA mimicking complex ecosystem behaviors in a
relatively simple mathematical way (Wolfram 1984) added to easy
representation in MC of succession development in a transition probability
matrix (Hulst 1979), make STMS a good candidate to model vegetation
dynamics in this study. Besides, because STMC is able to incorporate two
important dimensions of ecological processes, spatial dependence and
temporal dependence (Baltzer et al. 1998), stochastic processes such as plant
interaction and dispersion mechanisms (Logofet and Lesnaya 2000), as well as
the deterministic aspect linked to successional evolution (i.e. future stages
depend on past stages) may be accommodated into the model (Colasanti and
Grime 1993).
The construction of the STMC follows three steps: first, conceptual model
of vegetation succession is described based on expert knowledge and
literature; second, this theoretical model is transformed in mathematical
statements; and third, the model parameters are calibrated using inverse
modelling.
88
88
Conceptual successional model
Principles The present study proposes a conceptual model on vegetation
development in ATTZs in the Northern Pantanal. The framework of this
successional model was developed based on expert knowledge, field
observations, and literature (cf. Yemshanov and Perera. 2002). Although the
model concepts were created based on generic concepts on the vegetation
dynamics of the Pantanal, vegetation communities used in this study to illustrate
such dynamics are somewhat specific for our research area. Succession was
conceived here as a mechanistic process, by modelling the chances of seral
community types to change into a place as a result of the environmental
condition and spatiotemporal biotic interactions (Guisan and Zimmermann
2000).
The conceptual model is created based on a number of rules that drive
spatio-temporal vegetation changes and bound our interpretation of the model
outputs. These are: 1) vegetation communities (VC) are discrete entities; 2)
there is a fixed set of possible transitions between VCs, i.e. ‘succession’; 3)
speed of successional development is given by transition probabilities; 4)
transition probabilities depend on environmental site condition and
neighborhood effect.
Rule 1: Aggregation of species according to some defined criterion (e.g.
regeneration requirement) can be a big deal in modelling of tropical vegetation
types because of its rich flora (Acevedo et al. 1996). Discrete states derived
from this aggregation can be defined according to species composition,
89
89
successional stages, vegetation classification units or land cover types
(Yemshanov and Perera. 2002). Succession was here approached at broad
level, where few discrete vegetation states, here considered seral community
types, are defined by their physiognomic aspect and dominant species.
Rule 2: We assumed that there is a limited number of discrete states in which
succession moves on and back, constrained by biological species life traits,
what allowed us to describe and interpret succession in a mechanistic and
feasible way (Moore 1990). With a finite number of states, Markov chains can
be created by determining an order followed in succession. In our model, late
successional stages have chance to return to initial ones, what allowed us to
simulate cyclical succession (Hulst 1979), influence by disturbance (i.e. fire,
flooding) and natural regeneration processes of vegetation.
Rule 3: Once established, a seral community type takes some time to develop
to another one over the successional course. This time lag before transition
depends on the life span, growth rate and mortality of the individuals of a
community (Moore 1990, Acevedo et al. 1996) and varies in response to the
flood regime (Parolin 2009). In such time-dependent Markov model, transition
probabilities are controlled by these waiting times (Logofet and Lesnaya 2000),
that, in turn, govern the speed of transitions. The long the waiting time is, the
lower the probability of change among states becomes.
Rule 4: External and internal ecological forces are important constraints for
plant establishment and successional development (Tilman 1994). Flood
duration is considered the main external cause of vegetation shifts in this study,
due to its influence in colonization, persistence, mortality and growth rates of
90
90
plants (van der Valk 1981, Kirkman et al. 2000, Kandus and Malvarez 2004,
Parolin 2009). Vegetation shifts can also be promoted by internal forces, such
as plant interactions (e.g. competition) and dispersal mechanisms (Grime 1979,
Tilman 1994). Both mechanisms are considered spatially dependent (Gardner
and Engelhardt 2008), because the chances of dispersion decreases with the
distance from the propagule source (Janzen 1970); and competition might occur
among species that share similar and limited resources (Colasanti and Grime
1993). These environmental and neighbouring effects are considered here the
main forces influencing transitions among vegetation states.
Vegetation states and transitions
Vegetation states are considered those vegetation types showed on the
existing map of vegetation (Fig. 3.2A). These vegetation types are frequently
found in other ATTZ in the Pantanal, what makes the studied site relevant to
test our conceptual successional model for ATTZ of the Pantanal. They were
identified based on dominant species and structural characteristics of different
plant functional groups (i.e. shrub, tree, herbs, vines) (Arieira et al. in
preparation) and represent physiognomies rather than plant associations.
The scheme of succession elaborated in this study is shown in figure 3.3.
Grassland represents the starting point of our successional model. It is
characterized by predominance of an herbaceous layer. Woody species are
completely absent or exist in very low densities. Presence in seed bank, high
growth rates and high fecundity makes herbaceous species to show high
regeneration ability for colonizing open areas and regenerating after
disturbance (Table 1). Despites we consider here Grassland (state 1) as a
91
91
single state, a variety of assemblages with differences in dominant species,
biomass content and density of woody species are included in this state. Mostly
composed by flood-tolerant species, such as Scleria leptostachya, Euphorbia
tymifolia, Setaria geniculata, Cyperus campestris, Paspalum hydrophyllum,
Panicum spp, Axonopus purpussi, Axonopus leptostachyus, Euphorbia
tymifolia, Grassland can occur in different positions on the flooding gradient. As
flooding duration decreases, density of woody species increases. This change
is usually associated to transitions from Grassland to other savanna types,
namely, Open savanna (state 2) and Dense savanna (state 5) (Ratter et al.
1988), suggesting that under long-lasting flooding, successional development
towards woody savanna may be delayed, or even avoided by keeping
succession in a cyclical dynamic (Kirkman et al. 2000). This manner, it is
expected that drier years in the Pantanal will favor colonization by savanna
pioneer woody species, such as Byrsonyma orbygniana, Curatella americana
and Sclerolobium aureum. Consequently Grassland (state 1) will be succeeded
by Open savanna (state 2) and Dense savanna (state 4), unless increasing of
fire events follows these dry years and prevent succession development
(Loehle and LeBlanc 1996). On another hand, wetter years may promote the
rapid colonization by pioneer and flood-tolerant species, pushing succession
towards two possible trajectories: 1) Grassland is succeeded by Monodominant
forest (State 6) due to the spread of the fast-growing and gap-requiring tree
species V. divergens (Table 1) (Nunes da Cunha and Junk 2004); 2) Grassland
is succeeded by Shrubland (state 3), due to the very dynamic response of shrub
species, such as Mimosa pellita, Laetia americana and Albizia polycephala, in
92
92
areas under strong disturbance. Once established, Shrubland can persist for
many decades (Nunes da Cunha and Leitão-Filho 2007), due to the inhibition
effect caused by the shrub layer over tree and herb establishment. Its stable
state can be disrupted when new disturbances displace large areas dominated
by this community. Unlike Shrubland, Monodominant forest persistence has no
long-term guarantee (Arieira and Nunes da Cunha 2006), since regeneration
ability of V. divergens is limited to gap areas and depends on seed availability.
Even though, succession of Monodominant forest upwards more mature
vegetation states has not been recorded yet. In spite of there may be
overlapping between Shrubland and Monodominant forest occupancy,
Shrubland appears in low-lies position, while Monodominant forest mostly
occupy intermediary inundation periods. Succession in Alluvial forest (state 7)
looks like a classical facilitation model of succession described by Connell and
Slatyer (1977), where a set of species tolerant to stress and with high
colonization ability such as Sapium obovatum (Wantzen et al. 2005) and
Cecropia pachystachya, favor the establishment of better competitor species
with slow growth rate, short dispersion capability and shade tolerant such as
Cupania castaneifolia, Trichilia catigua, Mouriri guianensis, Inga vera and
Brosimum lactescens. The community state present in the first moment is called
here Secondary forest (state 5), but is represented in the vegetation map by
Alluvial low forest due to its transitional state toward Alluvial forest (Pott 2007).
Due to different life strategies of the set of species belonged to these two
sequential stages, these stages are frequently found in different flooding
conditions. Secondary forest may succeed Grassland everywhere but its
93
93
chances of developing to Alluvial forest increases as the flooding duraton
decreases. Most of the species of Alluvial forest tolerates both short-time
waterlogging and periods of water deficiency (Damasceno-Junior et al 2005),
although severe droughts, or extreme inundation events, may result in
increasing of seedling mortality (Condit et al. 1995). Possible diebacks between
stages are considered in the model, illustrating a dynamic ecosystem influenced
by disturbance (i.e. fire) and regeneration cycles.
Figure 3.3 Conceptual model of vegetation dynamics on Aquatic-Terrestrial
Transitional Zones in the Pantanal Mato-grossense. Successional changes
(solid arrows) occur from an initial herb dominated stage toward tree dominated
stages. Disturbance, such as fire and exceptional flood events may set back
succession to previous stages (arrows in dotted lines). Transition probabilities
grassland
monodominant forest
open savanna
alluvial forest
secondary forest
dense savanna
1
2
4 5
6 7
scrubland3
30-yr / 40-yr
0.5 / 0.8
80-yr / 80-yr 30-yr / 300-yr
70-yr / 60-yr 180-yr / 100-yr
300-yr / 200-yr300-yr / 100-yr
0.6
/0.1
0.05 / 0
.6
0.28 / 0.14
0.95
/ 0.
4
0.99 / 0.9
0 .01
/ 0.
1 0.10
/0.
12
1.0 / 1.0
0.02 / 0.64
1.0 / 1.0
0.5 /0.2
1.0
/1.0
94
94
among vegetation states ( ( )jip lk ,→ ) and waiting times before transitions ( kw , in
years) vary according with the vegetation position on wetter (right side values)
or drier parts (left side values) of the flood duration gradient. The strength of the
neighboring effect on transition probabilities is determined by a neighborhood
effect parameter term ( m = 18).
Mathematical formalization of the conceptual model
Model structure
The mathematical formalization of the conceptual model begins with the
definition of the spatial scale to which we are going to look at the process. We
have used a 60 km2 grid with i row number and j column number subdivided
into cells of 40 m (L) x 40 m (L). Each cell (L2), located at a specific coordinate
(x,y), is occupied by a community/ state (I) that have probability of changing into
other states in discrete time intervals (t=1,2,…,n; years).
As mentioned earlier, transition probabilities among states are driven by
both: habitat environmental suitability (i.e. flood duration) and the spatial
interaction between neighboring cells. This environmental and spatial
dependence of transition probabilities enables that the parameter values into
the transition matrix change according with the environmental site condition and
number of neighbors, as it is going to be shown.
95
95
Table 2.1. Historical life traits and flood tolerance of characteristic species of the
seven successional states found at the experimental area. Adapted from
classification given by Budowski (1965) and Acevedo et al (1996).
Characteristic species Growth
speed Life span*
gap
requiring
Tolerance to
flooding
1. Grassland
Paspalum hydrophilum/
/Panicum spp/ Scleria
mitis
Axonopus purpusii/
Leersia hexandra
very fast very short, less
than 10 yr yes very tolerant
2. Open
savanna
Byrsonima orbygniana /
Annona cornifolia /
Axonopus purpusii
moderate usually 40-100
yr, some more yes Tolerant
3. Shrubland
Laetia americana /
Mimosa pellita / Peritassa
dulcis / Albizia
polycephala
fast very short, less
than 10 yr yes very tolerant
4. Secondary
forest
Crataeva tapia /
Ruprechtia brachysepala /
Sapium obovatum /
Cecropia pachystachya
fast usually 40-100
yr, some more yes Tolerant
5.
Monodominant
forest
Vochysia divergens/
Duroia duckei fast
usually 40-100
yr, some more yes Tolerant
6. Dense
savanna
Curatella americana/
Hymenae stigonocarpa /
Cordia glabrata/
Astronium fraxinifolium
slow very long, 100-
1000 no Tolerant/intolerant
7. Alluvial
Forest
Mouriri guianensis/
Ocotea diospyridolia/
Brosimum latescens
slow very long, 100-
1000 no Tolerant
96
96
The number of input parameters used in the model is initially selected
based on the understanding of the system acquired in literature (Ratter et al.
