Study of Land-Use and Deforestation In Central African Tropical Forest Using High Resolution SAR Satellite Imagery
AbstractDeforestation in tropical Africa is driven by a variety of socio-economic and environmental factors, and has resulted in land cover changes that threaten biodiversity, water and energy resources, and contributes to trace-gas emissions. Several conservation and development studies have concluded that the deforestation in Africa is closely tied to demographic conditions such that the greatest loss of rain forests has occurred in countries with higher population growth. However, lack of reliable data and survey information in some countries, has made the estimation of areas of intact forest and/or under land use change and their relation to economic indicators surprisingly difficult to establish. Consequently, the extent and rate of deforestation in Africa are less well known than other regions of tropics.
In this study, we propose to use high resolution satellite imagery to map areas of forest clearing and general land cover types for the entire Central African tropical region. Images acquired by JERS-1 SAR (Synthetic Aperture Radar) instrument during its global rain forest mapping (GRFM) phase will be the main source of data for this study. By using a combination of radar backscatter and texture analysis, and a SAR-specific classifier, we propose to classify the images into five categories of forest, nonforest, savanna, flooded forest, and open water. Approximately 4000 high resolution (12.5 m) images for two periods of dry and wet seasons will be processed and delivered to the NASA/Jet Propulsion Laboratory and the European Commission Joint Research Center in Italy for generating a geocoded 100 m resolution mosaic of image data and a land cover map. The thematic interpretation of JERS-1 data and the validation of the land cover map will be supported by available Landsat TM images acquired by NASA pathfinder project, ERS-1 mosaic of Africa generated by ESA TREES (Tropical Ecosystem Environment Monitoring by Satellites) project, and vegetation maps and field data provided by collaborators from national institutions in the region. The resulting land cover map will be integrated in a socio-economic study to assess the trends and patterns of deforestation by including demographic and social and environmental policies.
RationaleThe amount of tropical rain forests cleared for colonization, conversion into agricultural fields, and the use of wood for fuel is important for assessing the net release of carbon into the atmosphere. The forest conversion affects the biogeochemical cycle and global atmospheric concentration of carbon dioxide in two phases: (1) the initial clearing of forests (burning and logging) releases significant amount of carbon; and (2) the type and intensity of subsequent agricultural use determines the long-term carbon cycling. The conversion/intensification sequence impacts other important processes such as biotic control of water and energy exchange between the land surface and atmosphere (Houghton, 1983, 1991).
As the destruction of tropical rain forests increased during the past 25 years, attempts in understanding its causes have multiplied. In general, the agents of deforestation can be found among the commercial loggers, cattle ranchers, and slash-and-burn cultivators. Since mid-1960s, the shifting cultivators of traditional type that practiced the small scale migratory agriculture has been gradually replaced by what is called the 'shifted' cultivators according to Myers (1984). These are farmers who moved from long-established farmlands of territories far distant from the forests into unoccupied lands available to them in tropical forests. The cause of this move resides in population growth, misdistribution of established farmlands, lack of agrotechnologies for intensive cultivation, and inadequate rural developments (Burgess, 1983). These shifted cultivators have grown exceptionally numerous in recent decades (World Resources Institute and World bank, 1985). The mass immigration of about 1.5 million small-scale farmers in Rondonia, Brazil between 1975 and 1988 is a notable example (Myers, 1991).
There is a general agreement among environmental experts and social science scholars that social configurations influence land use (Hecht, 1985; Hecht and Cockburn, 1989). The relation between social structure and ecological processes is key to understanding the degradation or enhancement of natural resources. The clearance of African rainforest is strongly tied to the socio-economic forces within the developing countries of the region, and is influenced by current and future environmental, political, and demographic conditions. The population of Africa is doubled in the past 25 years, it will double in 24 years and it will not stabilize until the year 2050 (World Bank, 1992). The greatest loss of forest in recent years has been in countries with high population growth. World Bank research suggests that shifting cultivation and conversion of the natural forest cover into perennial cash crops are the direct causes of forest degradation in the region. These causes have their origin in the rapid population growth, depending on the magnitude of the internal and external migration and the agricultural stagnation (Cleaver et al., 1992; Cleaver and Schreiber, 1994).
