bs bs banner research global vegetation biomass change...

RESEARCHPAPER

Global vegetation biomass change(1988–2008) and attribution toenvironmental and human driversYi Y. Liu1,2,3,4*, Albert I. J. M. van Dijk4,5, Matthew F. McCabe1, Jason P. Evans2

and Richard A. M. de Jeu3

1Water Research Centre, School of Civil and

Environmental Engineering, University of New

South Wales, Sydney, NSW 2052, Australia,2Climate Change Research Centre, University

of New South Wales, Sydney, NSW 2052,

Australia, 3Department of Hydrology and

Geo-Environmental Sciences, Faculty of Earth

and Life Sciences, Vrije Universiteit

Amsterdam, Amsterdam, 1181 HV, The

Netherlands, 4CSIRO Land and Water, Black

Mountain Laboratories, Canberra, ACT 2601,

Australia, 5Fenner School of Environment and

Society, The Australian National University,

Canberra, ACT 0200, Australia

ABSTRACT

Aim Vegetation optical depth (VOD) is an indicator of the water content of bothwoody and leaf components in terrestrial aboveground vegetation biomass that canbe derived from passive microwave remote sensing. VOD is distinct from opticalvegetation remote sensing data such as the normalized difference vegetation indexin that it is: (a) less prone to saturation in dense canopies; (b) sensitive to bothphotosynthetic and non-photosynthetic biomass; and (c) less affected by atmos-pheric conditions. Our primary objective was to analyse a recently developed long-term VOD record and investigate how the vegetation water content of variousland-cover types responded to environmental changes and human influences from1988 to 2008.

Location Global.

Methods We first conducted Mann–Kendall trend tests on annual average VODto identify regions with significant changes over the period 1988–2008. To diagnosethe underlying cause of the observed changes, patterns for these identified regionswere further compared with independent datasets of precipitation, crop produc-tion, deforestation and fire occurrence.

Results (1) Over grassland and shrubland, VOD patterns corresponded stronglyto temporal precipitation patterns. (2) Over croplands, annual average VODshowed a general increase that corresponded to reported crop production patternsand was attributed to a combination of precipitation patterns and agriculturalimprovements. (3) Over humid tropical forest, the spatial pattern of VOD declineagrees well with deforestation patterns; the 2005 Amazon drought correspondedwith a temporary VOD decrease. (4) Over boreal forests, regional VOD declines areattributed to a combination of fires and clear cutting.

Main conclusions Passive microwave remote sensing of VOD can be used tomonitor global changes in total aboveground vegetation water content and biomassover various land-cover types. This new observational record can help in hydro-logical, agricultural, ecological and climate change studies, and provides newinsights into large-scale vegetation change and its drivers.

KeywordsCrop production, deforestation, fire, long-term change, passive microwave,precipitation, vegetation optical depth.

*Correspondence: Yi Liu, Water ResearchCentre, School of Civil and EnvironmentalEngineering, University of New South Wales,Sydney, NSW 2052, Australia.E-mail: [email protected]

INTRODUCTION

Vegetation dynamics play an important role in terrestrial

ecology, atmospheric processes and the water and carbon cycles

(Cox et al., 2000; van Dijk & Keenan, 2007; Alessandri &

Navarra, 2008). At continental to global scales, satellite observa-

tion offers the only feasible approach for monitoring vegetation

dynamics. For long-term studies, the most commonly used

bs_bs_banner

Global Ecology and Biogeography, (Global Ecol. Biogeogr.) (2012) ••, ••–••

© 2012 Blackwell Publishing Ltd DOI: 10.1111/geb.12024http://wileyonlinelibrary.com/journal/geb 1

global observations of vegetation dynamics are based on the

normalized difference vegetation index (NDVI) derived from

the Advanced Very High Resolution Radiometer (AVHRR) (e.g.

Tucker et al., 2005), which extends back to the early 1980s.

NDVI, computed as a normalized ratio of reflectance in the

near-infrared and red portions of the electromagnetic spectrum,

provides a measure of chlorophyll abundance at the earth’s

surface, and as such offers an indirect measure of energy absorp-

tion and vegetation density (Myneni et al., 1995). There are

some well-established limitations to NDVI and other measures

derived from optical observations: NDVI only senses the top of

the vegetation canopy; it is affected by atmospheric influences

such as clouds and aerosols (Kobayashi & Dye, 2005); and the

relationship between NDVI and leaf biomass breaks down over

dense vegetation (Turner et al., 1999).

Vegetation properties can also be retrieved from satellite

passive microwave observations that are sensitive to water

content at the surface. Owe et al. (2001) developed a method to

retrieve vegetation optical depth (VOD) from passive micro-

wave emissions that can be used across all low-frequency micro-

wave bands (< 20 GHz). In this approach, VOD is interpreted as

being directly proportional to total vegetation water content in

all the aboveground biomass, including both woody and leafy

components. VOD has characteristics that are quite distinct

from NDVI. It is sensitive to water rather than chlorophyll and

hence to both photosynthetic and non-photosynthetic above-

ground biomass (Shi et al., 2008; Jones et al., 2011). Also,

because of the long microwave wavelengths, it penetrates deeper

into the canopy (and therefore saturates at a higher biomass

level than NDVI), and is much less sensitive to atmosphere and

weather conditions. The low energy of microwave emissions

does, however, require a larger footprint of several kilometres in

diameter, resulting in a relatively coarse spatial resolution com-

pared with optical and thermal techniques.

