fra 2000 f c m & m noaa-avhrr other coarse … · 5.5 firs – forest information from remote...

Forestry DepartmentFood and Agriculture Organization of the United Nations

Forest Resources Assessment Programme Working Paper 29Rome 2000

FRA 2000

FOREST COVER MAPPING& MONITORING WITH

NOAA-AVHRR& OTHER COARSE SPATIAL

RESOLUTION SENSORS

Rome, 2000

The Forest Resources Assessment Programme

Forests are crucial for the well-being of humanity. They provide foundations for life on earth through ecologicalfunctions, by regulating the climate and water resources, and by serving as habitats for plants and animals.Forests also furnish a wide range of essential goods such as wood, food, fodder and medicines, in addition toopportunities for recreation, spiritual renewal and other services.Today, forests are under pressure from expanding human populations, which frequently leads to theconversion or degradation of forests into unsustainable forms of land use. When forests are lost or severelydegraded, their capacity to function as regulators of the environment is also lost, increasing flood and erosionhazards, reducing soil fertility, and contributing to the loss of plant and animal life. As a result, the sustainableprovision of goods and services from forests is jeopardized.FAO, at the request of the member nations and the world community, regularly monitors the world’s foreststhrough the Forest Resources Assessment Programme. The next report, the Global Forest ResourcesAssessment 2000 (FRA 2000), will review the forest situation by the end of the millennium. FRA 2000 willinclude country-level information based on existing forest inventory data, regional investigations of land-coverchange processes, and a number of global studies focusing on the interaction between people and forests.The FRA 2000 report will be made public and distributed on the world wide web in the year 2000.The Forest Resources Assessment Programme is organized under the Forest Resources Division (FOR) atFAO headquarters in Rome. Contact persons are:

Robert Davis FRA Programme Coordinator [email protected]

Peter Holmgren FRA Project Director [email protected]

or use the e-mail address: [email protected]

DISCLAIMER

The Forest Resources Assessment (FRA) Working Paper Series is designed to reflect the activitiesand progress of the FRA Programme of FAO. Working Papers are not authoritative information sources –they do not reflect the official position of FAO and should not be used for official purposes. Please refer to theFAO forestry website (www.fao.org/fo) for access to official information.

The FRA Working Paper Series provides an important forum for the rapid release of preliminary FRA2000 findings needed for validation and to facilitate the final development of an official quality-controlled FRA2000 information set. Should users find any errors in the documents or have comments for improving theirquality they should contact either Robert Davis or Peter Holmgren at [email protected].

mailto:[email protected]

mailto:[email protected]

3

Table of Contents

ABSTRACT......................................................................................................................................................................... 6

1 INTRODUCTION ..................................................................................................................................................... 7

2 UTILITY OF NOAA-AVHRR FOR FOREST MAPPING/CHANGE DETECTION........................................ 8

2.1 AREA OF COVERAGE, IMAGE AVAILABILITY, TEMPORAL RESOLUTION .................................................................. 82.2 SPECTRAL RESOLUTION........................................................................................................................................... 92.3 SPATIAL AND GEOMETRIC CHARACTERISTICS....................................................................................................... 10

3 MAPPING CONSIDERATIONS........................................................................................................................... 12

4 CLASSIFICATION METHODOLOGIES .......................................................................................................... 14

4.1 SINGLE IMAGE VISUAL INTERPRETATION.............................................................................................................. 144.2 SINGLE-IMAGE RADIANCE OR BRIGHTNESS TEMPERATURE THRESHOLDING........................................................... 144.3 SINGLE-IMAGE DIGITAL SUPERVISED OR UNSUPERVISED CLASSIFICATION............................................................. 144.4 MULTI-TEMPORAL, MULTI-SPECTRAL CLASSIFICATION ......................................................................................... 154.5 REGRESSION ANALYSIS AND MIXTURE MODELLING ............................................................................................. 164.6 FOREST MONITORING AND CHANGE ANALYSIS CONSIDERATIONS........................................................................ 16

5 CASE STUDIES ...................................................................................................................................................... 18

5.1 GENERAL............................................................................................................................................................... 185.2 IGBP-DISCOVER: A GLOBAL LAND COVER CLASSIFICATION AT 1KM RESOLUTION ............................................. 18

5.2.1 Purpose............................................................................................................................................................. 185.2.2 Legend .............................................................................................................................................................. 185.2.3 Data Sources .................................................................................................................................................... 205.2.4 Classification and Mapping Methodology........................................................................................................ 205.2.5 Validation and Accuracy .................................................................................................................................. 215.2.6 Coverage, Resolution and Availability ............................................................................................................. 225.2.7 Problems Or Constraints in Use....................................................................................................................... 22

5.3 FAO FRA 2000 GLOBAL FOREST COVER MAP.................................................................................................... 235.3.1 Purpose............................................................................................................................................................. 235.3.2 Legend .............................................................................................................................................................. 235.3.3 Data Sources .................................................................................................................................................... 235.3.4 Classification and Mapping Methodology........................................................................................................ 235.3.5 Validation and Accuracy .................................................................................................................................. 245.3.6 Coverage, Resolution and Availability ............................................................................................................. 255.3.7 Problems Or Constraints In Use ...................................................................................................................... 25

5.4 TREES (TROPICAL ECOSYSTEM ENVIRONMENT OBSERVATION BY SATELLITES).................................................. 255.4.1 Purpose............................................................................................................................................................. 255.4.2 Legend .............................................................................................................................................................. 255.4.3 Data Sources .................................................................................................................................................... 265.4.4 Classification and Mapping Methodology........................................................................................................ 265.4.5 Validation and Accuracy .................................................................................................................................. 265.4.6 Coverage, Resolution and Availability ............................................................................................................. 265.4.7 Problems Or Constraints In use ....................................................................................................................... 27

5.5 FIRS – FOREST INFORMATION FROM REMOTE SENSING: A FOREST INFORMATION SYSTEM FOR THE ENTIRE EUROPEAN CONTINENT........................................................................................................................................... 27

5.5.1 Purpose............................................................................................................................................................. 275.5.2 Legend .............................................................................................................................................................. 275.5.3 Data Sources .................................................................................................................................................... 275.5.4 Classification and Mapping Methodology........................................................................................................ 275.5.5 Validation and Accuracy .................................................................................................................................. 28

5.6 OTHER FOREST COVER MAPS................................................................................................................................ 285.7 SUMMARY ............................................................................................................................................................. 28

4

6 COMBINING HIGH AND LOW SPATIAL RESOLUTION CLASSIFICATIONS........................................ 30

7 BIBLIOGRAPHY.................................................................................................................................................... 32

FRA WORKING PAPERS .................................................................................................................................................. 42

Paper drafted by: Dr. Giles D'Souza, Director Geographic Data Support LimitedIn cooperation with the staff of the FAO FRA 2000 Programme

Editorial production: Patrizia Pugliese, FAO FRA-Programme

5

Abbreviations

BEF Biomass Expansion Factor

BV Biomass of inventoried volume

CATIE Centro Agronómico Tropical de Investigación y Enseñanza

Cirad Centre de coopération internationale en recherche agronomique pour le développement

EDC Eros Data Centre

FAO Food and Agricultural Organization of the United Nations

FORIS Forest Resources Information System

FRA Forest Resources Assessment

GIS Geographic Information System

NOAA -

AVHRR

National Oceanic and Atmospheric Agency – Advanced Very High Resolution

Radiometer

SNU Sub National Unit(s)

UN-ECE United Nations Economic Commission for Europe

VOB Volume Over Bark

WD Wood Density

WCMC World Conservation Monitoring Centre

6

Abstract

This paper represents a review of the use of coarse spatial resolution satellite data, mainly NOAA-AVHRR, for forest cover mapping and monitoring all over the globe. Information about the sensor’sutility in respect of this application, appropriate scales, differentiation of forest classes, coverage(global, tropical, regional, national, etc.), accuracy of the results, integration with high spatialresolution results and problems/constraints in use, are all discussed.

Other programmes that have used this instrument for forest mapping and monitoring (i.e. TREES,IGBP, FAO and other regional programmes) are reported on, and a brief review of the applicabilityof new coarse spatial resolution data (e.g. VEGETATION, ATSR) is also presented.

7

1 IntroductionForests are crucial for the well-being of humanity, and the need to map and monitor their extent andcondition in a timely fashion at local, national, regional and global scales is growing increasinglyurgent. However, forests are very complex and difficult biomes to classify and map, evenconventionally (eco-floristically). Because of the urgent need to map forest extent and condition,several organisations have turned to remote-sensing methods from which to derive their information.And because the areas that need to be mapped are typically large, many have turned to coarse spatialresolution data obtained from sensors flown onboard polar-orbiting satellites. In this paper we willreview the use of coarse spatial resolution data from the U.S. National Oceanic and AtmosphericAdministration's (NOAA) Advanced Very High Resolution Radiometer (AVHRR) sensor in studiesof forest mapping and monitoring.

Although specifically designed for meteorological applications, data from the NOAA-AVHRR intheir various forms, have also been widely used for a variety of land applications, for example cropmonitoring in the Nile Delta (Tucker et. al., 1984a), grassland monitoring in Senegal (Tucker et. al.,1983) and the Sahel (Prince et. al., 1990), continental or global-scale vegetation classification(Townshend et. al., 1991), forest and bush fire mapping and monitoring (Gregoire, 1995), andtropical deforestation monitoring in South East Asia and South America (Malingreau et. al.,1989).

Among the reasons cited for their use in land applications are:

• the suitability of their spatial, spectral and temporal resolutions,• the large area of coverage offered,• the operational and guaranteed nature of the system (guaranteeing continuity of data in the same

format to users from 1981 to at least 2005 (NOAA,1998),• the relative cheapness and accessibility of the data, and• the manageable data rates for use over large areas.

For the forest mapping and monitoring over large areas at national or regional scales as required byprogrammes such as the FAO Forest Resources Assessment 2000, these reasons are equallyapplicable.

8

2 Utility of NOAA-AVHRR for forest mapping/change detection

2.1 Area of Coverage, Image Availability, Temporal Resolution

In its rawest full-resolution form, known as High Resolution Picture Transmission (HRPT) or LargeArea Coverage (LAC), a single swath of data from the AVHRR sensor spans about 2700km in thecross-track direction, and a single recording of 10 minutes covers more than 3600 kilometres in thealong-track direction. So, for example, the whole of the Amazon Basin could be covered by only twosuch scenes. These would have a spatial resolution ranging from 1.1 by 1.1km at centre swath (near-nadir) to about 2.4 by 6.9 km at the edge of swath, i.e. at the AVHRR's most extreme view angles of55.4°=off-nadir (Belward, 1991). Assuming the spatial and spectral resolutions were detailed enoughfor the purpose at hand this would offer considerable time and cost advantages over the purchase andprocessing of over 200 Landsat TM or 1800 SPOT images to cover the same area.

Although the orbital characteristics of the NOAA series of satellites are very similar to those of theLandsat series (i.e. 14.1 orbits per day, sun-synchronous, near-polar orbit at 99.1° orbitalinclination), the higher altitude (nominally 833km), the wider swath-width of the AVHRR sensor,and the fact that the NOAA series of satellites usually operate in complementary pairs, togetherpermit a much higher theoretical frequency of image acquisition for the same area (albeit withdifferent look angles). Potentially one image per day may be acquired if afternoon overpasses onlyare considered, or 4 images per 24-hour period if night-time and early morning overpasses are alsosuitable (for example in applications requiring thermal waveband detection only). Even consideringjust afternoon images that are reasonably near-nadir, at least one image every 5 days should bepossible (Goward et. al., 1991).

This high image repeatability, or temporal resolution, of the data is very important in tropical areaswhere cloud cover is high and/or where intra-seasonal or phenological variations in time have to bemonitored. For most land applications, the images acquired by early morning overpasses of theNOAA satellites are not very suitable since at that time (about 0730 local solar time), the light levelsare too low and the thermal contrast between differing land cover types is not sufficiently developed.The light levels and thermal contrast are more suitable at the time of the NOAA satellite afternoonoverpass time (about 1430 local solar time). However, this is normally also a time of intense localconvection in tropical areas and there is therefore also an increased likelihood of cloud and hazecontamination at that time. In sub-tropical or more extreme latitudes, frontal cloudiness and lowerlight levels for longer periods of the year provide other restrictions to suitable image acquisition(Kasischke and French, 1997).