1981, Ab’Saber 1988, Ponce and Nunes da Cunha 1993, Zeilhofer and Schessl
1999, Nunes da Cunha and Junk 2004, Damasceno-Junior et al. 2005, Arieira
and Nunes da Cunha 2006, Nunes da Cunha et al. 2006, Nunes da Cunha and
Leitão-Filho 2007, Pott 2007) and provided by ‘expert knowledge’. Knowledge
about tolerance to flooding (flood-tolerant; intermediate; flood-intolerant) and life
traits of dominant species of communities/states (Table 1) are used to
characterize succession by determining the transition probabilities among
vegetation states, the strength of the neighboring effect (neighboring effect
parameter) and waiting times to transitions (Yemshanov and Perera 2002,
Weaver and Perera 2004).
Transition probabilities are calculated in three steps: 1) waiting times are
defined for each vegetation state on the basis of the life span and the flood
tolerance of dominant species; 2) the environmental effect is accommodated in
the model by calculating the transition probabilities as a function of the flood
duration; and last, 3) the probabilities calculated in the second step are adjusted
to include the neighboring effect on the transition probabilities.
Waiting times
We have included in the mathematical model the parameter waiting time
( wk (i, j) , years), with k = 1, 2, ..,n and n, the number of vegetation communities,
to control the time in which one state/community remains until it changes to
another (Loehle and LeBlanc 1996). Maximum (mk) and minimum (nk) waiting
97
97
times for each seral community are associated to the range of variation in flood
duration found in the study site, according with the impact of flooding on growth
rates and mortality of dominant species. These maximum and minimum values
are then used to calculate the final waiting time of each community in the cell,
as follows:
( ) ( ) ( )jiemnmjiw kkkk ,, ⋅−+= (1.1)
In Eq. (1.1), mk and nk are the values of the waiting time (years) for
community k for e(i,j) = 0 and e(i,j) = 1, respectively. e(i,j) correspond to
standardized flood duration values (d(i,j)) for each grid cell and are calculated
based on the accumulative influence of 38 years of flooding in the studied
floodplain and shown in figure 3.2B. Standardized flood duration (e(i,j), -) is
calculated as:
e(i, j) = d(i, j)− dmin( ) dmax − dmin( ) (1.2)
In Eq. (1.2), dmin is the minimum value of all d(i,j) values in the area, and
dmax the maximum value.
Environmental effect
The effect of flood duration on the transition probabilities among states is
included in the model by calculating probabilities as a function of the number of
98
98
flooded days per year (d(i,j), days/year) and the waiting time (wk(i,j)). With a
time step of a year, the probability pk→ k (i, j) that the cell remains unchanged is:
pk→ k (i, j) = 1− 1 wk (i, j) (1.3)
The probability that the cell changes into another vegetation community
is calculated as a linear function of the flood duration:
pk→ l (i, j) = 1− pk→ k (i, j)( )⋅ pk→ l ,e=0 + pk→ l ,e=1 − pk→ l ,e=0( )⋅ e(i, j)( ), for each l ≠ k
with (1.4)
pk→ l ,e=0l≠ k∑ = 1 , and
pk→ l ,e=1l≠ k∑ = 1
In Eq. (1.4), pk→ l ,e=0 and pk→ l ,e=1 are the transition probabilities of vegetation
community k to change to vegetation community l given that the cell changes to
another state (defined by Eq. 1.3), for e(i,j) = 0 and e(i,j) = 1, respectively.
Neighboring effect
The neighboring effect is mathematically formalized by adjusting the
transition probabilities previously determined in the Eq. (1.3) and (1.4). First, for
each vegetation community k, the probability of transition to each of the
vegetation communities l, l=1,2,.., n is adjusted relative to the conditions in the
99
99
neighbourhood. This is calculated for each cell, resulting in adjusted values
ak→ l (i, j)
ak→ l (i, j) = pk→ l ⋅ 1+ nl (i, j)( )⋅m , for each l (1.5)
with, nl(i,j) the number of cells with vegetation community l in the neighborhood
of the cell (i,j) under consideration. The neighborhood is defined as cells with a
spatial index (i+1,j), (i-1,j), (i,j+1), and (i,j-1). In Eq. (1.5), m is a neighborhood
effect parameter that control the strength of the neighboring influence on the
transition probability from k to l. The higher m is, the higher the neighboring
effect will be. It is assumed to be the same for all transition probabilities. Finally,
the ak→ l (i, j) values are standardized, resulting in transition probabilities
p*k→ l (i, j) , i.e. the transition probability for each cell from class k to class l, taking
into account the state of the neighboring cells:
p*k→ l (i, j) =
ak→ l (i, j)
ak→ l (i, j)l=1
n
∑ , for each k and l (1.6)
The transition matrix is implemented using Python programming language (cf.
Karssenberg et al. 2007) and run in the program PCRaster (PCRaster 2002).
100
100
Neighboring effect
Spatial interaction between neighboring communities was
accommodated in our model by stating that the chance of a certain state
persists in the same state over time increases as the number of neighbors in
the same state increases. We consider the influence of the four cell neighbors
(first order neighborhood). This is mathematically formalized by adjusting the
transition probabilities, previously determined. First, for each vegetation
community k, the probability of transition to each of the vegetation communities
l, l=1,2,.., n is adjusted relative to the conditions in the neighbourhood. This is
calculated for each cell, resulting in adjusted values ak→ l (i, j)
ak→ l (i, j) = pk→ l ⋅ 1+ nl (i, j)( )⋅m , for each l (1.5)
with, nl(i,j) the number of cells with vegetation community l in the neighborhood
of the cell (i,j) under consideration. The neighborhood is defined as cells with a
spatial index (i+1,j), (i-1,j), (i,j+1), and (i,j-1). In Eq. 1.5, m is a neighborhood
effect parameter. It is assumed to be the same for all transition probabilities.
Finally, the ak→ l (i, j) values are standardized, resulting in transition probabilities
p*k→ l (i, j) , i.e. the transition probability for each cell from class k to class l, taking
into account the state of the neighboring cells:
101
101
p*k→ l (i, j) =
ak→ l (i, j)
ak→ l (i, j)l=1
n
∑ , for each k (1.6)
The model was implemented using Python programming language (cf.
Karssenberg et al. 2007) and run in the GIS program PCRaster.
3.4. Model calibration
Procedure
Calibration of model parameters is an important step in modelling
procedures, providing parameter values best fitted to the reality. The model
parameters are calibrated using inverse modelling (IM). IM is suggested as the
only method available for model calibration in the absence of empirical
information on parameter values (Karssenberg 2002). Unlikely transitions, as
between Shrubland and Dense savanna, are considered fixed values in the
transition matrix and are not calibrated. IM yields posterior calibrated distribution
of the model parameters through an iteration of three steps: 1) selection of a set
of input maps (i.e. flooding duration map) and parameters (i.e. previous
parameters) for the dynamic model, 2) running the dynamic model with this set
of input data and parameters; 3) assessment of the model performance by
verifying the degree of agreement between model and observed outputs
(Hamann and Wang 2006). The observed output consists of the vegetation map
shown in figure 3.2A. This comparison between observed and model outputs
indicates if the model parameters are well adjusted or not. If the answer is
negative, the model parameters are changed and there is a loop back to step 1.
102
102
The number of iterations can be reduced when results of previous iterations are
used in a better selection of inputs and parameters in step 1.
The comparison of the outputs is performed by quantifying the spatial
heterogeneity of the observed and modeled vegetation maps using: the fraction
of occupied cells per state/community along the flood duration gradient, the total
occupied area and the mean patch size of each state. The landscape patterns
are quantified using the software PCRaster. Mean patch size is calculated by
dividing total landscape per number of patches. Because vegetation patterns in
the studied area are associated to ecological processes, such as biotic
interactions and flood constraints (Arieira et al. in preparation), we assume that
similarities between observed and simulated patterns indicate that our
successional model efficiently represents the undergoing ecological processes
conducting vegetation development toward the current landscape.
Model performance
The calibrated model parameters are shown in figure 3.3. They
generated acceptable model predictions. The model runs cover a period of
15500 yr to reach a similar spatial pattern as the observed (Fig. 3.4). This long
time for achieving the observed pattern might be related to our assumption of a
dynamic steady state controlled by disturbance, and the large experimental
area used for modeling.
103
103
Figure 3.4. (A) Spatial pattern of community distribution identified by Arieira et
al. (in preparation) and (B) resulted from our model. Uncertainty in vegetation
classification of the map in A is shown in (C), as two maps resulted from Monte
Carlo simulation (see Arieira et al. in preparation).
56 18’Wo 56 23’Wo
0 500 1000 meter
16 46’So
56 18’Wo 56 23’Wo
16 57’So
N
104
104
The model seems to incorporate the main ecological forces driving succession
in the study ATTZ. The distribution of communities along the flood duration
gradient is represented satisfactorily in the modeled map (Fig. 3.5).
Figure 3.5. Model calibration. Comparison between original (bar) and modeled
(line) fraction of occupied area (log scale) by each vegetation states at different
classes of flood duration (monthly intervals).
Grassland
Secondary forest
Alluvial forestShrubland
Open savanna
Monodominant forest
Dense savanna
frac
tion
of o
ccup
ied
area
(log
10(n
))
flood duration class0-2 >2-4 >4-6 >6-8 >8
1
0
10
100
frac
tion
of o
ccup
ied
area
(log
10(n
))
flood duration class
1
0
10
100
frac
tion
of o
ccup
ied
are
a (lo
g10(
n))
flood duration class
1
0
10
100
frac
tion
of o
ccup
ied
are
a (lo
g10(
n))
flood duration class
1
0
10
100
frac
tion
of o
ccup
ied
area
(log
10(n
))
flood duration class
1
0
10
100
frac
tion
of o
ccup
ied
area
(log
10(n
))
flood duration class
1
0
10
100
frac
tion
of o
ccup
ied
area
(log
10(n
))
flood duration class
1
0
10
100
0-2 >2-4 >4-6 >6-8 >8
0-2 >2-4 >4-6 >6-8 >80-2 >2-4 >4-6 >6-8 >8
0-2 >2-4 >4-6 >6-8 >8 0-2 >2-4 >4-6 >6-8 >8
0-2 >2-4 >4-6 >6-8 >8
105
105
The observed total fraction of the landscape occupied by each state (95 to 2127
ha) does not differ substantially from the simulated (23 - 2660 ha) (Fig. 3.6A). In
contrast, simulated patterns yielded mean patch sizes ten times higher than the
observed (Fig. 3.6B). Differences between simulated and observed spatial
patterns indicate some restrictions of the model and suggest that further
calibration can be necessary. The relatively simple mathematical formalization
of our succession model, related to the limited number of parameters used, may
have restricted the model representation of complex vegetation dynamics.
Including new parameters into the model could have improved the model ability
to simulate vegetation dynamics, but at the expense of an increasing in model
complexity and time spent in calibration.
Figure 3.6. Comparison of spatial patterns between the original and modeled
distribution of community states. A) total occupied area; B)mean patch size.
A B
106
106
3.5. Model spatiotemporal behaviour
The ability to highlight emergent properties of ecological systems,
through few set of rules determining the interactions among the system
components is an important contribution of spatially explicit models (Scheffer
2009). Emergent system properties, such as diversity and stability are broadly
supported by ecological theories (Hutchinson 1961, Tilman 1994, Scheffer
2009), but the examination of the theory applicability is still an ecological issue.
We use theoretical background on Ecology to discuss the space-time behaviour
of our model. The model behaviour is examined by observing the frequency of
transitions among vegetation states, each year, over 5000 years (timesteps)
and the frequency distribution of ‘number of neighbors’ in different situations of
flood duration. Flood duration values on the map in Figure 3.2B are divided in
four classes separated at monthly flood intervals: class 1: from 0 to 2 months;
class 2: greater than 2 to 4 months; class 3: greater than 4 to 6 months; class 4:
greater than 6 months. We have used PCRaster to obtain values of frequency
of transitions and number of neighbors from grid cells of each flood class.
Figure 3.7 illustrates the space-time patterns resulted from this analysis.