While NDVI is sensitive to chlorophyll abundance, VOD is an

indicator of biomass water content. A change in canopy green-

ness does not necessarily mean that the vegetation water content

changes in the same direction, and vice versa (Ceccato et al.,

2002). Therefore considering both NDVI and VOD records

together can provide a more comprehensive characterization of

vegetation dynamics. There have been several studies using

NDVI to analyse temporal and spatial variations in vegetation

distribution and monitor land degradation, landscape fragmen-

tation and the ecological effects of anthropogenic activities and

weather events (e.g. Bai et al., 2008; Seaquist et al., 2009; Milesi

et al., 2010). By contrast, there are few studies based on micro-

wave vegetation remote sensing (Shi et al., 2008; Jones et al.,

2011). One likely reason is that no single continuous satellite-

based microwave instrument covers a period comparable to the

AVHRR. However, there have been a series of complementary

passive microwave sensors extending back over two decades, and

Liu et al. (2011a) demonstrated that data from three successive

passive microwave satellite instruments could be merged to

create a historical VOD record from July 1987 onwards – the first

long-term observational record of total aboveground vegetation

water content, which is closely related to vegetation biomass.

This record provides new opportunities to study vegetation

dynamics at regional to global scale.

In addition to developing the VOD merging method, Liu et al.

(2011a) briefly compared VOD and NDVI patterns, but did not

attempt to analyse global vegetation dynamics and trends in

detail. Here, we aim to quantify spatiotemporal changes in VOD

from 1988 to 2008 at a global scale and identify the drivers

behind these changes. The underlying objective was to better

understand how VOD patterns have responded to environmen-

tal changes and human activity over the last two decades. A

subsidiary goal was to test the usefulness of the newly developed

VOD record for the study of macroecological patterns and

related applications to ecological, hydrological and agricultural

questions.

DATASETS AND METHODS

VOD data used

The VOD retrievals used in this study were derived from the

SSM/I (Special Sensor Microwave Imager, 1988–2007), TMI (the

microwave instrument onboard the Tropical Rainfall Measuring

Mission satellite, 1998–2008) and AMSR-E (the Advanced

Microwave Scanning Radiometer – Earth Observing System,

July 2002–08) sensors using the land parameter retrieval model

(LPRM), developed by the VU University Amsterdam, the Neth-

erlands in collaboration with National Aeronautics and Space

Administration (Owe et al., 2001). Full details on the merging

processes can be found in Liu et al. (2011a, b) and are only

briefly summarized here.

The VOD data were retrieved based on: (1) the vertically

polarized signal of 36.5/37 GHz (a common band of all three

instruments) and (2) the vertically and horizontally polarized

signals of the lowest microwave frequency of each instrument.

Due to differing specifications in wavelength, viewing angle and

footprint spatial resolution between the instruments, the abso-

lute values of VOD retrievals vary. However, the relative dynam-

ics are highly correlated, making it possible to adjust the data to

a common reference and merge these into a continuous long-

term time series. The AMSR-E VOD retrieval was selected as

reference, as it has a relatively low measuring microwave fre-

quency that increases VOD retrieval accuracy. The cumulative

distribution frequency (CDF) matching technique was applied

to rescale SSM/I and TMI against AMSR-E, after which data

were merged into a single continuous time series. In the merged

data set, AMSR-E retrievals are used from July 2002 to Decem-

ber 2008 at the global scale. TMI retrieval covers the period from

January 1998 to June 2002 between 38° N and 38° S. Elsewhere,

SSM/I retrievals are used. It is noted that VOD accuracy within

this long-term series is likely to have increased over time as the

design of successive instruments improved.

Data processing

The merged daily VOD data were resampled to 0.25° (roughly

25 km) resolution and aggregated into monthly and then annual

Y. Y. Liu et al.

Global Ecology and Biogeography, ••, ••–••, © 2012 Blackwell Publishing Ltd2

averages (Fig. 1a). VOD retrievals are not available when the

surface temperature is below 0 °C or where snow cover exists;

annual VOD values over high-latitude regions are the averages of

warm months only. The presence of open water affects the

microwave emissions and may lead to underestimates of VOD

values (Jones et al., 2011). Because of this, regions with extensive

lakes, reservoirs, rivers and flooded vegetation (cf. Lehner &

Doll, 2004) were masked out when the long-term trends were

examined (Fig. 1a). A 25-km buffering zone outside the mapped

open water boundaries was masked out to account for the pos-

sible influence beyond the nominal satellite footprint.

The University of Maryland (UMD) 14-class global 0.05°-

resolution land-cover product (MCD12C1; Friedl et al., 2002)

was aggregated to nine classes and 0.25° resolution by dominant

class (Fig. 1b). The distribution of VOD for the nine land-cover

classes (see Appendix S1 in Supporting Information) shows the

highest values for forests and the lowest values for barren

regions, as expected.

Several additional data sources were used to help attribute

global trends and patterns in VOD:

Precipitation

Where water limits growth, temporal patterns in vegetation are

likely to be influenced strongly by antecedent precipitation.

Three 0.5° gridded monthly precipitation data sources were

Figure 1 (a) Annual average vegetation optical depth (VOD) for 1988–2008. Regions probably affected by open water are masked in grey.(b) Biome classes based on Friedl et al. (2002) but aggregated to nine classes (‘evergreen and deciduous forest’ includes both needleleaf andbroadleaf forest; ‘shrublands’ both open and closed shrublands; ‘savannas’ both woody and non-woody savannas; and ‘barren’ includesurban, built-up and sparsely vegetated classes in the original classification). The Behrmann cylindrical equal-area projection is applied forboth panels.