In practice, therefore, the theoretical rate of daily image acquisition is rarely achieved so a largenumber of daily AVHRR images have to be examined and/or composited to obtain sufficientlycloud-free scenes for forest mapping and monitoring (Malingreau et. al., 1989). In the past, thepotential daily acquisition for certain areas may also have been hampered by lack of data capture andarchiving facilities, but the number and distribution of local and regional receiving and archivingfacilities for NOAA data acquisition has continued to grow and improve. So have the facilities forexchange and transmission of data acquired all over the world, particularly with the IGBP-DISproject – a concerted global effort to collect full-resolution AVHRR data for all land surfaces on adaily basis for over 54 months in 1992 – 6 (Townshend, 1992; Belward et. al., 1999)

Another characteristics of the NOAA-AVHRR systems that should be noted, is the marked orbitaldrift which causes the Equator crossing time of the orbit to change by several hours during a three or

9

four year period (Price, 1991; Privette et. al., 1995). This together the gradual degradation of thesensor in the visible channels over time (D’Souza, 1996; Rao and Chen, 1999) should be borne inmind when examining long quantitative time series of data from the AVHRR sensor.

2.2 Spectral Resolution

All of the AVHRR sensors have had four to five channels, sensing data in the visible red, near-infrared, shortwave-infrared and thermal channels. In AVHRR land applications the first twochannels of the radiometer, Channel 1 (0.58-0.68 µm ) sensitive to red reflected light, and Channel 2(0.725-1.10µm), sensitive to infra-red reflected light, have been the most widely used.Photosynthetically-active vegetation will typically yield a low reflectance at red wavelengths(chlorophyll pigment absorption) and a high reflectance at near-infrared wavelengths (caused byscattering of leaf surface and internal structure). Because it is the only material on the Earth’s surfacethat will exhibit this contrast, the two channels are useful for discriminating and monitoring thecondition of vegetation.

Often the reflectances from the two channels are combined mathematically to form a spectralvegetation index. Spectral vegetation indices attempt to discriminate between vegetated and non-vegetated areas by enhancing the spectral contribution of green vegetation whilst minimising thecontribution of the background. Numerous spectral vegetation indices have been developed basedupon combinations of the first two channels. The normalised difference vegetation index (NDVI) isthe most commonly used AVHRR vegetation index and is defined as the difference between near-infrared and red reflectances divided by their sum (e.g. Curran, 1983). The NDVI has beendemonstrated to be a robust and sensitive vegetation measure and is the only vegetation index thathas been used widely for continental and global-scale vegetation dynamics examination (i.e. Tuckeret. al., 1984; Justice et. al., 1985, etc.., and see box 4). The widespread use of the NDVI is, in part,due to its “ratioing” properties, which cancel out a large proportion of signal variations attributed tochanging irradiance conditions.

However, the uncalibrated wide-wavelength bands used for the first two channels of the AVHRR areparticularly sensitive to the smoke, water vapour, aerosols and other atmospheric particles that areprevalent in the atmosphere in tropical regions in particular. Canopy structure and background (soil,snow, wetness, litter), thin and sub-pixel cloud, illumination and viewing geometry effects have alsobeen shown to effect the AVHRR NDVI values. Furthermore, near infrared/red vegetation indices areoften more sensitive to leaf-area than to standing biomass (Sellers, 1986; Goward et. al., 1991; Loset. al., 1994; Meyer et. al., 1995; Walter-Shea et. al., 1997). Therefore dense forest areas may yieldsimilar (saturated) vegetation indices to recently regenerated or agricultural areas although in termsof standing biomass these two are vastly different (Tucker et. al., 1984b). Furthermore, the NDVIand other vegetation indices have been found to be particularly insensitive in dense forest areas orareas of complex canopies with several layers of shadow interactions and effects (Singh, 1987).Depending on the understorey, closed conifer cover often shows an inverse relationship with NDVI(Ripple, 1994; Ripple et. al., 1991, Spanner et. al., 1990).

On some single date NOAA-AVHRR images, channel 3 (3.5-3.9µm), sensitive to reflected andemitted shortwave-infrared light, has been found to be more useful for forest/ non-forest delineation(Kerber & Schutt, 1986). This channel has advantages (at least when used in conjunction with thevisible and near-infrared channels) because:

• this wavelength is not so contaminated by aerosols and smoke particles (Bird, 1984);

10

• it is sensitive to emitted (thermal) energy, where non-forested areas are characterised by a highertemperature than the surrounding cool forest because of differences in the sensible-latent heatbalance;

• it is also sensitive to reflected energy, where non-forested areas are normally characterised by ahigher albedo than the forested areas where the inter-canopy shadow effects act to trap visiblelight.

Thus the presence of dense forest areas work to reduce both the reflected and the emitted parts of thesignal in channel 3 while non-forested areas generally yield increased components of both. One ofthe disadvantages of channel 3 is that the signal obtained in this wavelength can be prone to noise,especially in earlier satellites (NOAA-7 and NOAA-9) (Warner, 1989) and the mixedemitted/reflected signal is also occasionally difficult to interpret.

Some work carried out on splitting the channel 3 response to its reflected and emitted parts (Holbenand Simabukuro, 1993; Kaufman and Kendall, 1994) has showed some promise, but has beenhampered by the need to know the thermal emissivity of the surface features being examined beforereliable splits can be achieved.

Channel 3 is also very sensitive to the presence of very hot areas within pixels, either from active orrecent fires (Gregoire et. al., 1988; Malingreau et. al., 1985). The occurrence of fire, especiallywithin dense forested areas is a good indicator of human activities (e.g. forest clearing forcultivation), and is often used as "circumstantial evidence" of forest degradation activities(Malingreau, 1990).

The fully-calibrated channels 4 (10.5 – 11.3µm) and 5 (11.5 – 12.5µm) (the thermal sensitive bands)of the AVHRR have also been used for fire detection and characterisation (Matson, 1987; Matson &Holben, 1987). Although they show some sensitivity to forest/non-forest areas (Kerber, 1983; Kerber& Schutt, 1986), these channels are also often heavily contaminated by the water vapour which isever present in the tropical atmosphere. They have therefore been little used for characterisation orclassification of tropical forest areas on single-date AVHRR images. However, the seasonal variationof surface temperature as estimated from AVHRR channels 4 and 5 in multitemporal data sets hasbeen shown to be closely related to the land cover type in West Africa (Achard & Blasco, 1989). Ithas also been used in conjunction with NDVI time series for land cover classification over Africa(Lambin and Ehrlich, 1996) and for forest classification over Europe (Roy et. al., 1997).

The other use for channels 4 and 5 in land applications has been in masking out areas of cloud orhaze (Saunders & Kriebel, 1990).

2.3 Spatial and Geometric Characteristics

Raw NOAA-AVHRR images are characterised by a very distorted geometry (Emery et. al., 1989).They have a nominal spatial resolution of approximately 1.1 by 1.1 km. However, because of thelarge swath width, the very off-nadir viewing angles, and the Earth's curvature, the spatial resolutiondecreases to about 2.4km (in the along-track direction) by 6.9 km at the most off-nadir viewing angleof the sensor. The varying pixel resolution compounded by the over sampling of the pixels in theacross-track direction (Mannstein and Gessell, 1991; Breaker, 1990) makes it difficult to use the datain any detailed spatial or textural pattern analyses (Belward and Lambin, 1990), but if the use of thedata is restricted to that acquired at less than 25° off-nadir the problem is considerably reduced(Goward et. al., 1991; Vogt, 1992).

11

Much literature refers to NOAA AVHRR data as 1.1km resolution, and several products are made ata nominal pixel cell size of 1km. In practice, because of the spatial characteristics of the imagery, theorbital characteristics of the NOAA satellite and the subsequent processing which may includeautomated remapping of the data, the actual spatial resolution of the data is somewhat coarser. Evenwith the most advanced geometric correction methods of the data, the locational accuracy ofremapped data, especially if carried out over large areas, is probably closer to 1.5 – 2.0 km (D’Souzaand Sandford, 1995). Where several images are composited (i.e. for multi-temporal analyses), theindividual image mis-registration is compounded, and the effective pixel size of a multi-datecomposite could be as high as 5.1km, assuming a view zenith angle cutoff of 57 degrees and amisregistration root mean square of 1km (Cihlar et. al., 1996).

However, despite the geometrical problems and the somewhat poor spatial resolution compared withthat of other Earth Resources satellites, the imagery has been found to be suitable for the detection ofa number of features relevant to forest monitoring, especially where single-date , good quality imagesare used (see box 3). Many useful maps and datasets have been published at scales ranging from 1: 1million to 1 : 5 million, and these will be described in detail later.

12

3 Mapping considerationsRemotely sensed images offer a new perspective on the land surface and environment, and a trainedimage interpreter can draw an enormous amount of information from even a qualitative examinationof an enhanced image. However, the demand for statistical measures, and for comparisons withconventionally-derived maps normally leads to a process of classification of remotely-sensed datainto a thematic map. It is widely recognised that a thematic map is in itself a considerablecondensation of the information available from a survey (Goodchild, 1995). One of the majorchallenges in the use and adoption of remotely sensed data is to reconcile the type of observationsmade from satellite-based sensors with classifications that have been derived from direct ground-based eco-floristic classifications. Additionally, eco-floristic definitions and classifications often varyacross national or regional boundaries.

The problem is probably more marked with coarser spatial resolution data. Ground-basedclassification schemes are often based on direct detailed observation and physical measurement ofeco-floristic characteristics over quite small and widely distributed sample areas or transects, oftenjust a few metres wide. Conventional vegetation cover maps are often produced by interpolation andgeneralisation of point or transect data into polygons, with abrupt and distinct boundaries shownbetween different cover types, i.e. forest to non-forest. Contrast this to a complete coverage of largebut remote and indirect samples of light from 1.1km squares that form the basis of NOAA AVHRRdata. Satellite-based data are continuous in space, but indirect measures of what is on the ground,averaged to the pixel size sample of measurement. Both datasets have their advantages anddisadvantages, and the real benefit is to use the two systems synergistically.

NOAA AVHRR data are normally resampled to a grid with a nominal cell size of 1km (pixel). Theclassification of these data then involves an attempt to assign a category to this 1km pixel, accordingto some predetermined legend. Statistics can then be generated by summing the number of pixelsassigned to this category and thematic maps can be produced. Some classification methodologiesassign more than one value to a pixel, for example, a category, and a reliability, or proportion. Themethods used for the categoric assignment are discussed in the next section. However, classificationtechniques applied to remotely-sensed data, especially to those of coarse spatial resolution such asNOAA AVHRR, have the inherent problem that pixels seldom contain exclusively one distinctground cover class (e.g. Foody et. al., 1992). Furthermore, classification techniques are not wellsuited for the estimation of forest cover variables, which are more continuous than discrete in nature.Numerous studies have shown that the accuracy of classification of NOAA AVHRR data is directlyrelated to the purity of the ground cover in the AVHRR size pixels (Cihlar et. al., 1996).

Accuracy assessments of digital land cover classifications are typically based on contingency tables,or confusion matrices, where accuracy is expressed in terms of errors of omission and commission, orin terms of agreement analysis using the Kappa test statistic (Stehman, 1996a; Congalton, 1991;Rosenfeld and Fitzpatrick-Lins, 1986). The contingency table is created by comparing on a class byclass basis the land cover classification with an independent data source - field observations, existingmaps, higher resolution imagery - collected using a statistically valid sampling strategy (Robinson et.al., 1983; Stehman, 1996b; Rosenfeld, 1986). Whilst such methods are well established they havegenerally only been applied to local scale classifications, occasionally to regional scale work andonly in isolated instances on scales greater than these (Hay, 1988; Manshard, 1993; Estes andMooneyhan, 1994; Jeanjean et. al., 1996). The general method of validating NOAA AVHRR-basedclassifications by employing classification of higher spatial resolution remotely-sensed data is nowwell established (Fitzpatrick-Lins, 1980; Rosenfield et. al., 1981) and has been employed extensively

13

at local and regional scales (Borella et. al., 1982; Estes et. al., 1987). However, it is often based onthe assumption that the results derived from the high spatial resolution imagery is completelyaccurate and reliable. This is obviously not the case – they themselves will include some imprecisionand perhaps mis-classification, especially when seasonal patterns are prevalent but the purchase ofmulti-date high resolution imagery is prohibitive. In general though, it is generally felt that dataderived from the higher spatial resolution imagery should be more reliable than those derived fromcoarser spatial resolution data, at least for the detailed spatial representation of the different landcover patterns. Furthermore, it is much more convenient to carry out field checking with higherspatial resolution imagery (say at scales of 1: 25,000) than with coarse spatial resolution data. Errorsin the AVHRR-based classifications due to scale differences are often measured by inter-comparisons of the two co-registered sets of data (see section 6 below).