Here, it is apparent that there is a trend in frequency distribution of number of
neighbors, accordingly with the cell position on the flood gradient. The longer
the flood duration is, the higher the proportion of cells completely surrounded by
the same vegetation state becomes (four neighbors). Contrarily, the highest
proportion of cells with the highest diversity of neighbors (number of neighbors
equal to zero) occurs where flooding lasts from 0 to 2 months. Intermediate
flood sites (flood class 2 and 3), i.e., sites where flooding lasts from 2 to 6
107
107
months dominate the study site and show more intense dynamics. This might
result from the more balanced transition probabilities among community states
(Fig. 3.3) at these intermediate areas. On the other hand, the slower vegetation
dynamics in places that are flooded over very long periods (>6 months year-1) or
short periods (from 0 to 2 months year-1) might be controlled by long waiting
times and high probability of transitions among few vegetation states, generally,
Grassland and Shrubland (Fig. 3.3). In response to this slow dynamics,
Shrubland creates an aggregated distribution pattern (Fig. 3.4) that, in turn,
increases its chance of permanence on the site. Regardless the flood condition,
the mean frequency of change among vegetation states over 5000-yr (Fig. 3.7)
was low, varying between 4.7 and 6 changes over 5000-yr. This result suggests
that the modeled system reached an overall stability.
108
108
Figure 3.7. Spatio-temporal model behaviour. Frequency of transitions
among vegetation states, each year, over 5000 years (dots) and frequency
distribution of ‘number of neighbors’ of 500 grid cells (bars), in four classes of
flood duration. Flood duration classes were derived from the map in Figure
3.2B: class 1: 0 to 2 months; class 2: greater than 2 to 4 months; class 3:
greater than 4 to 6 months; class4: greater than 6 months. Mean frequency of
changes among vegetation states is highest at intermediary flood sites.
109
109
3.6. Scenarios
Scenarios for flood regime change
We have used the calibrated succession model to simulate short (i.e.
less than 102 yr) and long-term (i.e. greater than 102 yr) vegetation responses to
flooding regime shifts. Due to the above-mentioned restrictions of our model,
simulation outputs might be interpreted as suggestive of the sensitivity of
vegetation to environmental changes, rather than providing an accurate
prognostic of the future states of vegetation (Reynolds et al. 2001). Currently,
projections of climate changes have been suggesting contrasting trends to the
Pantanal, what is related to uncertainties in model predictions (Milly et al. 2005,
Marengo 2008). Warmer temperatures and increase in river flow are two
possible perspectives (Hulme and Sheard 1999, Marengo 2008). Due to the
current doubts regarding the future environment of the Pantanal, we establish
flood scenarios to represent contrasting trends of changes in the hydrologic
regime (namely, duration of inundation). Four flood scenarios are defined: two
of them illustrate homogeneous spatial patterns of flood duration on the
floodplain: one represents a situation of average flood duration found in low-
lying flood zones, or wetter zones (WZ); and the other, in high-lying flood zones,
or drier zones (DZ). The two other scenarios illustrate historical scenarios into
the Pantanal, represented by a dry hydrologic year, recorded in 1971 (DY) and
a wet hydrologic year, recorded in 2006 (WY) (Fig. 3.8). Hydrological historical
data were provided by the Brazilian National Water Agency (ANA;
(http://hidroweb.ana.gov.br). The flood maps are created in the same way as
the flood map used to calibrate the successional model (Fig. 3.2B). The effects
110
110
of the scenarios of flood duration on vegetation dynamics are examined by
observing the fluctuation of the fraction of occupied sites (0≤F≤1) by each
community, at annual timescale.
Figure 3.8. Scenarios illustrating spatial patterns of flood duration on the study
site found in a historical dry year (A; 1971) and in a historical wet year (B;
2006). (C) Water level fluctuation in the River Cuiabá between 1969 and 2007 is
provided by Brazilian National Water Agency (ANA;
(http://hidroweb.ana.gov.br).
N
EW
S
111
111
Vegetation response to flood regime shifts
Spatially homogeneous scenarios
DZ - The greatest changes in the current patterns of community distribution
are observed in this scenario. A retraction of Shrubland is the first vegetation
response to the drier landscape (Fig. 3.9A). In the first 200-yr after
environmental shift, the fraction of occupied site by Shrubland drops from 0.48
to 0.18, and goes to zero in the subsequent years. Grassland gains extensive
areas just after retraction of Shrubland. The initial frequency of Grassland in the
studied site (F= 0.12) rises in the first 400-yr after the environmental shift (F=
0.48), but is reversed when new Grassland areas are succeeded by other
community states. The chances for the new occupants depend on both, how
close they are from the new empty areas (i.e. Grassland) and their tolerance to
drier habitats. Open savanna and Monodominant forest do not show any
change in response to the environmental shift for 200-yr, ever since Open
savanna becomes five times more frequent (F= 0.16) and Monodominant forest
duplicates its range (F= 0.64). Dense savanna presents a slightly expansion,
from 0.0068 to 0.0075, as well as Secondary forest, from 0.0022 to 0.0094.
Although the probability of establishment of Secondary forest is lower on drier
habitats, its waiting time is higher, what might have caused the increase of its
range. Otherwise, Alluvial forest remains almost unaffected by the new drier
situation, oscillating in gaining and loosing small portions of habitats.
WZ - When long-lasting flooding occurs everywhere in the landscape (scenario
WZ), changes in vegetation patterns are not as drastic as when flooded periods
are reduced everywhere (scenario DZ) (Fig. 3.9B). Grassland does not change
112
112
at all in the first 400-yr after the new wet scenario begins. Long waiting time of
Grassland at the wettest parts of the flood gradient holds the successional
development for 400-yr, ever since a slight retraction of Grassland is associated
with the expansion of Shrubland. The high probability of occupying the deepest
part of the flood gradient and the very long waiting time at this position (Fig.
3.3), led Shrubland to dominate more than half of the landscape in this
scenario. Contrarily, Open savanna remains in an unaltered situation for 150-yr.
After this time, its current proportion of occupied habitat (F= 0.03) is reduced
(F= 0.01). Dense savanna also shows a slight decline in its current occupancy
that might be linked to its shorter waiting time on wetter habitats. Similarly to
other communities, there is a time delay in the response of Monodominant
forest to the environmental shift. Only after 100-yr from the shift, Monodominant
forest retracts from 0.29 to 0.21. The fraction of occupied site of Secondary
forest falls from 0.0037 to 0.0029 over the first 100-yr and reaches 0.0007
during the next 900-yr. The frequency of Alluvial forest on the landscape keeps
almost unaltered for the first 200-yr after the beginning of the new scenario (F=
~ 0.05) and decreases (F= 0.01) at the subsequent years. The declines of
Secondary forest and Alluvial forest are related to their higher chances of dying
back to previous successional states and their lower waiting times on wetter
habitats.
113
113
Figure 3.9. Vegetation response to shifts in the hydrologic regime (namely,
duration of inundation) in the Pantanal; base realizations. A) spatially
homogeneous dry scenario (DZ), B) spatially homogeneous wet scenario (WZ).
Hydrological changes begin after 500 yr timesteps.
A
B0 400 800 1200
timestep (year)
0
0.2
0.4
0.6
0.8
1
fract
ion
ofoc
cupi
edsi
te
1500
b
a
c
df
e g
0 400 800 1200
tim es tep (y ea r )
0
0.2
0.4
0.6
0.8
1
fract
ionof
occu
pied
site
1500
c
b
fa
dge
GRASSLAND
MONODOMINANT FOREST
ALLUVIAL LOW FOREST
DENSE SAVANNA
ALLUVIAL FOREST
OPEN SAVANNA
SHRUBLAND
b
a
g
f
e
c
d
200-YR 400-YR 600-YR 800-YR 1000-YR
1km
50-YR
time (year)
114
114
Historical scenarios
DY – The vegetation changes observed in this scenario follows a similar trend
as the observed in DZ, but, here, the communities respond in a smoother way
to the environmental shift (Fig. 3.10A). Unlikely in DZ, Grassland oscillates in
gaining and losing areas for 200-yr after the hydrological dry year has begun,
indicating a rapid dynamic of creation and colonization of empty places. At the
end of 1000-yr, Grassland shows a slight increase of its occupied area.
Similarly to how we saw in DZ, the regret of Shrubland in this historical drought
scenario brings new opportunities for the landscape occupancy. After Shrubland
regression, vacant areas are most likely colonized by the nearest and abundant
neighbours, i.e., Monodominant forest (Fig. 3.10A). The spread of Open
savanna on the landscape is verified 800-yr after the environmental shift. This
expansion affects positively the expansion of Dense savanna. The slight decline
in occupied site by Secondary forest over the first 100-yr from the beginning of
the new scenario, is associated with its succession to Alluvial forest. The
frequency of Alluvial forest on the site is almost unaltered here.
WY – For this scenario, the changes in vegetation distribution are almost
imperceptible (Fig.3.10B). This might be related to the quite similar spatial
patterns of the flood maps representing this scenario (Fig. 3.8B) and that one
used to calibrate the model (Fig. 3.2B). The similarity between the flood maps
suggests that most of the past 38 years in the Pantanal evidenced large flood
events. The small vegetation changes observed under this scenario occur with
a delay of 200-yr, except for Shrubland that starts expanding after 50-yr. The
direction of the changes is very similar to that seen in the scenario WZ, where
115
115
A
B
0 400 800 1200
timestep (year)
0
0.2
0.4
0.6
0.8
1
fract
ion
ofoc
cupi
edsit
e
1500
c
b
f
g d
ae
0 400 800 1200
timestep (year)
0
0.2
0.4
0.6
0.8
1
fract
ion
ofoc
cupi
edsi
te
15001500
b
e dg
f
c
a
200-YR 400-YR 600-YR 800-YR 1000-YR
1km
50-YR
time (year)
GRASSLAND
MONODOMINANT FOREST
ALLUVIAL LOW FOREST
DENSE SAVANNA
ALLUVIAL FOREST
OPEN SAVANNA
SHRUBLAND
b
a
g
f
e
c
d
Secondary forest, Alluvial forest, Grassland, Open savanna and Dense savanna
regret, while Shrubland spread. But unlikely in WZ, here, Monodominant forest
shows an expansion that seems to be related to the maintenance of spatially
heterogeneous flood conditions on the landscape.
Figure 3.10. Vegetation response to shifts in the hydrologic regime (namely,
duration of inundation) in the Pantanal; base realizations: A) historical dry
scenario (DY; 1971); (B) historical wet scenario (WY; 2006); base realizations:
Hydrological changes are simulated after 500 yr timesteps.
116
116
3.7. Discussion
In this paper, we describe a vegetation successional model considering
spatial interaction between neighboring communities and duration of inundation
as important drivers of vegetation changes in the Pantanal. We seeks to gain
insights on wetland vegetation functioning by modelling succession as a
probabilistic and algorithm phenomena (Caswell and Etter 1999, Colosanti et al.
2007), extending the scope of Markov chain theory to spatio-temporal models
(Baltzer et al. 1998). We show that useful insights may be gained from simple
and few assumptions represented by transitional rules (Colosanti et al. 2007).
Despite our model may have some restricted application, because of the
coarse-scale representation of the system dynamic (Acevedo et al. 1996,
Baltzer et al. 1998), its practical methodology and easy computational
implementation make it a good alternative for modeling of vegetation dynamics,
when empirical data are not available. Literature and expert knowledge about
physiological plant limits, life history traits and ecosystem functioning are of
overwhelming importance in such a modeling approach, because these provide
information on the realized and potential range of vegetation responses to
environmental variability (Logofet and Lesnaya 2000), and, therefore, a more
realistic meaning for the model parameters.
The simulation of space time vegetation interactions with external (flooding)
and internal (neighborhood effect) ecological forces has resulted in an
understanding on how emergent system properties, such as landscape
diversity, stability and resilience may come up. These properties are related to
the capability of the system absorbing disturbance (Scheffer 2009). As
117
117
disturbance was included in our model by making transition probabilities vary
according with flood tolerance in dominant species, the resulting non-
homogeneous transition matrix could be used to simulate different vegetation
behaviors under different environmental conditions, and vegetation changes
resulted from environmental shifts (Logofet and Lesnaya 2000).
Spatial heterogeneity in flooding conditions was responsible for increasing
community diversity and dictate vegetation dynamics on the floodplain, because
of local uniqueness (Levin 1976) and the trade-off between the abilities of
community states to tolerate more or less floodable places (Luo et al., 2008).