Global vegetation biomass change

Global Ecology and Biogeography, ••, ••–••, © 2012 Blackwell Publishing Ltd 3

used, developed by the Global Precipitation Climatology Centre

(GPCC; Rudolf & Schneider, 2005; http://gpcc.dwd.de), the

University of Delaware (UDel; http://climate.geog.udel.edu/

~climate/) and the University of East Anglia Climatic Research

Unit (CRU; Mitchell & Jones, 2005; http://www.cru.uea.ac.uk/

cru/data/hrg/). The simple average of the three data sources was

used in the analyses.

Agriculture

Annual crop production data for various agricultural sectors

and individual countries were obtained from the Food and Agri-

culture Organization of the United Nations (FAO; http://

faostat.fao.org/site/567/default.aspx#ancor). These data were

compared with long-term VOD patterns for countries with

extensive cropland areas (Fig. 1b).

Tropical deforestation

Two datasets were used in this study to spatially map changes in

tropical forest cover between 1988 and 2008. A dataset delineat-

ing deforestation ‘hot spots’ in the humid tropics was derived

based on assessments from around 30 regional experts (Achard

et al., 1998; http://bioval.jrc.ec.europa.eu/products/deforest_

hotspot/deforesthotspotHT.php). More recently, Hansen et al.

(2008) developed a probability based sampling approach,

employing low- and high-resolution satellite datasets (MODIS

and Landsat, respectively) to quantify humid tropical forest

clearing between 2000 and 2005 (http://globalmonitoring.

sdstate.edu/projects/gfm/). Thirdly, to allow for a degree of tem-

poral interpretation, reported annual deforestation rates for all

states in Brazil from 1988 to 2008 were included (http://

www.obt.inpe.br/prodes/prodes_1988_2010.htm).

Fire

Two data sources were used to interpret VOD changes related to

fire. The MODIS-based global monthly fire location product

(MCD14ML; ftp://fuoco.geog.umd.au) contains the geographic

location (latitude, longitude), date and detection confidence

(0–100%) of fires identified from MODIS thermal band anoma-

lies. Fires with 100% detection confidence from 2001–08 were

used here, and accumulated to the number of fires per year for

each 0.25° grid cell. All fires detected within the same 0.25° grid

cell on the same day of year were summed. In addition, time

series of the annual number of fires in Alaska from 1988 to 2007

were available from the Bureau of Land Management, Alaska

Fire Service (http://agdc.usgs.gov/data/blm/fire/index.html).

Statistical methods

The nonparametric Mann–Kendall trend test is commonly used

to detect monotonic trends in environmental time series. It is

based on the relative rankings of the sample data rather than

absolute values, and does not require any assumptions about the

distribution of the data or the form of the monotonic trend

(Helsel & Hirsch, 2002). We applied the Mann–Kendall test to

annual average VOD for the period 1988–2008 at each grid cell

to identify areas with widespread and significant change

(P < 0.05) and selected these for further comparison against the

independent datasets described above.

The empirical orthogonal function (EOF) and singular value

decomposition (SVD) analysis techniques (Bretherton et al.,

1992; von Storch & Zwiers, 1999) were also applied. Both

methods produce a set of functions that represent the dominant

modes of variation in spatiotemporal data sets, and the relative

importance of each pattern in explaining observed variation

across space. EOF analysis was used to extract dominant modes

in a single variable, whereas SVD analysis was used to identify

the common modes of behaviour that relate two variables. Non-

parametric Spearman’s rank order correlation coefficients were

calculated to quantify the relationship between candidate

drivers and VOD or its dominant mode (von Storch & Zwiers,

1999).

RESULTS

Several regions showed significant change (P < 0.05) in annual

average VOD from 1988 to 2008 (Fig. 2). Decreases are seen over

the boreal forests in Alaska, west Russia and east Russia and over

the tropical forests of South America and Southeast Asia.

Declines are also seen over non-forested areas such as the south

central USA, south-east Australia and Mongolia. Increasing

changes are generally detected over large cropland regions in the

Canadian prairies, the USA, eastern Europe, southern Russia,

India and China. Several regions with grassland, shrubland and

savannas experienced increases in VOD, including Texas, south-

ern Argentina, southern Africa, the Sahel and north-west Aus-

tralia. Through comparison with the candidate driver (see

Datasets and Methods), the most likely causes for VOD trends in

the 19 outlined regions (Fig. 2) could generally be assigned to

three categories: precipitation, agriculture and deforestation.

These are discussed in more detail in the next sections.

Changes in water limited regions

Changes in VOD for nine out of the 19 regions (Fig. 2) seem to

be primarily attributable to variations in precipitation (see

Appendix S2). Over the five larger regions (i.e. Mongolia, north-

west and south-east Australia, southern Africa and the Sahel),

the SVD analysis was applied to investigate the covariance

between annual precipitation and VOD for all grid cells with

significant VOD changes. The first SVD mode explained

44–73% of the covariance between annual precipitation and

VOD for the different regions, while the correlation coefficients

(R) between the first SVD temporal mode of precipitation and

VOD ranged from 0.76 to 0.90 (Fig. 3). The spatial variability is

limited for the four smaller regions (i.e. south central USA,

Texas, southern Argentina and eastern Europe), which would

reduce the power of SVD analysis. Therefore the annual precipi-

tation and VOD between 1988 and 2008 for all grid cells with

Y. Y. Liu et al.


significant VOD changes are directly compared (Fig. 4); the cor-

relation between annual precipitation and VOD for these data

was 0.70–0.81.