Because the satellite-based data are in a digital and “raster” form, they could in theory be reproducedat any scale. However, it is generally accepted that the nominally 1km-resolution data are generallysuitable for reproduction of paper maps at one to one million scale. Most paper-based products ofNOAA AVHRR data tend to be at smaller scales than this, typically 1: 5 million scale (see forexample Stone et. al., 1993).

14

4 Classification methodologiesSeveral image classification methods for large-area forest delineation from AVHRR data are found inthe literature and these may be grouped into four main categories.

4.1 Single Image Visual Interpretation

This is the simplest and most flexible approach, since the eye-brain system can make allowances forthe inevitable large variations in feature appearance that occur with different viewing angles anddifferential bidirectional reflectance of various surfaces. At the same time local variations can bedetected for improved feature discrimination. Visual interpretations made by trained people familiarwith the relevant area and image data characteristics, can be accurate in many cases. However, thedelineation of boundaries is also time-consuming, difficult and somewhat subjective. If theinterpretations are made on a hard-copy image rather than on-screen image displays, then it oftenrequires digitisation for input into a GIS. A further drawback of visual interpretation of hard-copyimages is that it can be based on an image representation of, at most, three spectral bands at any onetime, and the results can be quite dependent on the contrast-stretch or enhancement used. However,this method is easy to understand and teach, and it requires no specialised hardware. Examples of thesuccessful use of this method are provided in Tucker et. al. (1984), Malingreau and Tucker (1990)and Malingreau et. al. (1989).

4.2 Single-image radiance or brightness temperature thresholding

Single-image radiance or brightness temperature thresholds applied to channels 3 and/or 4 have beenfound to delineate quite well the forest/non-forest boundary, at least in areas where the transition isvery sharp, e.g. Malingreau and Tucker 1988, Malingreau et. al.1989, Stone et. al. 1991; Woodwellet. al.1986, Amaral 1992, Cross 1990, Cross 1991). The technique is simple to apply and understand,but requires visual interaction with the data. Also, even in the cases cited, the thresholds used havediffered both in time and space, and the method is not reliable in areas of diffuse or complexforest/non-forest interfaces.

4.3 Single-image digital supervised or unsupervised classification

This technique offers an improvement on the previous methods, because it is said to be morequantitative, objective and can be based on more than three channels at a time. Examples of itssuccessful application are provided in Cross (1990), Stone et. al. (1991), Paivinen and Witt (1988)and Horning and Nelson (1993). Normally, a supervised classification requires recourse to quitedetailed geo-located reference data that define the locations and extent of representative land coverunits for classification training and accuracy assessment procedures. An unsupervised approach canbe more automatic in the first stage, but requires significant human interaction and reference data toassign meaningful classes to the resultant spectral clusters. In some cases, however, supervised andunsupervised approaches have been found to yield quite similar results, as in the mapping ofdeforestation in Rondonia state in Brazil (Skole 1993).

15

Because NOAA AVHRR images cover such large areas, the quality of classification may vary acrossthe image, or some parts of some images may be obscured by haze or cloud. In these cases, asuccessful approach by D’Souza et. al. (1995) was to carry out several image classifications foroverlapping areas, and assign a reliability value at each location. The classification with the mostreliable indicator was then retained in the final product.

4.4 Multi-temporal, multi-spectral classification

This method as used by Achard et. al. (1993) involves the examination of the behaviour of aparticular radiometric characteristic (e.g. vegetation index or surface temperature) through time. Thisis the most sophisticated and best method for making a seasonal/evergreen forest discrimination.However, it relies on a good set of high quality (cloud and haze-free) imagery throughout the seasonand is therefore normally only practical in tropical areas where there is a marked dry season, or whena large series of data is available. This technique usually involves the reduction of the AVHRR timeseries into weekly, 10-day or monthly maximum value composites (Holben, 1996), and until recentlywere performed on very low spatial (nominally 8km) sampled NOAA AVHRR data (e.g. Townshendet. al., 1987).

With the recent improvements in computing power, and in the global acquisition and processing ofNOAA AVHRR data (see D’Souza et. al., 1995), 10-day maximum value composites have recentlybeen developed at continental and global scales for the full spatial resolution data. Loveland et. al.,(1991) showed that such monthly full spatial resolution NDVI composites could be used tocharacterise land cover types at continental scales and a resolution of nominally 1km. In their workNDVI monthly composites were used to develop a land cover database for the conterminous UnitedStates. Eight monthly NDVI composites were used in an automatic unsupervised classificationscheme to produce 70 spectral temporal “seasonally-distinct” classes. These were then refined andused with expert knowledge and a wide variety of ancillary data to subjectively assign class labels toeach pixel. The labels assigned were taken from the USGS land cover classification. This work iscontinuing under the auspices of the International Global Biosphere Programme (IGBP). SeeBelward and Loveland (1995) and section later.

Other work that uses multitemporal NDVI composites seeks to extract phyto-phenological measuresor characteristics rather than categoric land cover classes from the data (see Lloyd 1990). Suchmeasures included maximum photosynthetic activity, the length of growing season, and mean dailyNDVI. DeFries et. al. (1995) also concluded that decision-tree classification of phyto-phenologicalmeasures proved to be more accurate than straight classification of monthly composited NDVI valuesalone. This method of classification is probably more reliably repeatable because it describes andcharacterises land cover types according to their radiometric behaviour through time, rather than theforcing a correspondence to a classification scheme developed on some other bases. However, it hasnot achieved widespread acceptance because the final classes do not correspond to the well-established land cover classifications.

It has also been suggested that the multi-temporal examination surface temperature together with aspectral index may increase the discrimination of regional land cover classes (e.g. Running et. al.,1994). Generally, surface temperature is observed to be inversely proportional to the amount ofvegetation canopy due to a variety of factors including latent heat transfer through evapotranspirationand the lower heat capacity and thermal inertia of vegetation compared with that of soil (Nemani et.al., 1993). Multitemporal NDVI and surface temperature have been used together to produce landcover classifications of the conterminous U.S. (Nemani and Running, 1997), a forest/non-forest

16

classification of Europe (Roy, 1997) and land cover change maps of Africa at 8km resolution(Lambin and Ehrlich, 1996)

4.5 Regression Analysis and Mixture Modelling

Regression analyses and mixture-modelling approaches may assign several classes in differentproportions to a single pixel. The results of regression analysis or mixture modelling are usually a setof images showing the concentrations of a different class component (called endmembers) in eachimage. Although they do not provide any more detailed about the location of forest/non-forestboundaries and sub-pixel forest locations, the overall accuracy of total forest cover estimation is oftenimproved. The mixture modelling approaches assume some form of scene reflectance model (implicitor explicit) that describes the mixing relationship between the different scene components. Iverson et.al. (1989) used an implicit model based on forest/non-forest estimates made from a sample ofoverlapping Landsat TM and NOAA AVHRR images. The percentage forest cover estimates madefrom TM classifications were related to the red and near-infrared AVHRR pixel values by linearmultiple regression. Then the resulting regression coefficients were used to estimate the percentageforest area over the rest of the AVHRR imagery. This approach has been used to make regional forestcover maps of the U.S. (e.g. Zhu and Evans, 1992; Iverson et. al., 1994; Ripple, 1994), of centralAfrica (DeFries et. al., 1997) and of the conterminous U.S. (Zhu and Evans, 1994). Regressionanalysis has also proven to be effective for biomass estimation of coniferous forests, but lesssuccessful where the target area includes conifers and broad leaves trees (Hame et. al., 1997).

Cross et. al. (1991) produced tropical forest proportion images by linear mixture modelling fourchannels of an AVHRR sub-image. The endmembers were derived by examination of the principalcomponents analysis of the AVHRR data. The resultant forest and non-forest proportion images werecompared with supervised classifications of co-registered Landsat TM images. Cross et. al. (1992)concluded that 60–70% of an AVHRR pixel must be forested before it is recognised as such. Holbenand Simabukuro (1993) presented a method to extract the reflected component of the mid-infraredAVHRR channel and used it with the red and near-infrared AVHRR channels to generate vegetation,soil and shade proportion images using a linear mixture model.

The attraction of mixture modelling and regression analysis approaches with NOAA AVHRR data isthat they appear to present a more realistic case of what exists on the ground. Over a 1km area, thereis likely to be a mix of vegetation types, and the assignment of just one class (e.g. forest) to the whole1km area seems unrealistic. The assignment of a proportion (e.g. 70% forest) is more satisfying.However, the application of pure mixture modelling techniques is not straightforward. The initialderivation of endmembers is crucial to the accuracy of the mixture modelling approach, but it iscomplex and not well understood. It is generally recognised that for successful mixture modellingmuch more spectral information is required where a complex mix of land cover types exist.

Regression analysis has been used to create an AVHRR-based forest probability map of the Pan-European area (Hame et. al., 1999). However, such techniques have a requirement for a large amountof independent, reliable and geo-coded reference data, and a set of very clean noise-free imagery.

4.6 Forest Monitoring and Change Analysis Considerations

If image classification is reliable and repeatable, then the change in land cover can be identifiedbetween classifications made at different times. However, in nearly all of the reported forest

17

classification exercises using NOAA AVHRR, the objective has been to achieve the classification.Very few studies have looked at or made repeated classifications at different time periods. Onenotable exception is the work of Cross (1991), in which he tested image difference, image division,image regression and edge difference techniques and applied them to AVHRR images of Rondoniaand Mato Grosso from 1984, 1988 and 1990. Large scale changes such as the establishment of new“fazendas” and other ranching and “fish-bone” and other forest clearances for agriculture wereclearly determinable from the examination of the individual channels on single-date AVHRRimagery.

Because of the lack of image classification repeatability tests in the literature, it is very difficult toassess the utility of NOAA AVHRR for forest extent or condition monitoring or change detection.However, it appears that where changes are abrupt, marked and larger than 1km square (e.g.denudations of dense forest and conversion to agriculture or bare soil) they are easily discerniblefrom individual, good quality single-date full-resolution AVHRR images. However, land conversionswhich are more gradual (e.g. degradation or reduction of forest cover percentage) or of a smallerextent would not be easily discernible, even with recourse to multi-temporal image examination. Formonitoring of such areas, some suggestions have been made as to the utility of AVHRR forcircumstantial or qualitative evidence gathering (e.g. GOFC 1999) to identify deforestation“hotspots”, either independently or in conjunction with a multi-level approach using NOAA AVHRRtogether with higher spatial resolution imagery and ground survey for measurement of the nature oramount of the observed change (see section later and, for example, Jeanjean and Achard 1997).

18

5 Case studies

5.1 General

In the last decade, several major initiatives have been launched concerning the use of coarse-resolution data for mapping and monitoring of forest cover extent and condition. Because they haddifferent objectives, time-scales and resources, the legends and methodologies adopted variedconsiderably. In the section we will provide a brief synopsis of each relevant project. In each, we willlist the legend, data sources, methodology and validation procedures adopted. We will also attempt tooutline any major problems or constraints in the use of the results from the various projects.

5.2 IGBP-DISCover: A Global land cover classification at 1km resolution

5.2.1 Purpose

The International Global Biosphere Programme (IGBP) global land cover classification was made tosupport a number of IGBP initiatives. Their stated objective was to provide a global land coverdataset that was more current, of known accuracy and that had higher spatial resolution and greaterinternal consistency than any other existing datasets (Loveland and Belward, 1997). Thus its primaryfocus was on fast delivery of a global product for use in other IGBP initiatives.

5.2.2 Legend

The legend was chosen to be exhaustive, so that every part of the Earth’s surface was assigned to aclass; exclusive, so that classes did not overlap; and structured so that classes were equallyinterpretable with 1km data, higher resolution remotely-sensed imagery, or ground observation. Thecategories were chosen so that they embraced the climate-independence and canopy componentphilosophy presented by Running et. al. (1994) but modified to be compatible with classificationschemes currently used for environmental modelling to provide, where possible, land useimplications and to represent landscape mosaics (Belward 1996). The legend comprises 17 so-calledDISCover classes and these are defined in Table1.

19

Table 1. IGBP DISCover Data Set Land Cover Classification System

CLASS CLASS NAME DESCRIPTION

1 EvergreenNeedleleaf Forests

Lands dominated by trees with a percent canopy cover >60% and heightexceeding 2 meters. Almost all trees remain green all year. Canopy isnever without green foliage.

2 EvergreenBroadleaf Forests

Lands dominated by trees with a percent canopy cover >60% and heightexceeding 2 meters. Almost all trees remain green all year. Canopy isnever without green foliage.