This high community diversity illustrates non-equilibrium coexistence in a
fluctuating environment (Shimda and Ellner 1984). Vegetation dynamics derived
from environmental heterogeneity and neighboring interaction led the modeled
system to a quasi-steady state (Scheffer 2009). At intermediate flood sites,
neighboring interaction affected the system dynamics by getting faster or slower
the processes of establishment and extinction of patches (Gardner and
Engelhardt 2008). Aggregated distribution of communities, such as
Monodominant forest and Shrubland, that resulted from high transition
probabilities and long waiting times caused long residence time and frequent
diebacks in successional stages. Cyclic vegetation dynamics under long flooded
periods, as simulated here, are recorded in other wetlands (Kirkman et al. 2000,
Stroh et al 2008). Contrarily, long-term persistence of communities with patchy
distribution, as is the case of Dense savanna and Open savanna, might depend
strongly on how long adverse site condition lasts, the species tolerance to
118
118
disturbance, the plasticity of response to environmental variability and genetic
characteristics (Loehle and LeBlanc 1996).
The vegetation response to shifts in hydrological regime was assessed
assuming homogeneous and heterogeneous spatial patterns in flood duration.
Critical transitions between vegetation states were seen in spatially
homogeneous scenario (Fig. 3.9A, B), while spatially heterogeneity historical
scenarios (Fig. 3.10A, B) smoothed vegetation response to environmental shift.
Sharp shifts in vegetation patterns occurred when flood duration decreased
everywhere in the floodplain, suggesting that drought imposes a critical impact
on the current landscape stability. Large differences in transition probabilities
and waiting times between wet and dry habitats are the main reasons for these
observed changes in vegetation patterns. The impact of drought on wetland
ecosystems has been broadly recorded (Kirkman et al. 2000, Casey and Ewel
2006, Junk et al. 2006b, Stroh et al 2008). Stroh et al (2008) noticed an
expansion of Grassland dominated by perennial emergent grass and meadow
species and the following colonization for facultative tree species after long
drought events in a depression wetland in U.S. Southeastern Coastal Plain. A
very similar response pattern was verified here, where Grassland gained
extensive areas at the expense of Shrubland and were subsequently colonized
by the flood-tolerant species Vochysia divergens (Monodominant forest).
Although our model has included probability of diebacks in vegetation states
and short waiting times to simulate the impact of drought (fire) on community
dynamics in the Pantanal (Junk 2002), including explicitly fire behavior and
post-fire succession in the model (c.f. Hobbs 1983, Hirabayashi and Kasahara
119
119
1987, Aumann 2007) could have provided a more reliable evaluation on this
impact.
Gradual vegetation changes were associated with the direction and the
spatial aspect of the environmental shift. Spatially heterogeneous scenarios
tended to smooth vegetation response to environmental shifts. Homogeneously
wet scenario (WZ, Fig. 3.9B) also caused a gradual vegetation response, but in
this case, strong vegetation inertia was responsible for this response. Inertia in
vegetation is associated with the ability of species to tolerate moderate to short-
term environmental fluctuation (101-103), by controlling growth rates at adverse
conditions (Loehle and LeBlanc 1996, Jackson and Overpeck 2000). The model
parameter waiting time regulated this delay in vegetation response, avoiding
overestimation of the vegetation sensitivity to environmental changes (Loehle
and LeBlanc 1996).
Although we have shown in this study, the ability of the STMC model to
simulate complex vegetation behaviours, the lack of empirical data to calibrate
parameters and the use of simple model assumptions have limited its
applicability. Succession knowledge obtained from literature has been used for
model calibration (Yemshanov and Perera 2002), but parameter values
acquired from this source do not represent exact estimates (Logofet and
Lesnaya 2000). The model assumptions of static communities and limited
number of vegetation states have restricted model ability to make accurate
projections on vegetation changes (Scheller and Mladenoff 2007). Despite
these restrictions, the successional model developed here can be considered a
‘null’ model (Yemshanov and Perera 2002), by focusing on how neighboring
120
120
interaction and flooding operate on vegetation dynamics (Yemshanov and
Perera 2002, Scheller and Mladenoff 2007). To overcome the above-mentioned
model limitations further calibration and inclusion of new parameters in the
model may be necessary (Baltzer et al. 1998, Yemshanov and Perera 2002,
van Nes and Scheffer 2005). It is important to point out that, by increasing the
number of parameters in the model, its complexity is also increased. Beside,
calibration techniques based on historical data require time series data that may
not be available.
Vegetation modeling addressed to broad-scale processes, as shown in this
work, gives opportunity to explore hypothesis about how natural or human-
made impacts on ecosystem functioning may affect vegetation response
patterns, this manner, supporting decision making and policy. The relatively
pristine condition of the Pantanal area (Junk et al. 2006a) allows insights on
natural forces of vegetation dynamic to be gained and sustainable alternative of
management be traced. Modeling vegetation dynamics in the Pantanal is still a
challenge for scientists, because of the lack of empirical data on physiology and
ecology of plants to support predictions.
3.8. Acknowledgments
The authors are grateful to the Brazilian governmental agencies, CAPES
and CNPq, for the financial support. Helpful comments and assistance were
provided by P. Girard, Peter Zeihofer and Wolfgang Junk. We also thanks to the
Social Service of the Commerce (SESC) and technicians and students of the
Federal University of Mato Grosso for the technical support in field work.
121
121
CAPÍTULO 4
SÍNTESE
4.1. Definição do problema
Paisagens são compostas por sistemas complexos com estrutura e
dinâmica influenciadas por fatores bióticos e abióticos, agindo em múltiplas
escalas espaciais e temporais. Indicar fatores chaves que determinem
heterogeneidade espacial de paisagens e quantificar padrões espaço-
temporais em escalas pertinentes ao manejo e conservação da biodiversidade
são demandas atuais e urgentes.
Paisagens de áreas úmidas estão entre as mais suscetíveis a mudanças
climáticas, devido ao papel regulador da hidrologia no funcionamento dos
ecossistemas. Estas áreas possuem uma grande variedade de serviços
ambientais associados à biodiversidade existente, tais como abastecimento de
água, regulação climática e fornecimento de alimentos. Conservar a
biodiversidade destas áreas necessita do entendimento de como a
heterogeneidade da paisagem é controlada pelo regime natural de distúrbio.
Predições acuradas de estados e mudanças em vegetação de áreas úmidas
requerem o desenvolvimento de métodos eficientes à aquisição e
processamento de informações relevantes sobre a variabilidade espacial e
temporal da vegetação.
O presente trabalho teve como principais objetivos descrever padrões
espaço-temporais para comunidades de plantas do Pantanal e entender suas
causas, usando modelos espacialmente explícitos. Dois tipos de modelos
122
122
foram desenvolvidos: um modelo preditivo de distribuição espacial da
vegetação e um modelo baseado em processos, usado para simular dinâmica
da vegetação. Estes modelos integram diferentes fontes de dados de
vegetação e ambientais, provenientes de amostragem de campo, de imagens
de sensoriamento remoto e informações adquiridas da literatura. Além disso, o
procedimento de modelagem usado para descrever processos ecológicos,
identificar padrões e realizar predições conta com diferentes tipos de análise de
dados espaciais e não-espaciais, tais como análise multivariada, análise de
variograma, técnicas de interpolação e análise de ordenação. Incertezas
associadas à abordagem de modelagem para predições de distribuição e
dinâmica da vegetação são quantificadas, informando possíveis causas e
apontando formas de melhoramento dos modelos.
As principais questões científicas que este estudo se propôs a discutir
foram:
Como processos espaciais e fatores ambientais afetam padrões espaço-
temporais da vegetação do Pantanal e qual é a acurácea de modelos
espacialmente explícitos para descrever dinâmica e distribuição da
vegetação?
Estas questões foram discutidas ao longo do desenvolvimento da tese,
através de sub-questões feitas nos capítulos 2 e 3 e que serão apresentadas a
seguir, nos itens 4.2 e 4.3.
123
123
4.2. Integrando amostragem de campo, sensoriamento remoto e
estatística espacial para predizer distribuição de comunidades de plantas
no Pantanal
Comunidades de plantas podem ser identificadas como entidades
discretas, com base na descrição de atributos estruturais de cinco formas
de vida de plantas?
A primeira parte do capítulo 2 visou identificar comunidades de plantas
descritas por Nunes da Cunha et al. (2006) numa planície de inundação no
Pantanal, através da amostragem de atributos estruturais (i.e. biomassa, grau
de cobertura, altura de copa e outros) de cinco formas de vida. Análise de
fatores foi usada para resumir os dados multivariados e ajudar na identificação
das sete comunidades de plantas presentes na área estudada. O resultado
desta análise mostrou que comunidades de plantas podem ser identificadas
usando atributos estruturais de conjuntos de espécies, agrupadas de acordo
com suas formas de vida, mas também revelou a importância de se incluir
dados de espécies dominantes para distinguir comunidades estruturalmente
similares, mas floristicamente distintas. Apesar de comunidades de plantas
terem sido tratadas como entidades discretas, o que pode ser útil para fins de
mapeamento e manejo da vegetação, sobreposição de escores de diferentes
comunidades nos eixos de fatores apontou para a existência de mudanças
graduais, antes que abruptas, em vegetação. Isto resultou em incertezas na
124
124
classificação final da vegetação. A despeito disto, a suposição de comunidades
de plantas com limites bem definidos como um conceito espacial é, na maioria
dos casos, suficiente para interpretar padrões da vegetação, como neste
estudo.
Imagens de sensoriamento remoto conseguem capturar padrões de
vegetação amostrados em campo?
Qual a importância destas imagens em modelos preditivos de vegetação?
Imagens de sensoriamento remoto têm sido usadas para construção de
modelos preditivos da vegetação, devido à sua capacidade de detectar
variabilidade em atributos ambientais relacionados à vegetação. Neste estudo,
imagens IKONOS-2 e Modelo de Elevação Digital (MED/SRTM) ajudaram a
predizer a distribuição de comunidades de plantas sobre a paisagem estudada,
em função de suas correlações com padrões da vegetação observados em
campo. Topografia de copa (MED) ajudou na separação de áreas de alta e
baixa biomassa arbórea, como áreas de floresta e não-floresta. Uma imagem
do segundo componente principal, derivada da transformação das primeiras
quatro bandas IKONOS, capturou variações em riqueza e cobertura de
herbáceas e ajudou no mapeamento de comunidades de Campos. Tipos de
floresta, como Floresta Monodominante e Floresta Aluvial, foram
principalmente diferenciadas com auxílio das imagens de topografia de copa
(MED) e NDVI. As relações entre imagens e dados de campo foram
125
125
formalizadas usando modelos de regressão linear múltipla. Estes modelos
foram usados em um procedimento de krigagem universal, reduzindo as
incertezas nas comunidades mapeadas.
Dependência espacial em dados biológicos tem sido, crescentemente,
incorporada a modelos preditivos de vegetação, devido ao seu poder
explicativo sobre alguns padrões de distribuição da vegetação. A estrutura
espacial dos resíduos da regressão entre variáveis de imagem e de campo
(escores de fatores), calculada em análise de variograma, foi usada para
investigar a autocorrelação espacial, associada aos padrões de distribuição da
vegetação. Esta informação foi acomodada ao modelo de krigagem universal a
fim de fazer predições espaciais de vegetação, em locais não amostrados do
sítio de estudo. Incluir dependência espacial no modelo preditivo, além de dar
um peso mais realístico ao poder explicativo das imagens sobre os padrões
observados, possibilita que parte da variabilidade da vegetação, não explicada
pelas imagens, seja incorporada no modelo como resíduo espacialmente
correlacionado (Miller et al. 2007). A estrutura espacial dos resíduos mostrou
que gradientes de vegetação, representados por diferentes camadas da
vegetação (i.e. herbácea, arbustiva, arbórea), variam em diferentes escalas
espaciais. Pequeno alcance de dependência espacial, associado a um padrão
Como padrão de distribuição de comunidades de planta pode surgir em
ecossistemas complexos?
126
126
de distribuição em pequenas manchas de herbáceas e arbustos, pode ter
resultado de mecanismos de dispersão e interações biológicas, tais como
competição, ou estar associada a uma percepção detalhada de variabilidade
ambiental (Miller et al. 2007). Por outro lado, padrões de distribuição em ampla
escala foram associados a um contínuo aumento na variância e resultaram em
gradientes suaves de distribuição de árvores. Longos alcances de dependência
espacial relacionados à distribuição de árvores sobre a planície inundável
sugerem que variáveis explanatórias usadas no modelo podem estar co-
variando, espacialmente, com a distribuição das árvores; ou que variáveis que
explicam esta distribuição, não foram incluídas no modelo (Miller et al. 2007). A
combinação destes padrões de vegetação, variando em fina e ampla escala,
influenciados por diferentes forças ecológicas, foi responsável por criar uma
estrutura de paisagem complexa, enfatizando a natureza multi-escalar dos
mais altos níveis organizacionais biológicos, tais como comunidades e
ecossistemas.