Interannual variations and long-term changes in VOD from

1988 to 2008 varied between the regions: two regions experi-

enced strong declines (Mongolia and south-east Australia); and

five regions experienced an increase (Texas, the Sahel, eastern

Europe, southern Africa and southern Argentina). Over the

south central USA, VOD decreased from 1988 to 2002 and

recovered thereafter, while over north-west Australia VOD

increased before 2000 and then declined. Despite the different

patterns among regions, interannual variations as well as the

long-term changes in VOD correspond to precipitation changes

between 1988 and 2008.

Changes in agricultural regions

Extensive agricultural areas exist in China, India, Russia, the

USA and Canada (Fig. 1b) and significant increases in VOD

(1988–2008) are generally observed over these regions (Fig. 2).

To further examine agricultural responses and to extract the

primary spatial and temporal VOD patterns, EOF analysis was

performed on annual VOD values over the grid cells with sig-

nificant changes falling in the cropland land cover (displayed in

Fig. 1b). The first EOF modes accounted for 42–88% of annual

VOD variance over these large-scale agricultural regions from

1988 to 2008 (Fig. 5). Time series of the first EOF modes were

compared with the annual production (i.e. crop yield times

harvest area) of the dominant crop for each region in terms of

harvested area based on FAO data, i.e. maize in China and the

USA, rice in India, wheat in Russia and wheat/rapeseed in

Canada (in some areas multiple crops are grown but it is prob-

lematic to sum production from crops with different

production-to-biomass ratios). The annual crop production

appears to explain the majority of VOD variations (Fig. 5, right

column). The drivers of both VOD and crop production include

both weather conditions and agricultural practices. The corre-

lation between VOD and precipitation was between 0.35 and

0.45 for the five regions, suggesting that precipitation explains

up to c. 20% of VOD variations in these regions, with the

remainder attributable to other factors. Overall, VOD shows the

same increasing trends as crop production.

Changes in tropical forests

The spatial distribution of significant VOD decreases (P < 0.05)

over tropical forests in the American Tropics and Southeast Asia

are shown in Fig. 6(a) and (b), respectively. In the Americas the

greatest declines are observed over the south-eastern Amazon

Basin (particularly the Brazilian states of Mato Grosso, Pará and

Rondônia). Bolivia, Paraguay, Guatemala, Guyana, Suriname,

French Guiana and coastal regions of Ecuador and Colombia

also experienced significant VOD decreases. In Southeast Asia

the most widespread declines are observed over Sumatra and the

Kalimantan islands in Indonesia. Malaysia, Vietnam, Cambodia,

Laos, Thailand and Myanmar also witness significant decreases

Figure 2 Areas with significant changes (P < 0.05) in annual average vegetation optical depth (VOD) for 1988–2008 (unit: VODchange/year). Nineteen regions selected for further analysis are outlined.



in VOD over tropical forests in their territories. Despite subtle

differences, the spatial patterns compare very well with the

tropical deforestation areas mapped by Achard et al. (1998;

Fig. 6c & d) and Hansen et al. (2008; Fig. 6e & f). The agreement

suggests that the significant declines in VOD over these tropical

regions are mostly associated with large-scale deforestation.

Spatiotemporal VOD dynamics over the Brazilian states with

the most severe VOD declines (i.e. Mato Grosso, Pará and

Rondônia) were examined further by applying EOF analysis for

the grid cells with significant changes within the state bounda-

ries. The first EOF mode explained 75% of the VOD variance

over these states and shows continuous VOD declines from 1988

to 2008 due to deforestation (see Appendix S3). The detrended

time series of EOF mode 1 (i.e. after removing the linear trend

from the time series of EOF mode 1) was calculated and com-

pared with interannual variations in deforested area and rainfall

Figure 3 Spatial patterns (left and right column) and time series (middle column) of the first singular value decomposition (SVD) modebetween vegetation optical depth (VOD) and rainfall for Mongolia, north-west and south-east Australia, southern Africa and the Sahel.

Y. Y. Liu et al.


(see Appendix S3). Results indicate that both contribute to the

observed VOD patterns.

To further investigate the influence of rainfall variations,

VOD dynamics for largely intact forests in the Amazon were

analysed separately. We focus on the widespread Amazon

drought that occurred in the 2005 dry season (July–November).

Forest response to this event has received substantial attention in

the literature (Huete et al., 2006; Saleska et al., 2007; Brando

et al., 2010; Samanta et al., 2010; Lewis et al., 2011). Most studies

used the enhanced vegetation index (EVI), a vegetation green-

ness index with characteristics similar to NDVI derived from

optical remote sensing by the MODIS instrument. Like any

optical observation, EVI is affected by vapour, clouds and aero-

sols in the atmosphere, and there has been debate about their

influence on inferred vegetation changes during the drought.

The microwave-based VOD record used here is not sensitive to

such effects and therefore provides a unique opportunity to

examine forest response during the 2005 drought that is com-

plementary to the vegetation index analysis. As pointed out

previously, VOD is underestimated when there are substantial

open water bodies in the observation footprint. Annual average

VOD over the Amazon region (Fig. 7a) show lower values over

areas along the main rivers in central Amazonia that, at least in

part, are probably due to open water contamination, and there-

fore observations in these regions need to be interpreted with

caution.