3 DeciduousNeedleleaf Forests

Lands dominated by trees with a percent canopy cover >60% and heightexceeding 2 meters. Consists of seasonal needleleaf tree communities withan annual cycle of leaf-on and leaf-off periods.

4 DeciduousBroadleaf Forests

Lands dominated by trees with a percent canopy cover >60% and heightexceeding 2 meters. Consists of seasonal broadleaf tree communities withan annual cycle of leaf-on and leaf-off periods.

5 Mixed Forests Lands dominated by trees with a percent canopy cover >60% and heightexceeding 2 meters. Consists of tree communities with interspersedmixtures or mosaics of the other four forest cover types. None of the foresttypes exceeds 60% of landscape.

6 Closed Shrublands Lands with woody vegetation less than 2 meters tall and with shrubcanopy cover is >60%. The shrub foliage can be either evergreen ordeciduous.

7 Open Shrublands Lands with woody vegetation less than 2 meters tall and with shrubcanopy cover is between 10-60%. The shrub foliage can be eitherevergreen or deciduous.

8 Woody Savannas Lands with herbaceous and other understorey systems, and with forestcanopy cover between 30-60%.The forest cover height exceeds 2 meters.

9 Savannas Lands with herbaceous and other understorey systems, and with forestcanopy cover between 10-30%.The forest cover height exceeds 2 meters.

10 Grasslands Lands with herbaceous types of cover. Tree and shrub cover is less than10%.

11 PermanentWetlands

Lands with a permanent mixture of water and herbaceous or woodyvegetation that cover extensive areas. The vegetation can be present ineither salt, brackish, or fresh water.

12 Cropland Lands covered with temporary crops followed by harvest and a bare soilperiod (e.g., single and multiple cropping systems. Note that perennialwoody crops will be classified as the appropriate forest or shrub landcover type.

13 Urban and Built-up Land covered by buildings and other man-made structures. Note that thisclass will not be mapped from the AVHRR imagery but will be developedfrom the populated places layer that is part of the Digital Chart of theWorld.

14 Cropland/NaturalVegetation Mosaics

Lands with a mosaic of croplands, forest, shrublands, and grasslands inwhich no one component comprises more than 60% of the landscape.

15 Snow and Ice Lands under snow and/or ice cover throughout the year.16 Barren Lands exposed soil, sand, rocks, or snow and never has more than 10%

vegetated cover during any time of the year.17 Water Bodies Oceans, seas, lakes, reservoirs, and rivers. Can be either fresh or salt water

bodies

20

5.2.3 Data Sources

Under the joint auspices of the Committee on Earth Observations Satellites, the U. S. GeologicalSurvey (USGS), National Aeronautics and Space Administration (NASA), National Oceanic andAtmospheric Administration (NOAA) and the European Space Agency (ESA), a concerted effort waslaunched to capture and process full-resolution daily NOAA AVHRR data for every part of theglobe’s land surface between April 1992 and September 1996 (with a gap between September 1994to February 1995 when there was no operational afternoon-pass NOAA satellite). Over 4.4Terrabytes of 1km resolution AVHRR data from 30 receiving stations were collected, assembled andprocessed into a coherent data set (Eidenshink and Faundeen, 1994). This dataset consists of tenprocessed channels: including the individual calibrated five channel data and the NDVI.

5.2.4 Classification and Mapping Methodology

The IGBP Land Cover Working Group (LCWG) adapted the methods of the USGS EROS DataCenter (EDC) and the University of Nebraska-Lincoln (ULN) (Loveland et. al., 1991; Brown et. al.,1993). The classification stages include:

1. The 10-day NDVI data from April 1992 to March 1993 were converted to monthly maximumvalue composites, which essentially depict “greenness” classes.

2. Artefacts in the composites due to mis-registration, mis-calibration, geometric distortions, andatmospheric effects were identified, and affected data were either excluded form the analysis orre-processed (Zhu and Yang 1996).

3. “Water Bodies” pixels were identified from the AVHRR Channel 2, and the “Snow and Ice” and“Barren Or Sparsely Vegetated” pixels were determined from their low values in a 12-monthmaximum value NDVI composite. “Urban and Built-Up” pixels, typically very difficult toclassify from remotely-sensed data, were separated by reference to the “urban” layer informationfrom the Digital Chart of the World (Danko, 1992). Loveland, et. al., (1999) provides full detailsof the classification methodology.

4. The remaining data were segmented into greenness classes using an unsupervised classificationclustering algorithm described in Kelly and White (1993). This identifies clusters of NDVI withsimilar temporal properties. Clustering was controlled by predetermined parameters for numbersof iterations and classes, and for measures of inter- and intra-cluster distance. The clusters weredefined by channel mean vectors and covariance matrices. For each continent the number ofclusters was based on empirical evaluation of several clustering trials.

5. Each cluster was then assigned a preliminary cover type label. Class labelling involves theinspection of spatial patterns and spectral or multi-temporal statistics of each class, comparingeach class with ancillary data, and making a final decision concerning cover type(s). A minimumof three independent interpreters label each class to avoid interpreter bias (McGwire 1992).Where differences exist, interpreters compare decision histories to arrive at a consensus.

6. Preliminary clusters with two or more cover types are then split into “single category” land coverclasses; at least 60 per cent of the greenness classes typically represent multiple land cover types(Running et. al., 1995). These can be subdivided on the basis of the relation between the mixedgreenness classes and selected ancillary data sets. Elevations, ecoregions, and climate data haveproved to be the most useful for sub-dividing mixed greenness classes (Brown et. al., 1993).Occasionally, clustering of individual AVHRR spectral channels in the IGBP-DIS dataset may beused to refine mixed clusters. Analyst-defined polygons are used as a last resort and call forextremely careful documentation.

21

7. Attributes for each “single category” class are determined. These include descriptive land coverinformation, NDVI and elevation statistics, and class relations to other land cover legends.

8. Related “single category” classes are then grouped under DISCover labels using a convergence ofevidence approach, in which all available documentation, including the “single category” classattributes, image maps, atlases and existing land cover/vegetation maps are compared with thespatial distribution of each DISCover class. Three interpreters are again used to avoid bias.

5.2.5 Validation and Accuracy

Considerable effort has been made to validate the IGBP global land cover classification (known asDISCover 1.0). Initial phases of validation included peer review of the datasets published on theInternet. Subsequent phases of validation using high spatial resolution imagery are reported in aspecial edition of the Journal of Photogrammetric Engineering and Remote Sensing (from whichmost of the following results have been taken).

A stratified random sample design was adopted for identifying validation sites with a goal ofidentifying 25 samples per DISCover class, for which to obtain high spatial resolution images forvalidation. A total of 379 Landsat TM or SPOT images were obtained, processed and interpreted tothe same IGBP legend as used in the global land cover classification. Then at a series of workshops,three regional Expert Image Interpreters independently verified each sample and a majority decisionrule was used to determine sample accuracy. For the 15 DISCover classes validated (excluding Waterand Snow And Ice classes), the average class accuracy was 59.4% with accuracies ranging between40.0% and 100%. The overall area weighted accuracy of the data set was determined to be 66.9%.When only samples which had a majority interpretation for errors as well as for correct classificationwere considered, the average class accuracy of the data set was calculated to be 73.5%.

The highest individual class accuracies were established in Class 16 (Barren; 1.00), Class 2(Evergreen Broadleaf Forests; .840), and Class 7 (Open Shrublands; .778). Classes 2 and 16 met theaccuracy goal established by IGBP for DISCover 1.0 of .85 accuracy (at 95% confidence). Theaccuracies for Class 4 (Deciduous Broadleaf Forests) and Class 9 (Savannas) were the lowest of the15 classes verified. Class 3 (Deciduous Needleleaf Forests) and Class 11 (Permanent Wetlands) alsohad low accuracy, but the number of samples validated for this class was well below the minimum 25samples specified in the validation protocol. Not surprisingly, larger and less fragmented classes hadhigher thematic accuracy (Scepan 1999, Scepan et. al., 1999)

22

Table 2. IGBP DISCover 1.0 Overall Accuracy (from Scepan, 1999)

DISCover Class VerifiedSamples

SamplesVerifiedCorrect

User’sAccuracy

ConfidenceInterval

(.95)PercentCover

OverallDISCoverAccuracy

1 - EvergreenNeedleleaf Forests

26 15 .58 0.38 - 0.77 .0482 .669

2 - EvergreenBroadleaf Forests

25 21 .84 0.69 - 0.99 .0916

3 - DeciduousNeedleleaf Forests

11 5 .46 0.15 - 0.76 .0123

4 - DeciduousBroadleaf Forests

25 10 .40 0.20 - 0.60 .0284

5 - Mixed Forests 27 15 .56 0.36 - 0.75 .04716 - ClosedShrublands

27 15 .56 0.36 - 0.75 .0198

7 - Open Shrublands 27 21 .78 0.62 - 0.94 .14898 - Woody Savannas 31 18 .58 0.40 - 0.76 .07509 - Savannas 26 11 .42 0.23 - 0.62 .075510 - Grasslands 26 15 .58 0.39 - 0.77 .083011 - PermanentWetlands

17 5 .29 0.07 - 0.52 .0075

12 - Cropland 28 18 .64 0.46 - 0.82 .102813 - Urban andBuilt-up

30 16 .53 0.35 - 0.72 .0032

14 - Cropland/Natural VegetationMosaics

26 13 .50 0.30 - 0.70 .1114

16 - Barren 27 27 1.00 0.87 - 1.00 .1452

5.2.6 Coverage, Resolution and Availability

The IGBP land cover classification covers the whole of the Earth’s land surface. It is at a nominalspatial resolution of 1km, but because it is based on multi-temporal composites its geometricaccuracy is more likely to be closer to 2 – 5km. The intermediate and final datasets are madeavailable on the Internet for download and/or validation on:http://edcwww.cr.usgs.goc/landdaac/http://www.cnrm.meteo.fr:800/igbp/dis_home/html

5.2.7 Problems Or Constraints in Use

Because the IGBP land cover classification is based on multitemporal composites, the results aresubject to all the image compositing errors (e.g. the geometric accuracy is probably of the order of 2– 5 km). Also, because the classification is based on an unsupervised classification of a whole year ofmonthly maximum value composites of NDVI, it is very difficult to repeat or amend theclassification for any change detection or monitoring purposes. The assignment of classes to theautomatically-determined clusters relies heavily on ancillary datasets, so these too must be kept up todate and validated themselves. Although there was an attempt to remove interpreter subjectivity by

23

using at least three expert interpreters would the same ones have to be employed for subsequentreclassifications?

5.3 FAO FRA 2000 Global Forest Cover Map

5.3.1 Purpose

The global FRA 2000 global forest cover mapping effort builds on the full IGBP/USGS seasonaldatabase, refining forest classes to reflect forest density classification. Its main objectives are to:

1) develop a general, actual forest cover map linking to the IGBP/USGS land cover database;2) map broad forest cover categories that can be consistently extracted from AVHRR 1 km

composite data; and3) use the map to support the FAO year 2000 global forest survey.

As a complete enumeration of forest cover in the world, it may be used to supplement regions lackingrecent, reliable forest inventory (FAO, 1995). As a new addition to the USGS land cover database,the effort should help improve applications of the database when information about forest cover isneeded.

5.3.2 Legend

Six land cover categories are initially targeted:

1. closed forest,2. open forest,3. fragmented forest,4. woody savannah,5. other land cover, and6. water.

These categories follow, to the extent possible, canopy density definitions that have been suggestedby FAO Forestry as required for global assessment using remotely sensed data (1995). The first threecategories are considered to be forested land cover.

5.3.3 Data Sources

The same data as used for the IGBP global land cover classification are used. Note that the sourcedata used for the IGBP land cover classification were data acquired in the period from March 1992 toApril 1993, thus in order for the global FAO forest cover map to be used consistently for FRA 2000purposes, the global land cover map may have to be updated with source data from a later period inthe 1990s.


As described in the previous section, vegetation classification in the IGBP land cover database arebuilt on characteristics of vegetation seasonality determined in terms of AVHRR NDVI. Themagnitude of integrated NDVI over the length of the temporal period helped separate successivelydecreasing vegetation biomass, from dense forestlands to sparse land cover. On the other hand,

24

seasonal variations were investigated to partially support identification of vegetation physiognomy(e.g. separating deciduous from evergreen forests).

Because the IGBP seasonal land cover database was not intended to optimise for forest cover, nodirect relationship exits to enable a simple conversion of the seasonal land cover classes to the sixFAO classes. Rather, a two-step methodology has been implemented which allows certaininteractive flexibility in deriving and correcting for the FAO classes (Zhu et. al., 1999).