Quais as vantagens da abordagem geoestatística para predizer distribuição
espacial da vegetação?
A grande vantagem da abordagem geoestatística para predição de
distribuição da vegetação é a possibilidade de se integrar informação biológica
detalhada com dados de alta qualidade, sensoriados remotamente. A
127
127
habilidade do modelo de fazer predições acuradas em padrões de vegetação
variou com o poder explicativo das imagens e a dependência espacial dos
dados, ambas incorporadas ao método de interpolação (krigagem universal).
Outra vantagem da abordagem geoestatística em mapeamento de vegetação é
que incertezas nos mapas produzidos podem ser quantificadas e
representadas em bases cartográficas. Diferente de abordagens clássicas para
mapear vegetação, os resultados da abordagem geo-estatística carregam
informação quantitativa sobre variabilidade da vegetação e apresentam um
caráter dinâmico, permitindo sua futura aplicação em modelos interessados na
resposta biológica a mudanças ambientais.
Quais são as principais fontes de incerteza no mapa de vegetação
produzido e como estas podem ser minimizadas?
Incertezas fazem parte de modelos que representam a realidade. Por
isso, perguntar sobre a existência de incertezas em modelos preditivos não é
relevante e, sim, de onde surgem, como podem ser quantificadas e
minimizadas. Neste estudo, incertezas na classificação da vegetação surgiram
durante todo o procedimento de mapeamento, desde a qualidade dos dados
adquiridos, até os erros provenientes das análises estatísticas e geoestatísticas
utilizadas. Cross-validation e simulação de Monte Carlo ajudaram a quantificar
estas incertezas e apontaram suas possíveis causas. A percentagem de
128
128
amostras classificadas corretamente em cross-validation variou entre 49% e
52%, indicando que configuração e densidade de amostras afetam a acurácea
das predições espaciais. Um mapa de incertezas, derivado de simulações de
Monte Carlo, mostrou que a qualidade da classificação varia espacialmente,
embora a proporção e o arranjo de comunidades observadas no mapa original
tenham sido preservados em grande parte. A quantificação e representação
cartográfica de incertezas, mostradas neste estudo, são ferramentas
interessantes para pesquisa e manejo, pois mostram o grau de confiança dos
dados e auxiliam a traçar estratégias para melhoramento dos mapas
produzidos. Aumentar o número de observações de campo; definir um desenho
amostral mais adequado às predições espaciais, que priorize a amostragem
das comunidades existentes na paisagem; incluir outras imagens relevantes à
explicação da variabilidade da vegetação no modelo preditivo - todas estas
estratégias poderiam melhorar a qualidade do mapa de vegetação produzido.
No entanto, os custos e benefícios associados ao aumento do esforço amostral
e à aquisição de novas imagens é algo que deve ser considerado em planos de
pesquisa.
Duração da inundação está relacionada ao padrão de distribuição das
comunidades de plantas na planície inundável?
A relação entre distribuição das comunidades de plantas e padrões de
duração de inundação foi avaliada através da comparação entre mapa de
129
129
vegetação e mapa de duração de inundação, de acordo com o método de
análise direta de gradiente. A distribuição preferencial de comunidades a certos
alcances do gradiente de inundação mostrou que duração de inundação é um
importante determinante da zonação de comunidades de plantas na planície.
Por outro lado, respostas não-lineares de comunidades ao gradiente de
inundação, como no caso de Campos, sugeriram que interações espaciais em
processos biológicos também podem influenciar certos padrões de distribuição
da vegetação na planície. Com base nos resultados obtidos, este estudo
sugere que zonação da vegetação na planície de inundação deve ser
influenciada, não apenas pelos limites fisiográficos ditados pela inundação,
como ressaltado na maioria dos estudos no Pantanal, mas também por
interações biológicas e mecanismos de dispersão. Isto ressalta a importância
de estudos de vegetação no Pantanal, focados na influência de processos
biológicos espaciais (competição, dispersão) nos padrões espaciais da
vegetação.
130
130
4.3. Modelando dinâmica de vegetação de áreas úmidas, baseado em
interações espaço-temporais e tolerância à inundação de comunidades de
plantas
Há informação disponível e suficiente sobre a vegetação do Pantanal
para modelar sua dinâmica?
A criação do modelo de sucessão vegetal descrito no capítulo 3 baseou-
se, principalmente, em dados de literatura e de especialistas, com relação à
dinâmica e características de vegetação de áreas úmidas. A criação de
modelos espaciais baseados em processos precisa estar embasada no
conhecimento dos mecanismos causadores de mudança na vegetação. Neste
estudo, informações disponíveis sobre traços de história de vida e tolerância à
inundação de espécies dominantes de comunidades de plantas do Pantanal
foram usadas para definir o efeito seletivo da inundação sobre estabelecimento
e permanência de estados da vegetação, e o efeito de vizinhança, simulando
dispersão local, sobre as chances de transição entre estes estados. A
escassez de dados sobre respostas fisiológicas, requerimentos ecológicos e
história de vida da grande parte das espécies de plantas do Pantanal, hoje,
limita o uso de abordagem de modelagem baseada no indivíduo, para estudar
mecanismos de sucessão. A falta de dados em séries temporais, sobre
mudanças da vegetação, restringe as possibilidades de calibração do modelo.
A despeito do modelo desenvolvido aqui apresentar restrições em sua
131
131
aplicação, devido à representação, em grossa-escala, da dinâmica do sistema,
sua metodologia, de cunho prático e de fácil implementação computacional,
torna-o uma boa alternativa para modelar dinâmica da vegetação, quando
dados empíricos não estão disponíveis. Somado a isto, com base em simples e
poucas regras ditando o comportamento do modelo, este trouxe importantes
‘insights’ sobre o papel da inundação e de interações espaciais sobre a
dinâmica da vegetação.
Como efeitos ambientais e espaciais afetam padrões espaço-temporais
da vegetação?
A influência da inundação e de interações espaciais sobre a dinâmica da
vegetação foi investigada, através do comportamento espacial e temporal do
modelo. Apesar de esta influência ter sido determinada a priori, durante a
formulação do modelo conceitual, padrões de vegetação não esperados podem
emergir durante as simulações. Esta investigação mostrou que
heterogeneidade espacial em condições da inundação pode aumentar a
diversidade de comunidades na planície de inundação, por causa das
habilidades competitivas de espécies em tolerar locais, mais ou menos
inundados. Isto sugere que, em caso de homogeneidade ambiental, a
paisagem seria ocupada por poucos estados de vegetação, com alta
probabilidade de estabelecimento; e aponta para a importância de formulações
espacialmente explicitas, em modelagem de dinâmica de vegetação. A
132
132
dinâmica da vegetação variou em função da posição no gradiente de duração
de inundação. Em locais sob mais forte distúrbio, devido à seca ou inundação
prolongada, dinâmicas sucessionais mais lentas e cíclicas devem predominar.
Estas dinâmicas resultaram dos longos tempos de espera, antes de transições
e altas probabilidades de estabelecimento de poucas comunidades.
Contrariamente, em posições intermediárias do gradiente de inundação,
probabilidades de transição mais balanceadas, entre as comunidades, levaram
a dinâmicas mais intensas de colonização e extinção de manchas. A inclusão
do efeito de vizinhança no modelo de sucessão, simulando mecanismos
biológicos dependentes do espaço (i.e. dispersão, competição), exerceu um
papel essencial na formação de manchas de vegetação e na criação do padrão
de distribuição espacial das comunidades na paisagem. Neste cenário, onde
heterogeneidade ambiental e interação espacial são importantes preditoras da
dinâmica do sistema, o futuro estado da paisagem, associado à capacidade da
vegetação de responder a freqüentes oscilações ambientais, dependerá,
fortemente, da configuração inicial da paisagem e da força e direção da
mudança ambiental.
Qual a validade do modelo criado?
Embora validação de modelos de simulação seja considerada, por
alguns cientistas, uma tarefa impraticável, por causa da impossibilidade de
simular dinâmicas complexas de sistemas naturais, ainda é uma etapa
133
133
indispensável na avaliação de modelos (Scheffer 2009). Apenas diante desta
‘validação’, modelos poderão predizer, com maior confiança, os impactos de
mudanças ambientais sobre a vegetação. Neste trabalho, a validação do
modelo de sucessão foi realizada, através da comparação quantitativa entre
padrões espaciais da vegetação, observados e simulados. O bom ajustamento
entre estes padrões indicou que o modelo incorpora forças ecológicas
importantes, que dirigem sucessão vegetal, em áreas de transição aquático-
terrestre (ATTZ) do Pantanal. A falta de dados empíricos para calibração de
parâmetros e as simples suposições do modelo limitaram sua previsibilidade;
afinal de contas, parâmetros estimados por ajustamento à realidade não
representam estimativas exatas, apenas aproximações (Logofet and Lesnaya
2000). A inclusão de novos parâmetros no modelo ou o uso de métodos de
calibração mais avançados, como calibração Bayesiana, são possíveis
maneiras de melhorar o modelo, mas os custos, em tempo e complexidade, ao
se aumentar número de parâmetros de um modelo é algo que sempre deve ser
considerado. A despeito destas restrições, o modelo de sucessão desenvolvido
nesta tese, deve ser pensado como um modelo ‘nulo’, cujo objetivo principal é
avaliar como interação entre vizinhança e duração da inundação podem afetar
dinâmica da vegetação em ATTZ no Pantanal.
Quais são os possíveis impactos de alteração no regime de inundação sobre
o padrão de distribuição da vegetação do Pantanal?
134
134
O modelo de sucessão da vegetação, desenvolvido e testado neste
trabalho, foi usado para simular os impactos de diferentes cenários de
inundação sobre a vegetação. Devido às restrições do modelo, os resultados
das simulações foram interpretados como sugestivos da sensibilidade da
vegetação a mudanças ambientais, antes que como verdadeiros prognósticos
de estados futuros da vegetação. Aqui foram avaliados os impactos de quatro
cenários de inundação: dois ilustram padrões de inundação espacialmente
homogêneos, representando condições mais secas e mais úmidas na planície;
e dois outros ilustram padrões de inundação representando anos históricos
mais secos (1971) e mais úmidos (2006) do Pantanal. Estas simulações
mostraram que a redução do período de inundação e a homogeneização das
condições ambientais sobre a paisagem devem trazer impactos mais críticos
sobre o estado atual da paisagem, mudando, substancialmente e de maneira
abrupta, sua configuração. Contrariamente, mudanças graduais em distribuição
da vegetação e atraso em resposta de comunidades às mudanças ambientais
ocorreram, tanto por causa da existência de heterogeneidade ambiental, bem
como pela ocasião de cenários mais úmidos encontrados na planície. Este
atraso simula a capacidade de espécies de áreas úmidas em absorverem
distúrbio causado pela inundação, resultando em maior inércia da vegetação.
Estes resultados sugerem que a dinâmica e a diversidade da vegetação do
Pantanal dependem fortemente da manutenção do pulso de inundação,
responsável por suas variações temporais e espaciais.
135
135
4.4. Considerações finais – impactos da pesquisa em futuros estudos e na
conservação dos recursos naturais do Pantanal
Ecólogos têm apresentado forte tendência a ver sistemas naturais como
resultantes de uma rede de processos atuando em múltiplas escalas e como
reflexo de múltiplas causas (Turner et al., 2001, Aumann, 2007). Planícies de
inundação permitem que estudos da vegetação sejam desenvolvidos sob uma
abordagem multi-escalar, destacando respostas florísticas diferenciais ao
distúrbio e, ao mesmo tempo, descrevendo e caracterizando comunidades de
plantas num contexto de paisagem (Gillet e Gallandat 1996). Representar tais
sistemas naturais complexos, através de modelos espaciais, é uma importante
contribuição ao planejamento, monitoramento e manejo da paisagem, pois,
com ajuda de uma série de ferramentas operacionais como SIG e
sensoriamento remoto, permitem analisar dados variados, comunicar
resultados em bases cartográficas, além de permitir predições, projeções e
definição de hipóteses sobre interações entre padrões e processos (Turner et
al., 2001; Scheller e Mladenoff, 2007).