Standardized anomalies of VOD were calculated for the early

(July–September, JAS) and late (September–November, SON)

dry season of 2005, that is, for each pixel, the anomaly ( xt , the

departure from the average for the same months for 2000–08

excluding 2005) was divided by the corresponding standard

deviation (st). The 2005 drought was concentrated in southern

Amazonia, and peaked during the early dry season (JAS)

(Saleska et al., 2007). Corresponding VOD anomalies are

evident in southern Amazonia (Fig. 7b), and increased during

the late dry season (Fig. 7c). The spatial pattern in negative

anomalies agrees very well with EVI anomalies reported in the

mentioned studies and the rainfall anomalies that caused them.

However, it appears that VOD took longer to recover from the

drought than did EVI, which is perhaps not surprising given the

greater source depth of the VOD signal.

It is noted that apparent positive anomalies occur in the vicin-

ity of the main rivers in central Amazonia (Fig. 7b,c). The mean

and 2005 seasonal cycle for the southern and central Amazon

regions indicated in Fig. 7(b) and (c) are shown in Fig. 7(d). In

the drought-affected southern Amazon, VOD peaks at the end of

the wet season and declines to a minimum at the end of dry

season (Fig. 7d); during 2005 it shows a more severe decline

from the start of the dry season, reaching a minimum in

October and recovering afterwards. By comparison, VOD over

the central Amazon has an almost opposite seasonal cycle, and

showed an increase in 2005 (Fig. 7d). We suspect that the mean

and seasonal cycle in VOD pattern in the central Amazon is

affected by seasonally varying inundation in this region (i.e. the

problem of underestimation of VOD is more apparent for a

larger extent of open water) and that the VOD increase in 2005

probably reflects a reduction in flooded area, rather than an

actual increase in vegetation biomass. This demonstrates that

VOD observations need to be interpreted with care in regions

prone to regular and extensive inundation.

Changes in boreal forests

The EOF analysis was applied to the grid cells with significant

VOD declines over the boreal regions (Alaska and east Russia; cf.

Fig. 2). The first EOF modes explained 68 and 58% of the annual

VOD variance during 1988–2008 over the respective regions.

The spatial patterns of the first EOF modes show good corre-

Figure 4 Annual precipitation (mm,bars) and average vegetation optical depth(VOD) (line) for 1988–2008 for fourdifferent regions. Also listed is theSpearman correlation coefficient (r)between precipitation and VOD.



spondence with remotely sensed fire occurrence (Fig. 8),

although it is noted that the fire and VOD reductions should not

necessarily be expected to be directly related or temporally coin-

cident. That is, a large number of fires does not necessarily lead

to large declines in biomass, whereas declines in living biomass

due to clearing or mortality can precede fire by many months

(due to its sensitivity to water content, VOD will be a better

indicator of living biomass than dead or total biomass). None-

theless, the general spatial agreement between significant VOD

declines and fire occurrence over Alaska and east Russia

(Fig. 8a–d) suggests that fire and/or preceding forest dieback is

the most likely cause for the VOD declines observed over these

regions.

To confirm this, the temporal patterns of the first EOF modes

were compared with the fire frequency (Fig. 8e,f). For Alaska,

the strongest VOD declines occurred around 1990–92 and

2004–06, and coincided with increased fire activity (Fig. 8e). In

eastern Russia, fire frequency data were only available from

remote sensing (2000 onwards) but the significant VOD decline

around 2002–03 again correspond with high fire activity

(Fig. 8f). The year 1998 also witnessed considerable decreases in

VOD, and severe forest fires in the same year or the year before

have been reported for this area (Spichtinger et al., 2004).

Finally, for western Russia (cf. Fig. 2), fire activity could not

explain the observed VOD declines. Reports of large-scale forest

clearing in the areas with the greatest VOD decreases (the

Figure 5 Spatial patterns (left column) andtime series (right column) of the first empiricalorthogonal function (EOF) mode of vegetationoptical depth (VOD) for croplands (i.e. thecropland grid cells with significant changes) inChina, India, the USA, Russia and CanadianPrairies. Also plotted is the normalized timeseries of annual production of dominant cropsin the cropland regions as shown in the leftcolumn.

Y. Y. Liu et al.


Figure 6 (a), (b) Magnitude of vegetation optical depth (VOD) trends (year-1, 1988–2008) for grid cells with significant trends (P < 0.05)over (a) the American tropics and (b) Southeast Asia; (c), (d) Deforestation hot spots (Achard et al., 1998); (e), (f) Estimated tropical forestloss rates (2000–05) (Hansen et al., 2008).



Arkhangel’sk and Komi regions; Achard et al., 2006) suggest that

logging is probably the main driver in this case.

DISCUSSION

The spatial and temporal patterns in VOD observed across mul-

tiple biomes could be attributed through comparison with data

on precipitation, crop production, deforestation and fires. VOD

dynamics over grasslands and shrublands responded strongly to

precipitation variations (Figs 3 & 4). Our interpretation con-

firms various previous studies documenting the effect of rainfall

variability on vegetation dynamics in water-limited ecosystems

(e.g. Evans & Geerken, 2004; Donohue et al., 2009; Good &

Caylor, 2011).

Over croplands, VOD and the production of major crops

showed corresponding patterns and increasing trends (Fig. 5).