� Adapting the IGBP seasonal land cover classes to the FAO classification. Zhu et. al. (1999)report that the full IGBP seasonal land cover classes are used as the baseline data, on the continent-by-continent basis (presumably the unsupervised classes). Then, refinement methods similar to thoseused in producing the IGBP classes are used to fit the derived unsupervised classes to FAOdefinitions. These refinement methods depend on local conditions of land cover and rely on a carefulstudy of all available evidence. The country-level forest database maintained by FAO is also used asa general reference for country level forest classification. � Estimating percent forest cover using two techniques. The first is what Zhu et. al. (1999) refer toas spectral unmixing, apparently applied to the bright pixels in AVHRR bands 2 (infrared) and 1(visible) spectral space, which were found to be primarily mixtures of forest, cropland, and bare soilhaving high reflectance in these bands. For the dark pixels in AVHRR bands 1 and 2 a scaled NDVIwas found to be a more representative indicator or amount of forest. Negative relationships ofamount to NDVI were found for closed forests, open forest and woody savannah areas. It is not clearin Zhu et. al. (1999) whether the spectral unmixing and scaled NDVI approach was applied tomonthly or the initial 10-day composites.

To provide the least atmospherically affected result, final percent forest cover was determined overthe course of the year based on the maximum monthly forest cover value achieved, regardless of themethods chosen (mixture analysis or scaled NDVI). In addition to the derived forest cover map,results also include look-up tables linking the USGS seasonal land cover classes to the six-categorymap legend. Comparisons with the FAO forest resource information system are also made at thecountry level to guide the mapping work. However, it was not the objective of this effort to calculateforest area statistics based on the digital map, recognizing that, with limited spatial and spectralresolution, the data are best suited for showing where major forested areas are at small scales, ratherthan quantity of forests.


Most of the seasonal land cover classes from the IGBP database were translated well into the FAOcategories. For South America, for example, only 34 out of 167 seasonal classes needed furtherprocessing under the second step. Accuracy assessment has been conducted for the IGBP land coverdatabase at the IGBP 17-class level based on a set of Landsat and SPOT scenes, as reported in theprevious section. Validation of the FAO forest cover map is believed to be forthcoming, and shouldbe of a higher accuracy since there are fewer classes used in the separation (this author’s opiniononly)

The IGBP global land cover database is a convenient mapping framework, from which a new, anddifferent, global forest cover map is derived. Factors affecting quality of the derived global forestcover map include misclassification errors from the IGBP product, errors of translation to the FAOlegend, data quality problems, and accuracy of the linear unmixing approach.

25


Presumably as with the IGBP land cover classification.

5.3.7 Problems Or Constraints In Use

Since the FAO global forest cover map is derived from the same data and largely with the samemethodology as employed in the IGBP land cover classification, the caveats associated with that dataand methodology also apply here.

5.4 TREES (Tropical Ecosystem Environment observation by Satellites)

5.4.1 Purpose

The TREES project has several objectives and goals, including:

- The compilation of a comprehensive pan-tropical forest map at about 1: 1million scale (or1km resolution);

- The study of seasonality effects in tropical forests as evidenced in relatively long-term seriesof satellite-derived information;

- The modelling of deforestation processes and dynamics;- The investigation of the usefulness of new sensors for topical forest mapping and monitoring,

in particular the potential of Synthetic Aperture Radar (SAR) and the Along-Track ScanningRadiometer (ATSR) flown onboard the ERS-1 satellite.

Of most interest in this context is the global tropical forest inventory from NOAA AVHRR at anominal spatial resolution of 1km.

5.4.2 Legend

After detailed reviews and preliminary examinations of spatial, spectral and temporal characteristicsof NOAA AVHRR imagery over tropical forest areas, it was felt that a meaningful, appropriate anduseful legend for the pan-tropical map should be based essentially on two parameters that were felt tobe determinable with a high degree of reliability for all tropical areas: the amount of forest coverwithin 1km pixels, and the phonological character of the forest area overall. Based on these twocriteria, the following five classes, on which the global TREES legend is based, were identified:

- Intact evergreen forest; to include areas of more than 70% forest cover mainly of an evergreennature;

- Fragmented evergreen forest; to include areas of between 40% and 70% forest cover, mainlyof an evergreen nature;

- Intact seasonal forest; to include areas of more than 70% forest cover mainly of a seasonalnature;

- Fragmented seasonal forest; to include areas of between 40% and 70% forest cover, mainly ofa seasonal nature;

- Non-Forest; to include all other areas of less than 40% forest cover.

Fore some areas, under certain conditions, more detailed classes can sometimes be determined (e.g.mixed forest/savannah in Africa, mixed deciduous or semi-evergreen and dry dipterocarp forests in

26

continental South-EastAsia, etc.). For continental-level maps published at 1:5million scale, thelegend above is merged with other vegetation maps to provide more detailed classes (see Eva et. al.,1999; Mayaux et. al., 1999 and Achard et. al., 1999).

5.4.3 Data Sources

In the TREES project, the selection, processing and subsequent classification of the AVHRR imagerywas considered in a convenient number of “image windows”, which provided a crude first levelstratification based on consideration of the prevailing ecological and climatological conditions whichultimately control the type of existing vegetation and likely acquisition of good quality imagery.

In all areas, channels 1 – 4 of the full spatial resolution NOAA AVHRR imagery were used. About300 single-date images for South-East Asia, 150 – 200 for Central and West Africa and over 150images for South and Central America were used in the first round of classifications. Considerabletime was spent at local receiving stations examining archives and only selecting images which hadthe least cloud and atmospheric contamination. Where possible images that were as close to nadir aspossible, and which allowed a representative sample of data over dry seasons where these occurredwere chosen in priority (Malingreau et. al., 1996). Most of the data were collected during 1992 –1994.


In most cases individual images or parts of images were classified using 3 – 4 channels of data andNDVI in an unsupervised approach. The resulting spectral classes were then assigned to a class onthe legend after detailed comparison with ancillary data such as (i) high spatial resolution imagery;(ii) published vegetation maps; (iii) ecoregions and other published information, held in thesupporting Tropical Forest Information System (TFIS). In areas of seasonal forest formations, multi-temporal classification was used.


The TREES project has adopted a multi-scale, multi-sensor approach for validation of the AVHRR-derived classification, and subsequent calibration of area measurements therein. For this purpose, anumber of classifications based on higher spatial resolution imagery (Landsat TM or SPOT) werecommissioned at a sample number of locations. The process of validation as defined in the TREESproject, involved qualitative comparison of the class assignments made on the AVHRR-derivedclassification, with the classifications made from high spatial resolution imagery, which in turn wasnormally assisted with fieldwork. This qualitative comparison assisted in the improved labelling ofautomatically derived AVHRR spectral classes.

Once the AVHRR classification was completed, a more rigorous quantitative comparison andsubsequent correction (called calibration) of forest area measurements was made. This involves thecomparison of AVHRR-based and TM- or SPOT-based forest/non-forest results over a number of 15x 15km areas evenly distributed over the high spatial resolution image extent (see section 6).


TREES results cover most of the pan-tropical area (except for some areas where tropical forest wasnot thought to be of major interest, e.g. the Indian sub-continent, China, East Africa). The results areavailable at 1km resolution, and because the extents of the assigned classes was made on

27

interactively-processed single-date imagery, the geometric accuracy is likely to be within 1 – 2km.Continental maps have been published at a scale of 1: 5 million for Latin America (Eva et. al., 1999);West and Central Africa (Mayaux et. al., 1999) and South-East Asia (Achard et. al., 1999) and thesehave been widely disseminated. A CD-Rom outlining the project and some of its results has also beenwidely disseminated. More details are available on:http://www.trees.gvm.sai.jrc.it/

5.4.7 Problems Or Constraints In use

The TREES project has only compiled data for the pan-tropical area, and within that has alsoexcluded some areas such as India, China and East Africa. Its chosen legend also differs from otherprojects. Finally, the data used were mostly from 1992 – 1995, so the maps may need updating.

5.5 FIRS – Forest Information from Remote Sensing: A forest information system for the entire European continent

5.5.1 Purpose

The FIRS project has a number of objectives and foundation actions (Kennedy and Folving, 1994).Of most interest here are the objectives:

• To produce a standardised European Forest Map at a scale of 1: 1 million to provide a generaloverview for small scale monitoring

• To develop methods and models for mapping and monitoring the dynamics of forested areasusing NOAA AVHRR data.

5.5.2 Legend

A simple forest / non-forest legend was used for the initial phases of the work (Roy 1997). In asubsequent phase, an assignment of the probability of forest cover was also assigned to each 1kmpixel (Hame et. al. 1999).

5.5.3 Data Sources

For the first map, maximum value monthly composite NDVI and surface temperature images wereused for an 8-month period (March – October 1993). The probability map was based on a compositemade from 49 individual images acquired between June and September 1996.


In the derivation of the first map, AVHRR data in each of 82 forest strata were classified separately.Where sufficient cloud-free data were available, all 8 NDVI and surface temperature data sampleswere used. Otherwise the best 3 months were selected. In cases where cloud cover proved to be verypersistent, the nest NDVI and surface temperature image was used independently. Training andverification sample pixels were chosen using various reference data (Roy 1997) and maximumlikelihood classification was used.

In the derivation of the probability-based map, an unsupervised clustering program (Hame et. al.1998) was run to automatically select training samples consisting of 2 x 2 pixels representing

28

homogeneous ground targets in the red /near-infrared mosaic, after water pixels were removed. Then,a maximum likelihood classification method with equal a priori probabilities was used to purify the 2x 2 pixel blocks in each cluster. For each remaining 2 x 2 pixel block of each cluster, the probabilityof forest was computed by reference to the CORINE land cover database (Kennedy et. al., 1999).Thus the final value of forest cover assigned to each cluster was obtained by calculating the mean ofthe values from its sample components.


For the first forest/non-forest map accuracy assessment was made by comparison with reported forestpercentages by EUROSTAT (Roy 1997). Small scale comparison of the forest map withindependently collected forest statistics gave a very high level of agreement (R-squared 0.88 to 0.91)for France and Germany when forest map misclassification biases were taken into account.

The application of the AVHRR-based probability mapping method, and the accuracy of the derivedinformation is clearly strongly dependent on the quality and reliability of ground data used as well asthe cleanness of the satellite data (Kennedy et. al. 1999). On quantitative evaluation of the results, itwas apparent that the forest area was generally underestimated, especially in France and theMediterranean region (Hame et. al. 2000).

5.6 Other Forest Cover Maps

As well as the major projects listed above, there are a few other notable efforts in vegetation, and inparticular, forest cover mapping using coarse spatial resolution data. These include:

• Vegetation of South America, by Stone et. al. (1994);• Forest cover mapping for Mexico, by Eggen-McIntosh et. al. (1992);• AVHRR-based forest mapping for West Africa , by UNEP/GRID (1991);

5.7 Summary

A number of projects that sought to utilise coarse spatial resolution imagery (mainly NOAAAVHRR) for forest mapping and monitoring have been reviewed. Earlier efforts consisted of closeexamination and interactive identification of forest boundaries from single-date imagery over quitesmall areas. As computing power and the size of datasets increased, increasingly automated methodsover large areas have been implemented.

In most cases, “acceptable” maps of forest extent have been produced as scales of 1: 1 million to 1: 5million. On qualitative comparison, these seem to depict the extent of the forest biomes quite well,and in many cases, better than any other existing dataset. They certainly seem useful for stratificationof forest areas, perhaps to identify areas where more detailed investigation is necessary. However,because the different results were obtained with very different legends and for different purposes, it isvery difficult to make any quantitative comparisons.

Accuracy assessment has mostly consisted of peer review, comparison with other data sources, andcomparison with classification of higher spatial resolution imagery from a sample set of sites. Ingeneral, all show that forest cover estimates derived from AVHRR are underestimated if the forest is

29

fragmented and generally overestimated if the forest cover is homogenous and uniform over largeareas (D’Souza et. al., 1996; Kuusela and Paivinen, 1995).

Repeat classifications using the same methodology have rarely been tried, so it is difficult to assessthe repeatability of each classification. Therefore we cannot say much about the usefulness of thereviewed methods for forest extent or condition monitoring/change detection. However, given thedifficulties in geometric, spectral, classification methodologies, etc. probably only large and markedchanges (of about 3 – 5km, and of considerable spectral transformation) would be detectable on anoperational basis. Most operational monitoring projects (e.g. TREES-II) have gone to a phase of“hot-spot” detection, where apparent changes in AVHRR reflectance response are combined withother evidence such as the occurrence of forest fire (Achard et. al. 1998).