A presente tese enfoca dois aspectos relevantes de pesquisa científica
no Pantanal: primeiro, a relação entre processos ecológicos, em particular, a
influência da inundação e de interações espaciais em processos biológicos,
nos padrões espaciais e temporais de comunidades de plantas na planície
inundável; e, segundo, o desenvolvimento de modelos, espacialmente
explícitos e dinâmicos, para descrever e examinar esta relação. Esta tese
contribui para o conhecimento da vegetação do Pantanal de diferentes
136
136
maneiras: 1) investigando métodos e ferramentas eficientes na descrição de
padrões espaço-temporais da vegetação; 2) indicando necessidades e ‘gaps’
de informação na modelagem da distribuição e dinâmica da vegetação; 3)
fazendo previsões dos impactos de mudança no regime de inundação sobre a
vegetação; 4) criando mapas detalhados de vegetação e inundação para
descrever padrões espaciais sobre a planície; 5) investigando a adequação de
amostragem de dados e classificação da vegetação, através da eficiente
aquisição e processamento de dados; 6) descobrindo escalas de variação
espacial de diferentes gradientes de vegetação e suas possíveis causas; e,
finalmente, 7) quantificando incertezas e avaliando a validade dos modelos de
vegetação e apontando estratégias para melhoramento dos mesmos.
Modelagem preditiva da vegetação é uma abordagem promissora para
descrição e avaliação de variabilidade espacial em dados biológicos de larga
escala, pois possibilita a integração de observações pontuais de campo e
informações contínuas de imagens de sensoriamento remoto. Modelos que
permitem a inclusão da dimensão espacial e temporal, para entender e
visualizar padrões resultantes da interação em processos ecológicos, em
escalas temporais e espaciais amplas, são ferramentas insubstituíveis que
ajudam a compreensão de comportamentos complexos de sistemas
ecológicos. Tomadas de decisão e definição de políticas em conservação e uso
de recursos naturais, calcados neste conhecimento, poderão tornar-se menos
arbitrárias e traçando um futuro menos incerto para este ecossistema
extraordinário que é o Pantanal Matogrossense.
137
137
LITERATURA CITADA
Ab’Saber, A.N., 1988. O Pantanal Matogrossense e a teoria dos refúgios. Rev.
Bras. Geog. 50, 9-57.
Acevedo, M. F., D. L. Urban, H.H. Shugart 1996. Transition and gap models of
forest dynamics. Ecological Applications, 5: 1040-1055.
Alvarenga, S.M., Brasil, A.E., Pinheiro, R., Kux, H. J. H., 1984. Estudo
geomorfológico aplicado à Bacia do alto Rio Paraguai e Pantanais
Matogrossenses. Boletim Técnico Projeto RADAM/BRASIL. Série
Geomorfologia, Salvador, 187: 89–183.
Arieira, J., C. Nunes da Cunha 2006. Fitossociologia de uma floresta inundável
monodominante de Vochysia divergens Pohl (Vochysiaceae), no
Pantanal Norte, MT, Brasil. Acta Botanica Brasílica, 20: 569-580.
Arieira, J., D. Karssenberg, S.M. De Jong, E.A. Addink, E.G. Couto, C. Nunes
da Cunha, J.O. Skøien (in preparation). Integrating field sampling, spatial
statistics and remote sensing to map floodplain vegetation in the
Pantanal, Brazil. Ecology.
Arieira, J., Karssenberg, D., Couto, E.G., Nunes da Cunha, C. (in preparation).
Modeling wetland vegetation dynamic based on spatio-temporal
interaction and flooding tolerance of plant communities in the Pantanal
Mato-Grossense (Brazil). PhD Thesis, Federal University of Mato
Grosso, Cuiabá, Brazil/ Utrecht University, Utrecht, The Netherlands.
138
138
Assine, M. L., and P. C. Soares. 2004. Quaternary of the Pantanal, west-central
Brazil. Quaternary International, 114: 23-34
Aumann, C. A. 2007. A methodology for developing simulation models of
complex systems. Ecological Modelling, 202: 385-396.
Austin, M.P., 1968. An ordination study of a chalk grassland community. J. Ecol.
56, 739-757.
Austin, M.P., 2002. Spatial prediction of species distribution: an interface
between ecological theory and statistical modelling. Ecol. Model. 157,
101-118.
Austin, M.P., Smith, T.M., 1989. A new model for the continuum concept.
Vegetatio 83, 35-47.
Balzter, H., P. W. Braun, and W. Kölhler, 1998. Cellular automata models for
vegetation dynamics. Ecological Modelling, 107:113–125.
Baptist, M. J., M. Haasnoot, P. Cornelissen, J. Icke, G. v. d. Wedden, H. J. d.
Vriend, and G. Gugic´. 2006. Flood detention, nature development and
water quality along the lowland river Sava, Croatia. Hydrobiology 565:
243–257.
Barbosa, R.I., Ferreira, C.A.C., 2004. Biomassa acima do solo de um
ecossistema de “campina” em Roraima, norte da Amazônia Brasileira.
Acta Amazon. 34, 577 - 586.
Bascompte, J., Solé, R.V., 1996. Habitat fragmentation and extinction
thresholds in spatially explicit models. J. Anim. Ecol. 65, 465-473.
139
139
Bascompte, J., R. V. Solé (eds.) 1998. Modeling spatiotemporal dynamics in
ecology. Springer, Berlin.
Bourennane, H., King, D., Couturier, A., Nicoullaud, B., Mary, B., Richard, G.,
2007. Uncertainty assessment of soil water content spatial patterns using
geostatistical simulations: an empirical comparison of a simulation
accounting for single attribute and a simulation accounting for secondary
information. Ecol. Model. 205, 323-335.
BRASIL, Ministério do Meio Ambiente dos Recursos Hídricos e da Amazônia
Legal. 1997. Plano de Conservação da Bacia do Alto Paraguai
(Pantanal) – PCBAP. Diagnóstico dos meios físicos e biótico, meio
biótico. Vol. 2, Tomo 3. MMA/PNMA, Brasília, 433p.
Bray, J.R., Curtis, J.T., 1957. An ordination of the upland forest communities of
southern Wisconsin. Ecol. Monogr. 27, 325-349.
Brzeziecki, B., Kienast, F., Wildi, O., 1993. A simulated map of the potential
natural forest vegetation of Switzerland. J. Veg. Sci. 4, 499-508.
Budowski, G. 1965. Distribution of tropical American trees in the light of
successional process. Turrialba 15: 40-42.
Bullock, J., 1996. Plants. In: Sutherland, W.J. (Ed.), Ecological census
techniques: a handbook. Cambridge Univ. Press, NY., pp. 111–138.
Bureau, R. I. 1998. What are wetlands? Ramsar Information Paper 1.
Burke, M. K., S. L. King, D. Gartner, and M. H. Eisenbies. 2003. Vegetation,
soil, and flooding relationships in a blackwater floodplain forest.
Wetlands, 23 (4): 988-1002.
140
140
Burrough, P., and R. McDonnell. 1998. Principles of Geographic Information
Systems. Oxford University Press, New York.
Casanova, M. T. and M. A. Brock 2000. How do depth, duration and frequency
of flooding influence the establishment of wetland plant communities?
Plant Ecology, 142: 237–250.
Casey, W. P., and K. C. Ewel. 2006. Patterns of succession in forested
depressional wetlands in north Florida, USA. Wetlands, 26:147–160.
Caswell, H., and R. Etter 1999. Cellular Automaton Models for Competition in
Patchy Environments: Facilitation, Inhibition, and Tolerance. Bulletin of
Mathematical Biology, 61(4): 625-649.
Chambers, J.Q., Asner, G.P., Morton, D.C., Morton, D.C., Anderson, L.O.,
Saatchi, S.S., Espírito-Santo, F.D.B., Souza Jr., M.P.C., 2007. Regional
ecosystem structure and function: ecological insights from remote
sensing of tropical forests. Trends Ecol. Evol. 22, 414-423.
Chave, J., Andalo, C., Brown, S., Cairns, M.A., Chambers, J.Q., Eamus, D.,
Folster, H., Fromard, F., Higuchi, N., Kira, T., Lescure, J.P., Nelson,
B.W., Ogawa, H., Puig, H., Riera, B., Yamakura, T., 2005. Tree allometry
and improved estimation of carbon stocks and balance in tropical forests.
Oecologia 145, 87-99.
Chong, G.W., Reich, R.M., Kalkhan, M.A., Stohlgren, T.J., 2001. New
approaches for sampling and modeling native and exotic plant species
richness. West. N. Am. Naturalist 61, 328–335.
141
141
Colasanti, R. L. and J. P. Grime 1993. Resource Dynamics and Vegetation
Processes: A Deterministic Model Using Two-Dimensional Cellular
Automata. Functional Ecology, 7(2): 169-176.
Colasanti, R. L., R. Hunt, and L. Watrud. 2007. A simple cellular automaton
model for high-level vegetation dynamics. Ecological Modelling 203:363–
374.
Condit, R., S. P. Hubbell, and R. B. Foster. 1995. Mortality Rates of 205
Neotropical Tree and Shrub Species and the Impact of a Severe
Drought. Ecological Monographs, 65: 419-439.
Congalton, R.G., Green, K., 1999. Assessing the accuracy of remotely sensed
data: principle and practices. Lewis Publishers, BocaRaton, USA.
Connell, J.H., Slatyer, R.O., 1977. Mechanisms of succession in natural
communities and their role in community stability and organization. Am.
Nat. 111, 1119-1144.
Coutinho, L.M., 1982. Ecological Effect of fire in Brazilian Dense savanna. In:
Huntley, B.J., Walter, B.H. (Eds.). Ecology of Tropical Savannas,
Ecological Studies, 42, Springer, Berlin. pp. 273-291.
Curtis, J.T., McIntosh, R.P., 1950. The interrelations of certain analytic and
synthetic phytosociological characters. Ecology 31, 434-455.
Da Silva, C. J., and P. Girard. 2004. New challenges in the management of the
Brazilian Pantanal and catchment area. Wetlands Ecology and
Management, 12: 553–561.
142
142
Damasceno-Junior, G.A., J. Semir, F.A.M. Santos, H.F. Leitão-Filho 2005.
Structure, distribution of species and inundation in a Alluvial forest of Rio
Paraguai, Pantanal, Brazil. Flora, 200:119-135.
Davis, M. B., R. G. Shaw, and J. R. Etterson. 2005. Evolutionary Responses to
Changing Climate. Ecology,86: 704-1714.
De Musis, C.R., Júnior, J.H.C., Filho, N.P., 1997. Caracterização climatológica
da Bacia do Alto Paraguai. Geografia 22, 5-21.
Decocq, G., 2002. Patterns of plant species and community diversity at different
organization levels in a forested riparian landscape. J. Veg. Sci. 13, 91-
106.
Draper, N.R., Smith, H., 1998. Applied Regression Analysis. John Wiley &
Sons, Inc., New York, USA.
Efron, B., Tibshirani, R., 1986. Bootstrap methods for standard errors,
confidence intervals, and other measures of statistical accuracy. Stat.
Sci. 1, 54-75.
Ferrier, S., Drielsma, M., Manion, G., Watson, G., 2002. Extended statistical
approaches to modelling spatial pattern in biodiversity in northeast New
South Wales. II. Community-level modelling. Biodivers. Conserv. 11,
2309-2338.
Forman, R.T.T., Godron, M., 1986. Landscape Ecology. John Wiley & Sons,
Inc. 603p
Fortin, M.J., Olson, R.J., Ferson, S. et al., 2000. Issues Related to the Detection
of Boundaries. Landscape Ecology 15: 453-466.
143
143
Fortin, M.-J., Olson, R. J., Ferson, S., Iverson, L., Hunsaker, C., Edwards, G.,
Levine, D., Butera, K., Klemas, V., 2000. Issues related to the detection
of boundaries. Landscape Ecol. 15, 453-466.
Gardner, R.H., Engelhardt, K.A.M., 2008. Spatial processes that maintain
biodiversity in plant communities. Perspect. Plant Ecol. Evol. Syst. 9,
211–228.
Gillet F., Gallandat, J.D., 1996. Integrated synusial phytosociology: some notes
on a new, multiscalar approach to vegetation analysis. J. Veg. Sci. 7, 13-
18.
Gluck, M., Rempel, R., Uhlig, P.W.C., 1996. An Evaluation of Remote Sensing
for Regional Wetland Mapping Applications. Forest Research report 137.