Figure 7 (a) Spatial pattern of annual average vegetation optical depth (VOD) calculated from the period 2000–08 over the evergreentropical forest in South America shown in Fig. 1(b). (b), (c) Spatial distribution of standardized anomalies in the early (July–September)and late (September–November) dry season, 2005. The reference period used for anomaly calculation is 2000–08, but excluding 2005. (d)Seasonal cycle of VOD, calculated from the year 2000 through 2008 (with 2005 excluded) and the seasonal cycle of drought year 2005 overthe two outlined circles in (b) and (c).

Y. Y. Liu et al.


The long-term increases in VOD and crop production are at

least partly attributable to non-climatic factors, such as the

adoption of improved crop varieties and cropping practices (e.g.

Liu et al., 2010 for China); increased use of fertilizers (e.g.

Stewart et al., 2005 for North America); irrigation (e.g. Shah,

2009 for India); application of new technologies; and improved

management and agricultural policies. The long-term trends

were modulated by interannual climate variability; we found

that approximately 20% of VOD variations over cropping

regions could be explained by precipitation. This explained vari-

ance relates to interannual variations, but for some regions there

was also evidence of an influence from a climate trend over the

1988–2008 observation period. For example, increased agricul-

tural production in eastern Europe appears to have been assisted

by increases in rainfall (Fig. 3d), whereas decreasing precipita-

tion patterns reversed production trends in south-east Australia

(Fig. 3g-i) and may have attenuated them in China (Fig 5a, cf.

Liu et al., 2010). Other aspects of long-term climate change may

also have played a role. For example, Twine & Kucharik (2009)

found that 20–25% of the increase in US cereal production

between 1982 and 2002 could be explained by the combined

effects of lengthening growing season, reduced summer heat

stress and increased precipitation. Conversely, You et al. (2009)

estimated that rising temperatures caused a 4.5% decline in

wheat production in China over the period 1979–2000.

Over tropical forests, the spatial and temporal distributions of

VOD declines correspond well with known deforestation hot

spots and reported deforestation rates. Similarly, fires and

Figure 8 Spatial patterns and time seriesof the first empirical orthogonal function(EOF) mode of vegetation optical depth(VOD) for Alaska (left column) andeastern Russia (right column). Alsodisplayed are the average number of fireobservations per year (2001–08) estimatedfrom the MODIS fire product (middlepanel) and annual total number of fires(bottom panel). The total number of firesshown in (e) is provided by the Bureau ofLand Management, Alaska Fire Service,while total number of fires over easternRussia (f) is observed by MODIS (i.e. notavailable before 2000).



logging explain the observed VOD reductions in boreal forest

biomass. Fire incidence is likely to be a secondary process facili-

tated by tree mortality (e.g. due to ecological disturbances) or

drought conditions. Regardless of the cause, the VOD data accu-

rately reflect the decreases in biomass associated with removal or

burning. This suggests an opportunity, albeit at a coarse scale, to

overcome some of the challenges in detecting deforestation

using optical remote sensing, in particular for regions where the

forest and replacing land-cover type have similar spectral

properties.

Overall, the VOD data are clearly sensitive to biomass changes

and have the capacity to monitor vegetation dynamics over

various land-cover types as influenced by environmental and

human factors. The VOD record used here can be extended as

passive microwave observations continue to become available

from existing passive microwave missions. In addition, several

new satellites with microwave instruments have been launched

recently or are scheduled for future launch and can be expected

to provide vegetation retrievals with higher accuracy due to

improved instrumental design.

The VOD data have limitations and uncertainties associated

with the inherent characteristics of microwave radiative transfer

physics, satellite instruments and retrieval algorithm. An impor-

tant limitation is its relatively coarse spatial resolution

(> 10 km), necessitated by the low energy of the earth’s micro-

wave emissions. As a consequence, changes taking place at the

small scale are not detected. Therefore, long-term VOD moni-

toring is arguably more suitable for regional, continental and

global investigations. For smaller-scale studies, the passive

microwave data may assist in prioritizing regions where higher-

resolution remote sensing observations from optical or active

radar platforms can be used to provide more detailed analyses.

Second, open water can cause underestimation of microwave

emissions from the land, and as a consequence underestimation

of VOD (Walker et al., 2006). As such, VOD data should be used

with caution over regions that are permanently or seasonally

inundated (cf. Figs 1a & 7).

CONCLUSIONS

An analysis of global vegetation trends using a new satellite

microwave based long-term record of vegetation water content

and biomass of both woody and leafy components was under-

taken. Our results suggest that the VOD record captures changes

in biomass over diverse land-cover types. The observed spatial

and temporal patterns were verified against independent infor-

mation sources, and attributed to their main drivers: rainfall,

agricultural development, deforestation and fire. It is anticipated

that this new data source can help answer challenging questions

about the influence of large-scale vegetation dynamics on the

earth’s water, energy and carbon cycles.

ACKNOWLEDGEMENTS

This work was supported by a University International Post-

graduate Award (UIPA) from the University of New South Wales

and a top-up scholarship from the CSIRO Water for a Healthy

Country Flagship Program.

REFERENCES

Achard, F., Eva, H., Glinni, A., Mayaux, P., Richards, T. & Stibig,

H. (1998) Identification of deforestation hot spot areas in the

humid tropics. TREES Series B, No. 4, EUR 18079 EN. Euro-

pean Commission, Luxembourg.