30

6 Combining high and low spatial resolution classificationsAs seen from the reviews above, NOAA AVHRR data have been used operationally for providingglobal scale land-cover maps. However, the estimation of land-cover proportions from these mapswill be associated with a systematic bias due to spatial aggregation effects, especially in areas wherethe amount of spatial fragmentation is high (Nelson 1989; Gervin et. al., 1985). For this reason, anumber of researchers have indicated that NOAA AVHRR data should be used in conjunction withdata from higher spatial resolution data for mapping and monitoring purposes (e.g. Nelson andHolben 1986; Van Roessel 1990; Jeanjean and Achard 1997)

We have already discussed the use of higher spatial resolution data for calibrating and validatingvarious NOAA AVHRR sub-pixel classifications (see section 4.5). In this section we review themethods suggested for the use of image classifications from high spatial resolution data to improvethe results of the classifications obtained at coarse spatial resolution. A few studies have applied asimple “regression estimator” approach. This involves relating the fine scale forest cover directly tothe coarse scale forest cover (Nelson 1989; Stone and Schlesinger 1990; Amaral 1992; Stibig 1993).Forest/non-forest proportions are determined over blocks of AVHRR-size pixels, and related to theforest/non-forest proportions for the corresponding areas on the fine scale classifications. A simplelinear regression is sought and then applied inversely to the whole of the AVHRR classification tocorrect for the estimation bias. However, the regression parameters depend heavily on the type oflandscape analysed. When calculating the simple regression over large areas, ignoring the effect ofspatial organisation leads to a large instability of the results (Nelson 1989).

In more recent work, Mayaux and Lambin (1995) showed that different regressions were obtainedbetween Landsat TM-based and AVHRR-based forest percentages when the spatial pattern of theAVHRR-based forest/non-forest classification was taken into account. Mayaux and Lambin (1995)used AVHRR forest/non-forest classifications and Landsat TM classifications from 13 sites withinthe TREES project area. They aggregated forest proportion amounts in 13 x 13 km blocks, and used asimple spatial index (the Matheron Index as defined by Kleinn et. al., 1993) in a two-step correctionfunction. The various stages of the two-step correction process were to:

− Geometrically co-register fine and coarse resolution data− Overlay of a grid of contiguous pixel blocks. The block size selected is dependent on the

accuracy of co-registration. Kleinn et. al. (1993) demonstrated in simulation studies that,assuming a misregistration of Landsat TM and AVHRR data of around two pixels, theinfluence of misregistration on the calculation of proportional errors for block sizes of 13 x 13AVHRR pixels becomes negligible. If better registration can be achieved, smaller blocks maybe used.

− Measure in each pixel block: (i) the forest cover at fine resolution; (ii) the forest cover atcoarse resolution; (iii) the spatial pattern (e.g. Matheron Index) at coarse resolution

− Partition the population of pixel blocks in equal-sized subsets of similar spatial patterns− Make First step regressions: Linear regressions within each subset between forest cover

proportions at fine and coarse resolution. The observations are the pixel blocks.− Make Second step regressions: Regressions between the spatial measure and (i) the intercept

and (ii) the slope of the first-step regression. The observations are the subsets of the blocks.

Mayaux and Lambin (1995) were able to conclude that the integration of spatial information in acorrection model significantly improves the retrieval of proportions of cover types from coarseresolution data compared with a simple correction function relating directly proportions at coarse and

31

fine scale resolutions. This was true even is a simple spatial measure such as the Matheron Index wasapplied to the coarse resolution classification.

In a subsequent study, Mayaux and Lambin (1997) improved the two-step correction model byincluding textural measures derived from the coarse resolution data (e.g. minimum local variance for3 x 3 AVHRR channel 2 pixels) instead of the Matheron Index. In a detailed study that tested anumber of hypotheses they were able to show that:

− Area estimates of land-cover categories should not be extracted directly from coarse spatialresolution classifications because of the biases inherent in assigning a single class to a largepixel area;

− The correction of broad scale maps for the proportional errors introduced by spatial scalingdrastically improves the reliability of the estimation of forest-cover areas;

− If only AVHRR data are available, the two-step procedures integrating around pixel texturewith proportions estimated from a coarse resolution classification gave an equivalentperformance to a mixed pixel regression-based estimator. The mixed pixel estimator,however, did not require the aggregation to larger block sizes (of say 9 x 9) but wasdependent on the application of a suitable bulk correction for atmospheric effects;

− The two-step calibration procedure required a very large sample of high spatial resolutiondata since it depended on two levels of nested regressions;

− A more advanced concept of a land-cover map should consist of a multilayered databaserepresenting, for every layer, the calibrated proportion of a given land-cover class in a coarseresolution sampling unit. Or, pixel’s attributes should be vectors of within-pixel land-coverproportions rather than single land-cover categories.

32

7 BibliographyAchard, F. and Blasco, F., 1990. Analysis of vegetation seasonal evolution and mapping of forest

cover in West Africa with the use of NOAA AVHRR HRPT data. PhotogrammetricEngineering and Remote Sensing, 10: 1359-1365.

Amaral, S., 1991a. Deforestation estimates in AVHRR-NOAA and TM-Landsat images for a regionin Central Brazil. In Proceedings of the XVII Congress of the International Society ofPhotogrammetry and Remote Sensing, Washington, DC, August 2-14, 1992.

Amaral, S., 1991b. Imagens do sistema sensor AVHRR-NOAA na deteccao e avaliacao dedesmatamentos na floresta Amazonica- relacoes com dados do sistema TM-Landsat. MScThesis, INPE, Sao Jose dos Campos, Brazil, 166 pp.

Batista, G.T. and Tucker, C.J., 1992. Assessment of AVHRR data for deforestation estimation inMato Grosso (Amazon Basin). In Proceedings of the XVII Congress of the InternationalSociety of Photogrammetry and Remote Sensing, Washington, DC, August 2-14, 1992.

Belward, A.S., 1991. Remote sensing for vegetation monitoring on regional and global scales. In :Remote Sensing and GIS for Resource Management in Developing Countries (Belward,A.S.and Valenzuela,C.R. eds.), Kluwer Academic Publishers, pp. 169-188.

Belward, A.S. and Lambin, E., 1990. Limitations to the identification of spatial structures fromAVHRR data. International Journal of Remote Sensing, 11: 921-927.

Belward, A.S., Vogt ,J.V., Falk-Langermann, A., Saradeth, S. and Cambridge, H., 1992.Preparationof AVHRR GAC data sets for global change studies. In Proceedings of theEuropean 'International Space Year' Conference held at Munich, Germany, 30 March-4 April1992, ESA ISY-1, pp.19-24.

Belward, A., Ed. The IGBP-DIS Global 1Km Land Cover Data Set “DISCover”: Proposal andImplementation Plans. 1996. IGBP-DIS Working Paper 13. International Geosphere BiosphereProgramme. European Commission Joint Research Center, Ispra, Varese, Italy.

Bird, R.E., 1984. A simple, solar spectral model for direct-normal and diffuse horizontal irradiance.Solar Energy, 32: 461-471.

Beard, J.S., 1955. The classification of tropical american vegetation types. Ecology, 36, 89-101.

Breaker, L.C., 1990. Estimating and removing sensor-induced correlation from advanced very highresolution radiometer data. Journal of Geophysical Research, 95: 9701-9711.

Borella, H.M., J.E. Estes, C.E. Ezra, J. Scepan, and L. Tinney, 1982, Image Analysis for FacilitySiting: A Comparison of Low-and High-Altitude Image Interpretability for Landuse/LandcoverMmapping. United States Nuclear Regulatory Commission Publication, NUREG/CR-2861,S-744-R; 61p.

Box, E.O, Holben, B.N. and Kalb,V., 1989. Accuracy of the AVHRR vegetation index as a predictorof biomass, primary productivity and net CO2 flux. Vegetatio, 80: 71-89.

33

Cihlar, J., Manak, D., Dílorio, M., 1994. Evaluation of compositing algorithms for AVHRR data overland, IEEE Transactions on Geoscience and Remote Sensing, 32: 427-437.

Cochran, W.G., 1977. Sampling Techniques. Wiley Series in Probability and MathematicalStatistics. New York, NY.

Congalton,. R.G., 1991, A review of assessing the accuracy of classifications of remotely sensed data,Remote Sensing Environment , 37:35-46.

Congalton, R.G. and Kass Green, 1999. Assessing the Accuracy of Remotely Sensed Data:Principals and Practices. Lewis Publishers, New York, NY.

Conservation International, 1991. Biodiversity at Risk:Atlas and map. Consrvation International,Washington, DC, USA.

Cross, A., 1990. Tropical Deforestation and Remote Sensing: The use of NOAA/AVHRR data overAmazonia. Final report to the Commission of the European Communities (B946/88), Geneva,60 pp.

Cross, A., 1991a. NOAA/AVHRR as a data source for a tropical forest GIS: Deforestation inAmazonia.In Proceedings of IGARSS-90, held in Washington D.C.,U.S.A., May 21-25 1990,223-226.

Cross, A., 1991b. Tropical Forest Monitoring using AVHRR: Towards an automated system forchange detection. Final Report to UNEP-GRID, Geneva, 44pp.

Cross, A., Settle, J.J., Drake, N.A. and Paivinen,R.T.M., 1991. Subpixel measurement of tropicalforest cover using AVHRR data. International Journal of Remote Sensing, 12, 1119-1130.

Czaplewski, R.L., 1992, Missclassification bias in areal estimates, Photogrammetric Engineering andRemote Sensing , 58:189-192.

DeFries, R., and J.R.G. Townshend. 1994. NDVI-derived Land Cover Classifications at a GlobalScale. International Journal of Remote Sensing. Vol. 15. No.7. pp. 3443-3462.

DeFries, R.S., and S.O. Los, 1999. Implications of Land Cover Misclassification Parameter Estimatesin Global Land Surface Models: An Example from the Simple Biosphere Model (SiB2).Photogrammetric Engineering and Remote Sensing (special issue).

DeFries, R. S. and J. R. G. Townshend, 1994, NDVI-derived land cover classification at a globalscale, International Journal of Remote Sensing, 15: 3567-3586.

EEA Task Force. 1992. CORINE Land Cover. Brochure prepared as contribution to EuropeanConference of the International Space Year. Munich. 30 March - 4 April, 1992. 22p.

EEC Council Regulation 1615/89. Official Journal of the European Communities, 15.6.89. No. L.165/12-13.

EEC Council Regulation 2080/92. Official Journal of the European Communities, 30.7.92. No. L.215/96-99.

34

Eggen-McIntosh,S.E., Munoz,E.B, Ornelas de Anda, J.L, and others, 1992. Forest Cover Mapping ofMexici using Advanced Very High Resolution Radiometer Imagery. In Proceedings, Mappingand Monitoring global change, ASPRS/ACSM/RT’92 technical papers, vol 5, pp 410-416, Aug3 – 8 1992, Washington D.C.

Eidenshink, J.C. and J.L. Faundeen, 1994. The 1km AVHRR Global Land Cover Data Set: FirstStages in Implementation. International Journal of Remote Sensing. Vol. 15. No.7. pp. 3567-3586.

Emery, W.J., Brown, J. and Nowak, Z.P., 1989. AVHRR image navigation: summary and review.Photogrammetric Engineering and Remote Sensing, 55: 1175-1183.

ESA. 1993. ISY Forest Map of Europe Based on NOAA-AVHRR. Report No.9730/91/NL/FG.ESA/ESTEC. Noordwijk, The Netherlands.

Estes, J.E., J. Scepan, and L. Ritter, 1987. Evaluation of Low Altitude Remote Sensing Techniquesfor Obtaining Environmental Information. U.S. Nuclear Regulatory Commission PublicationNUREG/CR-3583; S-762-R.

Estes, J.E. and W. Mooneyhan, 1994, Of Maps and Myths, Photogrammetric Engineering andRemote Sensing, Vol. 60, No. 5, pp. 517-524.

European Commission, 1995. Regionalisation and Stratification of European Forest Ecosystems.Joint Research Centre of the European Commission, Institute for Remote Sensing Applications,Environmental Mapping and Modelling Unit. internal Publication, No.S.P.I.95.44.

Fearnside, P.M., 1991. Deforestation in Brazilian Amazon. In : The Earth in Transition: patterns andprocesses of biotic impoverishment (Woodwell,G.M. Ed.), Cambridge University Press, pp.211-238.