Ontario Forest Research Institute, Sault Ste Marie, Ontario, Canada.
Greig-Smith 1979. Pattern in vegetation. Journal of Ecology, 67: 755-779.
Grime, J. P. 1979. Plant Strategies and Vegetation processes, Chichester.New
york.Brismane.Toronto. pp. 222
Grime, J. P. 1979. Plant Strategies and Vegetation processes, Chichester.New
york.Brismane.Toronto. pp. 222
Grime, J. P. 1994. Plant Strategies and Vegetation Processes. John Wiley &
Sons, Inc. pp. 222.
Guisan, A., Thuiller, W., 2005. Predicting species distribution: offering more
than simple habitat models. Ecol. Lett. 8, 993-1009.
Guisan, A., Zimmermann, N.E., 2000. Predictive habitat distribution models in
ecology. Ecol. Model. 135, 147-186.
144
144
Hamann, A., and T. Wang. 2006. Potential Effects of Climate Change on
Ecosystem and Tree Species Distribution in British Columbia. Ecology
87:2773-2786.
Hamilton, S. K., S. J. Sippel and J. M. Melack, 1996. Inundation patterns in the
Pantanal wetland of South America determined from passive microwave
remote sensing. Archiv für Hydrobiologie, 137: 1–23.
Hamilton, S.K., Sippel, S.J., Calheiros, D.F., Melack, J.M., 1997. An anoxic
event and other biogeochemical effects of the Pantanal wetland on the
Paraguay River. Limnol. Oceanogr. 42, 257-272.
Hernandez-Stefanoni, J.L., Dupuy, J.M., 2007. Mapping species density of
trees, shrubs and vines in a tropical forest, using field measurements,
satellite multiespectral imagery and spatial interpolation. Biodivers.
Conserv. 16, 3817-3833.
Hiemstra, P.H., Pebesma, E.J., Twenhöfel, C.J.W., Heuvelink, G.B.M., 2008.
Real-time automatic interpolation of ambient gamma dose rates from the
Dutch Radioactivity Monitoring Network, Comput. Geosci. 35, 1711-
1721.
Hirabayashi, F., Kasahara, Y., 1987. A fire-spread simulation model developed
as an extension of a dynamic percolation process model
Simulation 49: 254-261
Hobbs, R. J., 1983. Markov models in the study of post-fire succession in
heathland community. Vegetatio, 56: 17-30
145
145
Hofmann, M., Gatu, C., Kontoghiorghes, E.J., 2007. Efficient algorithms for
computing the best subset regression models for large-scale problems.
Comput. Stat. Data An. 52, 16-29.
Hulme, M. and N. Sheard 1999. Climate Change Scenarios for Brazil Climatic
Research Unit, Norwich, UK, pp. 6
Hulst, R. 1979. On the dynamics of vegetation: Succession in model
communities. Plant Ecology 39:85-96.
Hulst, R. 1980. Vegetation dynamics or ecosystem dynamics: Dynamic
sufficiency in succession theory. Plant Ecology 43: 147-151.
Hutchinson, G. E. 1961. The paradox of the plankton. American Naturalist
95:137-145.
Iverson, L. R., and A. M. Prasad. 2001. Potential Changes in Tree Species
Richness and Forest Community Types following Climate Change.
Ecosystems 4: 186–199.
Ives, A. R., and S. R. Carpenter. 2007. Stability and Diversity of Ecosystems.
Science 317:58-62.
Jackson, S. T., and J. T. Overpeck. 2000. Responses of plant populations and
communities to environmental changes of the late Quaternary.
Paleobiology 26:194-220.
Jacobsen, K., 2006. SRTM Height Models. In: Thurston, Jeff (Ed.).
GEOconnexion International Magazine - Spatial data & landuse, 5. Mai
Ward, UK. pp. 20-21.
146
146
James, F.C., McCulloch, C.E., 1990. Multivariate analysis in ecology and
systematics: panacea or pandora's box? Annu. Rev. Ecol. Syst. 21, 129-
166.
Janzen, D. H. 1970. Herbivores and the Number of Tree Species in tropical
Forests. The American Naturalist 104:501-528.
Jensen, J.R., 2007. Remote Sensing of the Environment - an Earth Resource
Perspective. 2nd edition, Upper Saddle River: Prentice-Hall.
Jongman, R.H.G., ter Braak, C.J.F., Van Togeren, O.F.R., 1995. Data Analysis
in Community and Landscape Ecology, Cambridge at the University
press, Cambridge, UK.
Junk, W.J., 1989. Flood tolerance and tree distribution in central Amazonian
floodplains. In: Haynes, R.R., Holm-Nielsen, L.B., Balslev, H. (Eds.).
Tropical Forests. Botanical Dynamics, Speciation and Diversity.
Academic Press, London. pp. 47-64.
Junk, W.J., Bayley, P.B., Sparks, R.E., 1989. The flood pulse concept in river-
floodplain systems. In: Dodge, D.P. (Ed.). Proceedings of the
International Large River Symposium, Canadian Special Publicationof
Fisheries and Aquatic Sciences, Ottawa, Canada. pp. 110–127.
Junk W.J., 1993. Wetlands of tropical South America. In: Whigham, D.F.,
Dyrygova, D., Hejny, S. (Eds.). Wetlands of the World, Inventory and
Management. Kluwer Academic Publisher, Dordrecht. pp.679-740.
147
147
Junk, W. J., 2002. Long-term environmental trends and the future of tropical
wetlands. Env. Cons. 29: 414–435.
Junk, W. J. and M. T. F. Piedade 2004. Status of knowledge, ongoing research,
and research needs in Amazonian wetlands. Wetlands Ecology and
Management 12(6): 597-609.
Junk, W. J., and C. N. d. Cunha. 2005. Pantanal: a large South American
wetland at a crossroads. Ecological Engineering 24: 391-401.
Junk, W.J., Nunes da Cunha, C., Wantzen, K.M., Petermann, P., Strüssmann,
C., Marques, M.I., Adis, J., 2006a. Biodiversity and its conservation in the
Pantanal of Mato Grosso, Brazil. Aquat. Sci. 68, 278-309.
Junk, W. J., M. Brown, I. C. Campbell, M. Finlayson, B. Gopal, L. Ramberg, and
B. G. Warner. 2006b. The comparative biodiversity of seven globally
important wetlands: a synthesis. Aquatic Sciences 68:400–414.
Kandus, P., and A. I. s. Malvarez. 2004. Vegetation patterns and change
analysis in the lower delta islands of the Parana river (Argentina).
Wetlands 24:620–632.
Karssenberg, D. 2002. Building dynamic spatial environmental models (Het
maken van dynamische ruimtelijke landschapsmodellen). PhD thesis,
Universiteit Utrecht, Utrecht.
Karssenberg, D., De Jong, K. and Van der Kwast, J. 2007. Modelling landscape
dynamics with Python. Int. J. Geogr. Inf. Sci., 21: 483–495.
148
148
Keddy, P.A., L. H. Fraser, A. I. Solomeshch, W. J. Junk, D. R. Campbell, M. T.
K. Arroyo, and C. J. R. Alho. 2009 Wet and Wonderful: The World’s
Largest Wetlands Are Conservation Priorities. BioScience 59:39-51.
Keogh, T.M., Keddy, P.A. e Fraser, L.H. 1999. Patterns of Tree Species
Richness in Forested Wetlands Wetlands, 19(3): 639-647.
Kirkman, L. K., P. C. Goebel, L. West, M. B. Drew, and B. J. Palik. 2000.
Depressional wetland vegetation types: a question of plant community
development. Wetlands 20: 373–385.
Köppen, W. 1948. Climatologia. Buenos Aires, Fondo de Cultura, pp.152-192.
Legendre, P., Legendre, L., 1998. Numerical Ecology. Amsterdam, NL.
Leps, J., 1990. Can underlying mechanisms be deduced from observed
patterns? In: Krahulec F. Agnew A.D.Q. Agnew S. & Willems J.H. (eds.):
Spatial processes in plant communities. SPB Publ.. pp. 1-11.
Levin, S. A. 1976. Population Dynamic Models in Heterogeneous
Environments. Annual Review of Ecology and Systematics 7:287-310.
Loehle, C. and D. LeBlanc 1996. Model-based assessments of climate change
effects on forests: a critical review. Ecological Modelling 90(1): 1-31.
Logofet, D. O. and E. V. Lesnaya 2000. The mathematics of Markov models:
what Markov chains can really predict in forest successions. Ecological
Modelling 126(2-3): 285-298.
Luo, W., F. Song, and Y. Xie. 2008. Trade-off between tolerance to drought and
tolerance to flooding in three wetland plants. Wetlands 28:866–873.
149
149
Magnusson, W.E., Lim, A.P, Luizão, R., Luizão, F., Costa, F.R.C., Castilho,
C.V., Kinupp, V.F., 2005. RAPELD: uma modificação do método de
Gentry para inventário da biodiversidade em sítios para pesquisa
ecológica de longa duração. Biota Neotropica, 5.
http://www.biotaneotropica.org.br/v5n2
Marengo, J. A. 2008 Água e mudanças climáticas. Estudos Avançados 22: 83-
96.
Martinez, J.M., Toan, T.L., 2007. Mapping of flood dynamics and spatial
distribution of vegetation in the Amazon floodplain using multitemporal
SAR data. Remote Sens. Environ. 108, 209–223.
Mauro, R. A., Pott, A., Silva, M.P. 1998. Una propuesta de modelo de estados y
transiciones para una sabana tropical inundable: el Pantanal arenoso.
Ecotropicos, 10(2): 99-112.
McIntosh, R.P., 1958. Plant Communities. Science 128, 115-120.
Metzger, J. P. 2004. Delineamento de Experimentos numa Perspectiva de
Ecologia da Paisagem. In: Culler Jr L., Valladares-Pádua C. (eds.).
Métodos de Estudos em biologia da Conservação e Manejo da Vida
Silvestre. Ed. da UFPR. Fundação O Boticário de Proteção à Natureza.
pp. 539-553
Miller, J., Franklin, J., 2006. Explicitly incorporating spatial dependence in
predictive vegetation models in the form of explanatory variables: a
Mojave Desert case study. J. Geogr. Syst. 8, 411-435.
150
150
Miller, J., Franklin, J., Aspinall, R., 2007. Incorporating spatial dependence in
predictive vegetation models. Ecol. Model. 202, 225-242.
Milly, P. C. D., K. A. Dunne, and A. V. Vecchia. 2005. Global pattern of trends in
streamflow and water availability in a changing climate. 438:347-350.
Minchin, P.R., 1989. Montane vegetation of the Mt. Field Massif, Tasmania: a
test of some hypotheses about properties of community patterns. Plant
Ecol. 83, 97-110.
Mitsch, W. J., and J. G. Gosselink. 2000. Wetlands. Van Nostrand Reinhold,
New York. 3 ed.
Moore, A. D. 1990. The Semi-Markov Process: a Useful Tool in the Analysis of
Vegetation Dynamics for Management. Journal of Environmental
Management 30:111-130.
Mouillot, F., S. Rambal, and R. Joffre. 2002. Simulating climate change impacts
on fire frequency and vegetation dynamics in a Mediterranean-type
ecosystem. Global Change Biology 8: 423-437.
Mueller-Dombois, D., Ellemberg, H., 1974. Aims and Methods of Vegetation
Ecology. John Wiley & Sons, New York.
Nunes da Cunha, C., Junk, W.J., 1999. Composição florística de capões e
cordilheiras: localização das espécies lenhosas quanto ao gradiente de
inundação no Pantanal de Poconé, MT, Brasil. In: Anais do II Simpósio
sobre Recursos Naturais e Sócio-Econômicos do Pantanal - Manejo e
Conservação. Embrapa – CPAP, Corumbá, Brasil, pp. 387-406.
Nunes da Cunha, C., Junk, W.J., 2000. The impact of flooding on distribution of
woody plant communities in the Pantanal of Poconé, Mato Grosso,
151
151
Brazil. In: German-Brazilian Workshop on Neotropical Ecosystems -
Achievements and Prospects of Cooperative Research. Session 4: Living
Resource Management: Approaches, Techniques, Variability. Hamburg,
Germany. pp. 3-8.
Nunes da Cunha, C., and W. J. Junk. 2004. Year-to-year changes in water level
drive the invasion of Vochysia divergens in Pantanal grasslands. Applied
Vegetation Science 7:103-110.