Achard, F., Mollicone, D., Stibig, H.J., Aksenov, D., Laestadius, L.,

Li, Z.Y., Popatov, P. & Yaroshenko, A. (2006) Areas of rapid

forest-cover change in boreal Eurasia. Forest Ecology and Man-

agement, 237, 322–334.

Alessandri, A. & Navarra, A. (2008) On the coupling between

vegetation and rainfall inter-annual anomalies: possible con-

tributions to seasonal rainfall predictability over land areas.

Geophysical Research Letters, 35, 02718.

Bai, Z.G., Dent, D.L., Olsson, L. & Schaepman, M.E. (2008)

Proxy global assessment of land degradation. Soil Use and

Management, 24, 223–234.

Brando, P.M., Goetz, S.J., Baccini, A., Nepstad, D.C., Beck, P.S.A.

& Christman, M.C. (2010) Seasonal and interannual variabil-

ity of climate and vegetation indices across the Amazon. Pro-

ceedings of the National Academy of Sciences USA, 107, 14685–

14690.

Bretherton, C.S., Smith, C. & Wallace, J.M. (1992) An intercom-

parison of methods for finding coupled patterns in climate

data. Journal of Climate, 5, 541–560.

Ceccato, P., Flasse, S. & Gregoire, J.M. (2002) Designing a spec-

tral index to estimate vegetation water content from remote

sensing data – part 2. Validation and applications. Remote

Sensing of Environment, 82, 198–207.

Cox, P.M., Betts, R.A., Jones, C.D., Spall, S.A. & Totterdell, I.J.

(2000) Acceleration of global warming due to carbon-

cycle feedbacks in a coupled climate model. Nature, 408, 184–

187.

van Dijk, A.I.J. & Keenan, R.J. (2007) Planted forests and water

in perspective. Forest Ecology and Management, 251, 1–9.

Donohue, R.J., McVicar, T.R. & Roderick, M.L. (2009) Climate-

related trends in Australian vegetation cover as inferred from

satellite observations, 1981–2006. Global Change Biology, 15,

1025–1039.

Evans, J. & Geerken, R. (2004) Discrimination between climate

and human-induced dryland degradation. Journal of Arid

Environments, 57, 535–554.

Friedl, M.A., McIver, D.K., Hodges, J.C.F., Zhang, X.Y.,

Muchoney, D., Strahler, A.H., Woodcock, C.E., Gopal, S., Sch-

neider, A., Cooper, A., Baccini, A., Gao, F. & Schaaf, C. (2002)

Global land cover mapping from MODIS: algorithms and

early results. Remote Sensing of Environment, 83, 287–302.

Good, S.P. & Caylor, K.K. (2011) Climatological determinants of

woody cover in Africa. Proceedings of the National Academy of

Sciences USA, 108, 4902–4907.

Hansen, M.C., Stehman, S.V., Potapov, P.V., Loveland, T.R.,

Townshend, J.R.G., Defries, R.S., Pittman, K.W., Arunarwati,

B., Stolle, F., Steininger, M.K., Carroll, M. & DiMiceli, C.

Y. Y. Liu et al.


(2008) Humid tropical forest clearing from 2000 to 2005

quantified by using multitemporal and multiresolution

remotely sensed data. Proceedings of the National Academy of

Sciences USA, 105, 9439–9444.

Helsel, D.R. & Hirsch, R.M. (2002) Statistical methods in

water resources. Techniques of water-resources investiga-

tions, Book 4, Chapter A3. United States Geological Survey,

Reston, VA.

Huete, A.R., Didan, K., Shimabukuro, Y.E., Ratana, P., Saleska,

S.R., Hutyra, L.R., Yang, W.Z., Nemani, R.R. & Myneni, R.

(2006) Amazon rainforests green-up with sunlight in dry

season. Geophysical Research Letters, 33, L06405.

Jones, M.O., Jones, L.A., Kimball, J.S. & McDonald, K.C. (2011)

Satellite passive microwave remote sensing for monitoring

global land surface phenology. Remote Sensing of Environ-

ment, 115, 1102–1114.

Kobayashi, H. & Dye, D.G. (2005) Atmospheric conditions for

monitoring the long-term vegetation dynamics in the

Amazon using normalized difference vegetation index.

Remote Sensing of Environment, 97, 519–525.

Lehner, B. & Doll, P. (2004) Development and validation of a

global database of lakes, reservoirs and wetlands. Journal of

Hydrology, 296, 1–22.

Lewis, S.L., Brando, P.M., Phillips, O.L., van der Heijden, G.M.F.

& Nepstad, D. (2011) The 2010 Amazon drought. Science, 331,

554–554.

Liu, Y., Wang, E., Yang, X. & Wang, J. (2010) Contributions of

climatic and crop varietal changes to crop production in the

North China Plain, since the 1980s. Global Change Biology, 16,

2287–2299.

Liu, Y.Y., de Jeu, R.A.M., McCabe, M.F., Evans, J.P. & van Dijk,

A.I.J.M. (2011a) Global long-term passive microwave

satellite-based retrievals of vegetation optical depth. Geophysi-

cal Research Letters, 38, L18402.

Liu, Y.Y., Parinussa, R.M., Dorigo, W.A., de Jeu, R.A.M.,

Wagner, W., van Dijk, A.I.J.M., McCabe, M.F. & Evans,

J.P. (2011b) Developing an improved soil moisture dataset

by blending passive and active microwave satellite-

based retrievals. Hydrology and Earth System Sciences, 15, 425–

436.