Finnish Ministry of Agriculture and Forestry. 1993. The Resolutions and the General Declarations ofthe Second Ministerial Conference on the Protection of Forests in Europe. MinisterialConference on the Protection of Forests in Europe, 16-17 June, 1993. Helsinki.

Fitzpatrick-Lins, Katherine. The Accuracy of Selected Land Use and Land Cover Maps at Scales of1:250,000 and 1:100,000. Geological Survey Circular 829, 1980.

Folving, S., G. Ertner and T. B. Svendsen (eds.). 1992. European Collaborative ProgrammeWorkshop on Remote Sensing for Forestry Applications. Copenhagen, Denmark. 13-15November 1991. Brussels - Luxembourg: CEC- EUR 144445 EN.

Fusco, L., Muirhead, K. and Tobiss, G., 1989. Earthnet's coordination scheme for AVHRR data.International Journal of Remote Sensing, 10: 635-636.

Fung, I.Y., Tucker, C.J. and Prentice, K.C., 1987. Application of advanced very high resolutionradiometer vegetation index to study atmosphere-biosphere exchange of CO2. Journal ofGeophysical Research, 92: 2999-3015.

Gervin, J.C., Kerber, A.G., Witt, R.G., Lu, Y.C. and Sekhon,R., 1985. Comparison of level I landcover classification accuracy for MSS and AVHRR data. International Journal of RemoteSensing, 6: 47-57.

35

Goward, S.N. and Dye, D.G., 1987. Evaluating North American net primary productivity withsatellite observations. Advances in Space Research, 7: 165-174.

Goward, S.N., Tucker, C.J. and Dye, D.G., 1985. North American vegetation patterns observed withthe NOAA-7 advanced very high resolution radiometer. Vegetatio, 64: 3-14.

Goward, S.N., Kerber, A., Dye, D.G and Kalb, V., 1987. Comparison of North and South Americanbiomes from AVHRR observations. Geocarto, 2: 27-40.

Goward, S.N., Markham,B., Dye, D.G., Dulaney, W. and Yang, J., 1991. Normalized differencevegetation index measurements from advanced very high resolution radiometer.Remote Sensingof the Environment, 35: 259-279.

Goward, S.N., Dye,G., Turner, S. and Yang, J., 1992. Objective assessment of the NOAA GlobalVegetation Index data product. International Journal of Remote Sensing,, (in press).

Gregoire, J.M., Flasse, S. and Malingreau, J.P., 1988. Evaluation de l'action des feux de brousse, denovembre 1987 a fevrier 1988, dans la region frontaliere Guinee-Sierra Leone. Exploitationdes images NOAA-AVHRR. Report CCR SPI 88.39. CEC-JRC, Ispra, Italy.

Gutman, G.,1987. The derivation of vegetation indices from AVHRR data. International Journal ofRemote Sensing, 8: 1235-1243.

Gutman, G.,1991. Vegetation indices from AVHRR: An update and future prospects. Remote Sensingof the Environment, 35: 121-136.

Häusler, T., S. Saradeth and Y. Amitai. 1993. NOAA-AVHRR Forest Map of Europe. In:Proceedings of International Symposium Operationalization of Remote Sensing. 19-23 April1993, Enschede, The Netherlands. Eds. J.L. van Genderen, R.A. van Zuidam and C. Pohl, Vol.8., pp. 37-48. Enschede, ITC.

Holben, B.N., 1986. Characteristics of maximum-value composite images from temporal AVHRRdata, International Journal of Remote Sensing, 7:1417-1434.

Holben, B.N., Kaufman,Y.J. and Kendall, J.D., 1990. NOAA-11 AVHRR visible and near-IR inflightcalibration. International Journal of Remote Sensing, 11: 1511-1519.

I.B.G.E., 1988. Instituto Brasiliero de Geografia e Estatistica Mapa de Vegetacao do Brasil. 1:5mscale map, Rio de Janeiro, Brazil.

Husak, G.J., B.C. Hadley, and K.C. McGwire, 1999. Registration Accuracy of High-ResolutionSatellite Data Used in the IGBP Validation. Photogrammetric Engineering and Remote Sensing(this issue).

IUFRO, 1994. IUFRO World Series Vol. 5. IUFRO Guidelines for Forest Monitoring. R. P”ivinen,H.G. Lund, Simo Poso and T. Zawila-Niedzwiecki (Eds). IUFRO Secretariat, Vienna, 1994.

Justice, C.O., Townshend, J.R., Holben, B.N., and Tucker, C.O., 1985. Analysis of the phenology ofglobal vegetation using meteorological satellite data, International Journal of Remote Sensing,6: 1271-1381.

36

Justice, C.O., 1992. Satellite monitoring of tropical forests: a commentary of current status andinstitutional roles. In Proceedings International Space Year-World Forest Watch Conference,held in May 27-29 1992, Sao Jose dos Campos, Brazil, EUR 14561 EN, 19-33.

Justice, C.O., Townshend, J.R.G., Holben, B.N. and Tucker, C.J., 1985. Analysis of the phenology ofglobal vegetation using meteorological satellite data. International Journal of Remote Sensing,6: 1271-1281.

Justice, C.O., Dugdale, G., Townshend, J.R.G., Narracott, A. and Kumar, N., 1991. Synergismbetween NOAA-AVHRR and Metoesat data for studying vegetation in semi-arid West Africa.International Journal of Remote Sensing, 12: 1349-1368.

Kelly, M., J. E. Estes and K.A. Knight. 1999, Image Interpretation Keys for Validation of GlobalLand Cover Data Sets. Photogrammetric Engineering and Remote Sensing (this issue).

Kennedy, P.J., Paivinen, R., and Roihuvuo, L., 1995. Proceedings to Workshop on Designing aSystem of Nomenclature for European Forest Mapping, Joensuu, Finland, 13-15 June, 1994.Published by CEC, Luxembourg, 1995.

Kennedy, P.J., S. Folving, and N. McCormick. 1995. European Forest Ecosystems Mapping andForest Statistics: The FIRS Project. In: Proceedings of Workshop on Designing a System ofNomenclature for European Forest Mapping, Joensuu, Finland, 13-15 June, 1994. (Edited by P.Kennedy, R. Päivinen and L. Roihuvuo), Published by CEC, Luxembourg, 1995.

Kerber, A.G., 1983. Utility of the Advanced Very High Resolution Radiometer (AVHRR) middleinfrared and thermal channels in land cover mapping. In Proceedings of a National Conferenceon Resource Management Applications: Energy and Environment, VII: 32-40.

Kerber, A.G. and Schutt,J.B., 1986. Utility of AVHRR channels 3 and 4 in land-covermapping.Photogrammetric Engineering and Remote Sensing, 52: 1877-1883.

Kidwell, K.B., 1991. NOAA Polar Orbiter Data User's Guide. Washington, DC: National Oceanicand Atmospheric Administration, World Weather Building, Romm 100. (First issued 1986,revised 1991).

Kogan, F.N., 1990. Remote sensing of weather impacts on vegetation in non-homogeneous areas.International Journal of Remote Sensing, 11: 1405-1420.

Kuusela, K., 1994. Resources in Europe 1950-1990. Cambridge University Press.

Lambin, E.F., and Ehrlich, D., 1996, The surface temperature-vegetation index space for land coverand land cover change analysis, International Journal of Remote Sensing, 17: 463-487.

Leemans, R., and Cramer, W.P., 1992. USA database for mean monthly values of temperature,precipitation and cloudiness on a global terrestrial grid. Digital raster data on a 30 minutegeographic (Lat/Long) 360 x 720 grid. In: Global Ecosystems Database version 1.0: Disc A.Boulder, CO: NOAA National Geophysical Data Center. 36 independent single-attribute spatiallayers on CD-ROM, 15.6 MB (First published in 1991).

37

Loveland, T., Merchant, J., Ohlen, D., and Brown, J., 1991. Development of a land-covercharacteristics database for the contermious US. Photogrammetric Engineering and RemoteSensing, 57: 1453-1463

Loveland, T.R., Z. Zhu, D.O. Ohlen, J.F. Brown, B.C. Reed and L. Yang, 1999. An Analysis of theIGBP Global Land Cover Characterization Process. Photogrammetric Engineering and RemoteSensing (this issue).

Lloyd, D., 1990. A phenological classification of terrestrial vegetation cover using shortwavevegetation index imagery. International Journal of Remote Sensing, 11: 2269-2280.

Lunetta, Ross S., Congalton, Russell G., Fenstermaker, Lynn K., Jensen, John R., McGwire, KennethC., and Tinney, Larry R. Remote Sensing and Geographic Information System DataIntegration: Error Sources and Research Issues. Photogrammetric Engineering and RemoteSensing, Vol. 47, No. 6, June 1991, pp. 677-687.

Malingreau, J.P., 1990. The contribution of remote sensing to the global monitoring of fires intropical and subtropical ecosystems. In : Fire in the Tropical Biota, Ecosystem Processes andGlobal Challenges (Goldammer,J, Ed), Springer-Verlag, Berlin, pp.337-370.

Malingreau, J.P., 1992. Satellite based forest monitoring: A review of current issues. In ProceedingsInternational Space Year-World Forest Watch Conference, held in May 27-29 1992, Sao Josedos Campos, Brazil, EUR 14561 EN, 84 pp.

Malingreau, J.P. and Belward, A.S., 1992. Scale considerations in vegetation monitoring usingAVHRR data.International Journal of Remote Sensing, 13, 2289-2307.

Malingreau, J.P. and Tucker, C.J., 1987. The contribution of AVHRR for measuring andunderstanding global processes: large scale deforestation in the Amazon Basin. In Proceedingsof the IGARSS'87 conference held in Ann Arbor, Michigan, USA, on 18-21 May 1987 (NewYork: IEEE), 443-448.

Malingreau, J.P. and Tucker, C.J., 1988. Large scale deforestation in the Southeastern Amazon Basinof Brazil. Ambio, 17: 49-55.

Malingreau, J.P. and Tucker, C.J., 1990. Ranching in the Amazon Basin, Large-scale changesobserved by AVHRR. International Journal of Remote Sensing, 11: 187-189.

Malingreau, J.P, Stephens,G. and Fellows, L., 1985b. Remote sensing of forest fires: Kalimantan andNorth Borneo in 1982-3. Ambio, 14: 314-5.

Malingreau, J.P., Tucker, C.J. and Laporte, N., 1989. AVHRR for monitoring global tropicaldeforestation, International Journal of Remote Sensing, 10: 855-867.

Malingreau, J.P., DaCunha, R. and Justice,C. (Eds), 1992. Proceedings International Space Year-World Forest Watch Conference, held in May 27-29 1992, Sao Jose dos Campos, Brazil, EUR14561 EN, 35-47.

Mannstein, H. and Gessell,G., 1991. Deconvolution of AVHRR-data. In Proceedings of the 5thEuropean AVHRR Data Users Meeting, held at Tromso, Norway, 25-28 June 1991, EUM P09,ISBN 92-9110-003-X, 53-56.

38

Matson, M. and Holben, B.N., 1987. Satellite detection of tropical burning in Brazil. InternationalJournal of Remote Sensing, 8: 509-516.

Matson, M., Stephens,G. and Robinson, S., 1987. Fire detection using data from the NOAA-Nsatellites.International Journal of Remote Sensing, 8: 961-970.

Meyer-Roux, J., and Vossen, P., 1993. The first phase of the MARS project, 1988-1993: overview,methods and results, Proceedings of Conference on the MARS Project : Overview andPerspectives, Institute for Remote Sensing Applications, Villa Carlotta, Belgirate, LakeMaggiore, Italy, 17-18 November, 1983, pp. 33-81.

Moody, A. and Strahler, A., 1994, Characteristics of composited AVHRR data and problems in theirclassification, International Journal of Remote Sensing, 15: 3473-3491.

Muchoney, D., A. Strahler, J. Hodges and J. Locastro, 1999. The IGBP DISCover Confidence Sitesand the System for Terrestrial Ecosystem Parameterization: Tools for Describing andValidating Global Land Cover Data. Photogrammetric Engineering and Remote Sensing(1999).

Murai, S. and Honda,Y., 1992. World vegetation map from NOAA GVI data. In : Applications ofRemote Sensing in Asia and Oceania, Asian assoc. of remote sensing (Murai,S. Ed), pp. 3-9.

Nelson, R. and Holben, B., 1986. Identifying deforestation in Brazil using multiresolution satellitedata. International Journal of Remote Sensing, 7: 429-448.