Nunes da Cunha, C., and H. F. Leitão-Filho. 2007. Woody vegetation in the
Pantanal of Mato Grosso, Brazil: apreliminary typology. Amazoniana XIX
(3/4): 159-184.
Nunes da Cunha, C., Rawiel, P., Wantzen, K.M., Junk, W.J., Prado, A.L., 2006.
Mapping and characterization of vegetation units by means of Landsat
imagery and management recommendations for the Pantanal of Mato
Grosso (Brazil), north of Poconé. Amazoniana. XIX, 1-32.
O'Neill, R.V., Johnson, A.R., King, A.W., 1989. A hierarchical framework for the
analysis of scale. Landscape Ecol. 3, 193-205.
Ozesmi, S.L., Bauer, M.E., 2002. Satellite remote sensing of wetlands. Wetl.
Ecol. Manag. 10, 381–402.
Pardo-Iguzquiza, E., Chica-Olmo, M., 2008. Geostatistics with the Matérn
semivariogram model: a library of computer programs for inference,
kriging and simulation. Comput. Geosci. 34, 1073-1079.
152
152
Parolin P., Oliveira, A.C., Piedade, M.T.F, Wittmann, F., Junk, W.J., 2002.
Pioneer trees in Amazonian floodplains: three key species form
monospecific stand in different habitats. Folia Geobotanica 37: 225-238.
Parolin, P., 2009. Submerged in darkness: adaptations to prolonged
submergence by woody species of the Amazonian floodplains. Ann. Bot.
103, 359-376.
PCRaster, 2002. PCRaster Software. Info at http://www.geog.uu.nl/pcraster.nl.
Pebesma, E.J., Wesseling, C.G., 1998. GSTAT: a program for geostatistical
modelling, prediction and simulation. Computat. Geosci. 24, 17–31.
Pebesma, E. J., 2004. Multivariable geostatistics in S: the gstat package.
Comput. Geosci. 30, 683-691.
Pfeffer, K., Pebesma, E.J., Burrough, P.A., 2003. Mapping alpine vegetation
using vegetation observations and topographic attributes. Landscape
Ecol. 18, 759-776.
Pillar, V.D., Sosinski Jr., E.E., 2003. An improved method for searching plant
functional types by numerical analysis. J. Veg. Sci. 14, 323-332.
Ponce, V. M. and C. Nunes da Cunha 1993. Vegetated Earthmounds in
Tropical Savannas of Central Brazil: A Synthesis: With Special
Reference to the Pantanal do Mato Grosso. Journal of Biogeography
20(2): 219-225.
Pott, A., Pott, V.J. 2004. Features and conservation of the Brazilian Pantanal
wetland. Wetlands Ecology and Management 12: 547–552.
153
153
Pott, A. 2007. Dinâmica da vegetação do Pantanal. In: SEB (Ed.) Anais do VIII
Congresso de Ecologia do Brasil, Caxambu - MG.
Prance, G.T., Schaller, G.B., 1982. Preliminary study of some vegetation types
of the Pantanal, Mato Grosso, Brazil. Brittonia 34, 228-251.
R Development Core Team, 2009. R: A Language and Environment for
Statistical Computing. R Foundation for Statistical Computing, Vienna,
Austria, ISBN 3-900051-07-0, URL http://www.R-project.org.
Ratter, J.A., Pott, A., Pott, V.J., Nunes da Cunha, C., M. Haridasam 1988.
Observations on woody vegetation types in the Pantanal and at
Corumbá, Brasil. - Notes RBG, Edinb. 45(3): 503-525.
Reynolds, J. F., H. Bugmann, and L. F. Pitelka. 2001. How much physiology is
needed in forest gap models for simulating long-term vegetation
response to global change? challenges, limitations, and potentials.
Climatic Change 51:541–557.
Scheffer, M. 2009. Critical transitions in nature and society. Princeton University
Press, Princeton and Oxford.
Scheller, R., Mladenoff, D., 2007. An ecological classification of forest
landscape simulation models: tools and strategies for understanding
broad-scale forested ecosystems. Landscape Ecology 22 (4): 491-505.
Schöngart, J., Arieira, J., Fortes, C. F., Arruda, E.C., Nunes da Cunha, C.,
2008. Carbon dynamics in aboveground coarse wood biomass of
154
154
wetland forest in the northern Pantanal, Brazil. Biogeosciences Discus. 5,
1203-1230.
Shmida, A., and S. Ellner. 1984. Coexistence of plant species with similar
niches. Vegetatio 58: 29-55.
Silva, M.P.; Mauro, R.A.; Pott, A.; Boock, A.; Pott, V.J.; Ribeiro, M. , 1988. Una
sabana tropical inundable: El Pantanal arcilloso, propuesta de modelo de
estados y transiciones. Ecotropicos, 10(2): 87-98.
Silva, M.P., Mauro, R., Mourão, G.E, Coutinho, M., 2000. Distribuição e
quantificação de classes de vegetação do Pantanal através de
levantamento aéreo. Revista Brasileira de Botânica, 23 (2): 143-152.
Stein, A., Corsten, L.C.A., 1991. Universal kriging and cokriging as a regression
procedure. Biometrics 47, 575-587.
Stroh, C. L., D. D. Steven, and G. R. Guntenspergen. 2008. Effect of climate
fluctuations on long-term vegetation dynamics in Carolina Bay wetlands.
Wetlands 28:17–27.
Svenning, J.C., Kinner, D.A., Stallard, R.F., Engelbrecht, B.M.J., Wright, S.J.,
2004. Ecological determinism in plant community structure across a
tropical forest landscape. Ecology 85, 2526-2538.
Tilman, D., 1988. Plant Strategies and the Dynamic and Structure of Plant
Communities. Princeton University Press, Princeton, New Jersey.
Tilman, D. 1994. Competition and biodiversity in spatially structured habitats.
Ecology. 75(1): 2-16.
155
155
Tilman, D., P. Kareiva 1997. Spatial ecology - the role of space in population
dynamics and interspecific interaction, Princeton, New Jersey. pp. 365
Troxler, G.T., Childers, D.L., 2006. Relationships between hydrology and soils
describe vegetation patterns in seasonally flooded tree islands of
southern Everglades, Florida. Plant Soil 279, 271-286.
Tucker, C.J., 1979. Red and photographic infrared linear combinations for
monitoring vegetation. Remote Sens. Environ. 8, 127-150.
Turner, M.G., 1987. Spatial simulation of landscape changes in Georgia: a
comparison of 3 transitional models. Landscape Ecol. 1, 29-36.
Turner, M.G., 1989. Landscape ecology: the effect of pattern on process. Annu.
Rev. Ecol. Syst. 20, 171-197.
Turner, M.G., Gardner, R.H., O'Neill, R.V., 2001. Landscape Ecology in Theory
and Practice- Pattern and Process. Springer-Verlag New York, Inc.,
USA.
Usher, M. B. 1981. Modelling ecological succession, with particular reference to
Markovian models. Plant Ecology 46-47(1): 11-18.
Van der Valk, A. G. 1981. Succession in Wetlands: A Gleasonian Approach.
Ecology 62:688-696.
Van Nes, E.H., Scheffer, M. 2005. A strategy to improve the contribution of
complex simulation models to ecological theory. Ecological Modelling.
185: 153–164.
Velloso, H.P., Rangel Filho, A.L.R., Lima, J.C.A., 1991. Classificação da
Vegetação Brasileira Adaptada a um Sistema Universal. IBGE, RJ, BR.
156
156
Vogiatzakis, I.N., Griffiths, G.H., 2006. A GIS-based empirical model for
vegetation prediction in Lefka Ori, Crete. Plant Ecol. 184, 311 –323.
Wagner, H.H., Fortin, M.J., 2005. Spatial analysis of landscapes: concepts and
statistics. Ecology 86, 1975-1987.
Walther, G.-R., E. Post, P. Convey, A. Menzel, C. Parmesank, T. J. C. Beebee,
J.-M. Fromentin, O. Hoegh-GuldbergI, and F. Bairlein. 2002. Ecological
responses to recent climate change. Nature 416:389-395.
Wantzen, K.M., Drago, E., da Silva, C.J., 2005. Aquatic habitats of the Upper
Paraguay River-Floodplain-System and parts of the Pantanal (Brazil). Int.
J. Ecohydrol. Hydrobiol. 5 (2), 107-126.
Ward, J. V. and K. Tockner 2001. Biodiversity: towards a unifying theme for
river ecology. Freshwater Biology 46: 807-819.
Wassen, M.J., Peeters, W.H.M., Venteunt, H.O., 2002. Patterns in vegetation,
hydrology and nutrient availability in an undesturbed river floodplain
Poland. Plant Ecol. 165, 27-43.
Weaver, J.E., Bruner, W.E., 1945. A Seven-Year Quantitative Study of
Succession in Grassland. Ecol. Monogr. 15, 297-319.
Weaver, K. and A. H. Perera 2004. Modelling land cover transitions: A solution
to the problem of spatial dependence in data. Landscape Ecology 19(3):
273-289.
157
157
Werffeli, B., Roulier, C., Buttler, A., 1997. The methodology of integrated
synusial phytosociology applied to a floodplain sector of the Sarine river,
Switzerland. Global Ecol. Biogeogr. 6, 237-245.
Wesseling, C.G., Deursen, W.P.A., Burrough, P.A., 1996. A spatial modelling
language that unifies dynamic environmental models and GIS. In:
Proceeding, Third International Conference/ Workshop of Integrating GIS
and Environmental Modeling, Santa Fé, NM.
http://www.ncgia.ucsb.edu/conf/SANTA FE CD-ROM/main.html, Santa
Barbara, CA. National Center for Geographic Information and Analysis.
Whittaker 1967,
Whittaker, R.W., 1967. Gradient analysis of vegetation. Biol Rev. 49, 207-264.
Wiens, J.A., 1989. Spatial scaling in ecology. Funct. Ecol. 3, 385-397.
Wolfram, S. (1984). Cellular automata as models of complexity. Nature
311(5985): 419-424.
Yemshanov, D. and A. H. Perera 2002. A spatially explicit stochastic model to
simulate boreal forest cover transitions: general structure and properties.
Ecological Modelling 150(1-2): 189-209.
Zeilhofer, P., Schessl, M., 1999. Relationship between vegetation and
environmental conditions in the northern Pantanal of Mato Grosso. J.
Biogeogr. 27, 159-168.
Zeilhofer, P., 2006. Soil mapping in the Pantanal of Mato Grosso, Brazil, using
multitemporal Landsat TM data. Wetl. Ecol. Manag. 14, 445-461.
158
158
Curriculum Vitae
Julia Arieira nasceu em 03 de julho de 1976 no Rio de Janeiro, Brasil.
Graduou-se na Universidade Santa Úrsula, Rio de Janeiro, em 2001, no curso
de Bacharelado e Licenciatura em Ciências Biológicas. Nos dois anos
seguintes a sua formação, trabalhou em projetos de Educação ambiental,
prestou serviços como consultora em projetos de desenvolvimento sustentável
e avaliação de impactos de empreendimentos energéticos e como professora
de ensino fundamental e médio. Em 2003 começou seu mestrado em Ecologia
e Conservação da Biodiversidade na Universidade Federal de Mato Grosso,
sob orientação da Doutora Cátia Nunes da Cunha. Ingressou em 2005, logo
após o termino do mestrado, no doutorado em Agricultura Tropical pela UFMT,
sob orientação do Doutor Eduardo Guimarães Couto e co-orientação da
doutora Cátia Nunes da Cunha e Doutor Derek Karssenberg. Em seu
doutorado, participou do programa oferecido pela CAPES de doutorado
sanduíche, no qual permaneceu um ano no departamento de Geografia Física,
Universidade de Utrecht, Países Baixos, desenvolvendo novas habilidades em
análise espacial de dados biológicos, sob orientação do Doutor Derek
Karssenberg e colaboração do Doutor Steven de Jon e Doutora Elisabeth
Addink. Durante seu desenvolvimento acadêmico, publicou dois artigos
científicos, teve publicações técnicas junto à EMBRAPA e textos em jornais de
divulgação científica. Hoje, Julia participa de projetos de pesquisa vinculados
ao Programa de Pesquisas Ecológicas de Longa Duração e ao Instituto
Nacional de Áreas Úmidas.