Milesi, C., Samanta, A., Hashimoto, H., Kumar, K.K., Ganguly,

S., Thenkabail, P.S., Srivastava, A.N., Nemani, R.R. & Myneni,

R.B. (2010) Decadal variations in NDVI and food production

in India. Remote Sensing, 2, 758–776.

Mitchell, T.D. & Jones, P.D. (2005) An improved method of

constructing a database of monthly climate observations and

associated high-resolution grids. International Journal of Cli-

matology, 25, 693–712.

Myneni, R.B., Hall, F.G., Sellers, P.J. & Marshak, A.L. (1995) The

interpretation of spectral vegetation indexes. IEEE Transac-

tions on Geoscience and Remote Sensing, 33, 481–486.

Owe, M., de Jeu, R. & Walker, J. (2001) A methodology for

surface soil moisture and vegetation optical depth retrieval

using the microwave polarization difference index. IEEE

Transactions on Geoscience and Remote Sensing, 39, 1643–

1654.

Rudolf, B. & Schneider, U. (2005) Calculation of gridded pre-

cipitation data for the global land-surface using in-situ gauge

observations. In: Proceedings of the 2nd Workshop of the Inter-

national Precipitation Working Group IPWG, Monterey.

EUMETSAT, Darmstadt, Germany.

Saleska, S.R., Didan, K., Huete, A.R. & da Rocha, H.R. (2007)

Amazon forests green-up during 2005 drought. Science, 318,

612.

Samanta, A., Ganguly, S., Hashimoto, H., Devadiga, S., Vermote,

E., Knyazikhin, Y., Nemani, R.R. & Myneni, R.B. (2010)

Amazon forests did not green-up during the 2005 drought.

Geophysical Research Letters, 37, L05401.

Seaquist, J.W., Hickler, T., Eklundh, L., Ardo, J. & Heumann,

B.W. (2009) Disentangling the effects of climate and

people on Sahel vegetation dynamics. Biogeosciences, 6,

469–477.

Shah, T. (2009) Climate change and groundwater: India’s

opportunities for mitigation and adaptation. Environmental

Research Letters, 4, 035005.

Shi, J.C., Jackson, T., Tao, J., Du, J., Bindlish, R., Lu, L. & Chen,

K.S. (2008) Microwave vegetation indices for short vegetation

covers from satellite passive microwave sensor AMSR-E.

Remote Sensing of Environment, 112, 4285–4300.

Spichtinger, N., Damoah, R., Eckhardt, S., Forster, C., James, P.,

Beirle, S., Marbach, T., Wagner, T., Novelli, P.C. & Stohl, A.

(2004) Boreal forest fires in 1997 and 1998: a seasonal com-

parison using transport model simulations and measurement

data. Atmospheric Chemistry and Physics, 4, 1857–1868.

Stewart, W.M., Dibb, D.W., Johnston, A.E. & Smyth, T.J. (2005)

The contribution of commercial fertilizer nutrients to food

production. Agronomy Journal, 97, 1–6.

von Storch, H. & Zwiers, F.W. (1999) Statistical analysis in

climate research. Cambridge University Press, Cambridge.

Tucker, C.J., Pinzon, J.E., Brown, M.E., Slayback, D.A., Pak, E.W.,

Mahoney, R., Vermote, E.F. & El Saleous, N. (2005) An

extended AVHRR 8-km NDVI dataset compatible with

MODIS and SPOT vegetation NDVI data. International

Journal of Remote Sensing, 26, 4485–4498.

Turner, D.P., Cohen, W.B., Kennedy, R.E., Fassnacht, K.S. &

Briggs, J.M. (1999) Relationships between leaf area index and

Landsat TM spectral vegetation indices across three temperate

zone sites. Remote Sensing of Environment, 70, 52–68.

Twine, T.E. & Kucharik, C.J. (2009) Climate impacts on net

primary productivity trends in natural and managed ecosys-

tems of the central and eastern United States. Agricultural and

Forest Meteorology, 149, 2143–2161.

Walker, J.P., Merlin, O., Panciera, R. & Kalma, J.D. (2006)

National airborne field experiments for soil moisture remote

sensing. In: Proceedings of the 30th Hydrology and Water

Resources Symposium. Launceston, Tasmania, Australia.

Available at: http://users.monash.edu.au/~jpwalker/papers/

hwrs06-1.pdf (accessed 26 November 2012)

You, L.Z., Rosegrant, M.W., Wood, S. & Sun, D.S. (2009)

Impact of growing season temperature on wheat producti-

vity in China. Agricultural and Forest Meteorology, 149, 1009–

1014.



SUPPORTING INFORMATION

Additional Supporting Information may be found in the online

version of this article:

Appendix S1 Box plots of vegetation optical depth by major

land-cover types.

Appendix S2 Trends in annual precipitation between 1988 and

2008.

Appendix S3 Empirical orthogonal function analysis results for

the ‘arc of deforestation’ in Brazil.

BIOSKETCH

Yi Liu is a Senior Research Associate at the Water

Research Centre of the University of New South Wales,

Sydney, Australia. His research interest is in using

satellite-based observations to investigate the

hydrological cycle, e.g. precipitation, evaporation, soil

moisture, vegetation water content and groundwater

storage, for enhanced understanding of interactions

between different hydrological components.

Editor: Navin Ramankutty

Y. Y. Liu et al.


bs bs banner research global vegetation biomass change...

Documents