Nelson, R. and Horning,N., 1987. Monitoring tropical deforestation in Mato Grosso, Brazil, usingLandsat MSS and AVHRR data. Final Report to The Ecosystems Center, Marine BiologicalLaboratory, Woods Hole, Massachusetts, USA (MBL P.O. 19167225), 47 pp.

Nelson, R., Horning, N. and Stone,T.A., 1987. Determining the rate of forest conversion in MatoGrosso, Brazil, using Landsat MSS and AVHRR data. International Journal of Remote Sensing,8: 1767-1785.

NOAA, 1992. Polar Metsat planning launch schedule and satellitre imagers, Bulletin, 1992, 2pp.

Pereira, M.C and Setzer,A.W., 1992. Spectral characterization of forest fires in NOAA-AVHRRimages. International Journal of Remote Sensing,(in press).

Pearson, E.S. and Hartley, H.O. (Eds.). Biometrika Tables for Statisticians. Volume 1, 3rd Edition.Cambridge University Press, Cambridge, England. 1966.

Price, J., 1984. Land surface temperature measurements from the split window channels of theNOAA 7 advanced very high resolution radiometer, Journal of Geophysical Research, 89:7231-7237.

Prince, S.D., Justice, C.O. and Los, S.O., 1990. Remote Sensing of the Sahelian Environment. Ispra,Italy; Commission of the European Communities.

Rosenfield, George H., K. Fitzpatrick-Lins, and H.S. Ling. Sampling for Thematic Map AccuracyTesting. Photogrammetric Engineering and Remote Sensing, Vol. 48, No. 1, 1981, pp. 131-137.

39

Roy, D.P, Kennedy, P., Folving, S., 1996, An Operational Methodology for Mapping EuropeanForest Cover using Low Spatial Resolution Satellite Data, Proceedings of InternationalWorkshop on the Application of Remote Sensing in European Forest Monitoring, Institute forSurveying and Remote Sensing, University of Agriculture, Vienna, Austria, 14-16th October1996.

Roy, D.P., Kennedy, P., Folving, S., 1997, Combination of the normalised difference vegetationindex and surface temperature for regional scale European forest cover mapping using AVHRRdata. International Journal of Remote Sensing, 18:1189-1195.

Roy, D.P., 1997, Investigation of the maximum Normalised Difference vegetation Index (NDVI) andthe maximum surface temperature (Ts) AVHRR compositing algorithms for the extraction ofNDVI and Ts over forest. International Journal of Remote Sensing

Running, S.W., Justice, C.O., Salomonson, V., Hall, D., Barker, J., Kaufman, Y.J., Strahler, A.H.,Heute, A.R., Muller, J.-P., Vanderbilt, V., Wan, Z.M., Teillet, P., and Carneggie, D., 1994.Terrestrial remote sensing science and algorithms planned for EOS/MODIS, InternationalJournal of Remote Sensing, 15: 3587-3620.

Sader, S.A., Stone,T.A. and Joyce, A.T., 1990. Remote sensing of Tropical Forests: An overview ofresearch and applications using non-photographic sensors. Photogrammetric Engineering andRemote Sensing, 56: 1343-1351.

Saunders, R.U. and Kriebel, K.T., 1988. An improved method for detecting clear sky and cloudyradiances from satellite radiances.International Journal of Remote Sensing, 9: 123-150.

Scepan, J., G. Menz and M.C. Hansen (1999). The DISCover 1.0 Image Interpretation Process.Photogrammetric Engineering and Remote Sensing (Jan 2000).

Sellers, P.J., 1985. Canopy reflectance, photosynthesis, and transpiration. International Journal ofRemote Sensing, 6: 1335-1371.

Singh, A., 1987. Spectral separability of tropical forest classes. International Journal of RemoteSensing, 8: 971-979.

Skula, J., Nobre,C. and Sellers,P.J., 1990. Amazon Deforestation and Climate Change. Science, 247:1322-1784.

Stone, T.A. and Schlesinger,P., 1990. Monitoring deforestation in the tropics with NOAA AVHRRand Landsat data. In Proceedings of an International Symposium on Primary Data Acquisition,Manaus, 1990. International Archives of Photogrammetry and Remote Sensing, 28: 197-202.

Stone, T.A. and Schlesinger,P., 1992. Using 1km resolution satellite data to classify the vegetation ofSouth America. In Proceedings of a conference on Remote Sensing and permanent plottechniques for world forest monitoring, held on 13-17 January 1992, IUFRO, Pattaya, Thailand,85-93.

Stone, T.A. Foster-Brown,I; and Woodwell,G.M., 1991. Estimation, by remote sensing, ofdeforestation in central Rondonia, Brazil. Forest Ecology and Management, 38: 291-304.

40

Story, M. and R.G. Congalton, 1986. Accuracy Assessment: A User’s Perspective. PhotogrammetricEngineering and Remote Sensing, Vol. 53, No. 3, p. 397-399.

Tarpley, J.D., Schneider,S.R. and Money,R.L., 1984. Global vegetation indices from the NOAA-7meteorological satellite. Journal of Climate and Meteorology, 23: 491-494.

Tateishi, R. and Kajiwara,K., 1992. Global land cover classification by NOAA GVI data. In :Applications of Remote Sensing in Asia and Oceania, Asian assoc. of remote sensing (Murai, S.Ed), 3-9.

Teillet, P.M. and Holben,B.N., 1992. A multi-level electronic database for documentation anddissemination of time-dependent NOAA-AVHRR calibration coefficients for the solarreflective channels. In Proceedings of the 6th Australasian Remote Sensing Conference,Wellington, New Zealand.

Thomas, I.L., Benning, V.M., and Ching, N.P., 1987. Classification of Remotely Sensed Images,Hilger, Bristol, pp. 82-101.

Townshend, J.R.G., 1992 (Ed). Improved global data for land applications. IGBP Global ChangeReport No.20. IGBP, Stockholm, Sweden, 87 pp.

Townshend, J.R.G., C.O. Justice, D. Skole, J.-P. Malingreau, J. Cihlar, P.Teillet, F. Sandowski andS. Ruttenberg. 1994. The 1Km Resolution Global Data Set: Needs of the InternationalGeosphere Biosphere Programme. International Journal of Remote Sensing, Vol. 15, No. 17,3417-3442.

Townshend, J.R.G., Goff, T.E. and Tucker, C.J., 1985. Multitemporal dimensionality of images ofnormalised difference vegetation index at continental scales. I.E.E.E. Transactions onGeoscience and Remote Sensing, 23: 888-895.

Townshend, J.R.G., Justice,C.O. and Kalb,V.T., 1987. Characterization and classification of SouthAmerican land cover types using satellite data. International Journal of Remote Sensing, , 8:1189-1207.

Townshend, J.R.G., Justice, C.O., Li,W., Gurney, C. and McManus, J., 1991. Global land coverclassification by remote sensing: present capabilities and future possibilities. Remote Sensing ofthe Environment, 35: 243-256.

TREES, 1991. Tropical Ecosystem Environment observation by Satellite. SP-I, 90.31. CEC-JRC,Ispra, Italy, Special Publication.

Tucker, C.J., Vanpraet, C.L., Boerwinkle,E. and Easton, A., 1983. Satellite remote sensing of totaldry matter accumulation in the Senegalese Sahel. Remote Sensing of the Environment,13: 461-469.

Tucker, C.J., Gatlin, J.A. and Schneider, S.R., 1984a. Monitoring vegetation in the Nile Delta withNOAA-6 and NOAA-7 AVHRR imagery. Photogrammetric Engineering and Remote Sensing,50:53-61

Tucker, C.J., Holben, B.N. and Goff, T.E., 1984b. Intensive forest clearing in Rondonia, Brazil, asdetected by satellite remote sensing. Remote Sensing of the Environment, 15: 255-262.

41

Tucker, C.J., Townshend, J.R.G. and Goff, T.E., 1985. African land-cover classification usingsatellite data. Science, 227: 369-375.

UNESCO, 1981. Vegetation map of South America, 2 sheets of 1:5m map, plus documentation (189pp).

Van Roessel, J.W., 1990. First experiment for pilot test definition of large area forest monitoringmethodology through sampling and use of Landsat and AVHRR imagery with applications inBrazil. Report to EROS Data Center, Sioux Falls, USA, 20 pp.

Viovy, N., 1992. Etude spatiale de la biosphere terrestre: Integration de modeles ecologiques et demesures de teledetection. PhD Thesis.LERTS, CNES, France, 213 pp.

Viovy, N., Arino, O. and Belward,A.S., 1992. The best index slope extraction (BISE): a method forreducing noise in NDVI-time series. International Journal of Remote Sensing, 13, 1585-1590.

Vogt, J., 1992. Characterizing the spatio-temporal variability of surface parameters from NOAAAVHRR data: A case study for Southern Mali. PhD dissertation, EUR 14367 EN, CEC-JRC,Italy, 266 pp.

Walsh, T.A. and Burk, T.E, 1993, Calibration of satellite classifications of land area, Remote Sensingof Environment , 46:281-290.

Warren, D.E., 1989. AVHRR channel-3 noise and methods for its removel. International Journal ofRemote Sensing, 10: 645-652.

Woodwell, G.M., Houghton, R.A., Stone, T.A., Nelson, R.F. and Kovalick,W., 1987. Deforestationin the tropics: new measurements in the Amazon Basin using Landsat and NOAA AdvancedVery High Resolution Radiometer imagery. Journal of Geophysical Research, 92: 2157-2163.

Zhu, Z., and Evans, D., 1994. U.S. forest types and predicted percent forest cover from AVHRR data,Photogrammetric Engineering and Remote Sensing, 60: 525-532.

42

FRA Working Papers

0. How to write a FRA Working Paper (10 pp. – E)

1. FRA 2000 Terms and Definitions (18 pp. - E/F/S/P)

2. FRA 2000 Guidelines for assessments in tropical and sub-tropical countries (43 pp. - E/F/S/P)

3. The status of the forest resources assessment in the South-Asian sub-region and the country capacitybuilding needs. Proceedings of the GCP/RAS/162/JPN regional workshop held in Dehradun, India, 8-12June 1998. (186 pp. - E)

4. Volume/Biomass Special Study: georeferenced forest volume data for Latin America (93 pp. - E)

5. Volume/Biomass Special Study: georeferenced forest volume data for Asia and Tropical Oceania (102pp. - E)

6. Country Maps for the Forestry Department website (21 pp. - E)

7. Forest Resources Information System (FORIS) – Concepts and Status Report (20 pp. E)

8. Remote Sensing and Forest Monitoring in FRA 2000 and beyond. (22 pp. - E)

9. Volume/Biomass special Study: Georeferenced Forest Volume Data for Tropical Africa (97 pp. – E)

10. Memorias del Taller sobre el Programa de Evaluación de los Recursos Forestales en once PaísesLatinoamericanos (S)

11. Non-wood forest Products study for Mexico, Cuba and South America (draft for comments) (82 pp. – E)

12. Annotated bibliography on Forest cover change – Nepal (59 pp. – E)

13. Annotated bibliography on Forest cover change – Guatemala (66 pp. – E)

14. Forest Resources of Bhutan - Country Report (18 pp. – E)

15. Forest Resources of Bangladesh – Country Report (89 pp. – E)

16. Forest Resources of Nepal – Country Report (under preparation)

17. Forest Resources of Sri Lanka – Country Report (under preparation)

18. Forest plantation resource in developing countries (75 pp. – E)

19. Global forest cover map (14 pp. – E)

20. A concept and strategy for ecological zoning for the global FRA 2000 (23 pp. – E)

21. Planning and information needs assessment for forest fires component (32 pp. – E)

22. Evaluación de los productos forestales no madereros en América Central (102 pp. – S)

23. Forest resources documentation, archiving and research for the Global FRA 2000 (77 pp. – E)

24. Maintenance of Country Texts on the FAO Forestry Department Website (under preparation)

25. Field documentation of forest cover changes for the Global FRA 2000 (under preparation)

26. FRA 2000 Global Ecological Zones Mapping Workshop Report Cambridge, 28-30 July 1999 (53 pp –E)

27. Tropical Deforestation Literature:Geographical and Historical Patterns in the Availability ofInformation and the Analysis of Causes (17 pp. – E)

28. World Forest Survey – Concept Paper (under preparation)

Please send a message to [email protected] for electronic copies or download fromhttp://www.fao.org/FORESTRY/FO/FRA/index.jsp

http://www.fao.org/FORESTRY/FO/FRA/index.jsp

fra 2000 f c m & m noaa-avhrr other coarse … · 5.5 firs – forest information from remote...

Documents