The costs of the deforestation in the region is only beginning to be understood. The most obvious costs are the loss of future wood for forest industry, destruction of biological diversity, changes in biogeochemical cycles and climate by altering local rainfall and hydrological processes, and desertification. The degradation of vegetation cover has caused northern areas of some West and Central African countries that were previously under forest to alter to savanna grasslands and degraded savanna. Investigations show that Africa contributes to gas emissions from biomass burning more than other tropical region of the world (Andreae, 1990). Although most of the biomass burning occurs in a small number of countries and in dry regions, there is an uncertainty in the geographical locations and the type of the forest (closed or open forest) being burned. Given this uncertainty, the estimate of forest biomass burned in tropical Africa because of deforestation is about 130 MT dm (megatons dry matter) per year (Delmas, et al., 1991). This estimate depends on three parameters: the area annually cleared, the average phytomass, and the burning efficiency. There is an increasing demand to reduce the uncertainty in all three parameters, and monitoring land use changes and deforestation is a key element of this process. By improving this estimate, our understanding of the rate of trace gas release to the atmosphere, and the current status of global climate models, can also improve.
In general, it is difficult to establish how big the problem of deforestation is neither it is easy to say categorically why the forest is being cleared because the reasons may vary from one area to another and may change through time. A coordinated program has been designed to conduct rigorous comparative studies at the regional and local scales to understand the wide diversity of causes for tropical deforestation and to identify cross-national and regionally specific types of deforestation by quantitative measures (Turner et al., 1994). These studies depend heavily upon remote sensing data from high resolution satellite imagery and aerial photography backed up by field data (Myers, 1988; Skole and Tucker, 1993; Moran, 1994; Skole et al., 1994). Satellite data provides large scale quantitative data for estimating the extent of deforestation and other forms of land cover changes. Landsat Thematic Mapper (TM) has been the most effective instrument for deforestation studies and mapping land cover types (Malingreau and Tucker, 1985; Skole and Tucker, 1993). Currently, Landsat TM data are used for routinely estimating the rate of deforestation in the Amazon basin (INPE, 1992). Similar efforts are underway to study the deforestation in African tropical region. The NASA Landsat Pathfinder program has been collecting Landsat images for 1970s, 1980s, and 1990s period over Central Africa. However, due to continuous cloud cover the frequent acquisition of data has been impeded for some areas, making the estimation of the annual rate of deforestation in Africa is a difficult if not an impossible task (see Figure 1).
Synthetic aperture radar (SAR) data can be used to circumvent this problem. Moreover, it has been shown that SAR images are useful for land cover classification because of their sensitivity to vegetation structure and biomass (Saatchi et al., 1996a; Rignot et al., 1996). In late 1995, JERS-1 (Japanese Earth Observation Satellite) entered into its tropical rain forest mapping (GRFM) phase and has been collecting high resolution data (12.5 m resolution, L-band, HH polarization) over the entire tropical rainforest including West and Central Africa. The JERS-1 coverage of African rain forests consists of about 4000 frame images (75 km x 75 km) for two periods of high and low flood seasons that can be used to map general vegetation classes and areas of deforested in the region.
ObjectivesThe proposed study intends to address three distinct and interrelated scientific goals:
BackgroundRainforest of Central Africa
Tropical forest of Africa is 18 per cent of the world total and covers over 3.6 million square kilometers of land in West, East and Central Africa. This total area can be subdivided to 2.69 million square kilometers (74%) in Central Africa, 680,000 square kilometers (19%) in West Africa, and 250,000 square kilometers (7%) in East Africa (Sommer, 1976). In West Africa, a belt of rain forests up to 350 km long stretches from the eastern border of Sierra Leone all the way to Ghana. In Ghana the forest zone gradually dissipates near the Volta river, following a 300 km stretch of Dahomey savanna gap. The rain forest of West Africa continues from east of Benin through southern Nigeria and officially ends at the border of Cameroon along the Sanaga river. FAO study divides the two regions along the national border of Cameroon for practical purposes, however, the distribution of many West African plant and animal species ends at the Sanaga river (Richards, 1979; White, 1983; Hamilton, 1984).
Schematically the rain forest vegetation of the Guinea-Congolian transition region, extending from Senegal to western Uganda are constituted of two main types:
The most recent vegetation map of Africa was published by UNESCO (White, 1983), and the main vegetation features of Central African rain forest can be divided to the following categories:
The term deforestation refers to complete destruction of forest canopy cover through clearing for agriculture, cattle ranching, plantations, or other non-forest purposes. Other forms of land-use changes such as forest fragmentation (altering the spatial continuity and creating a mosaic of forest blocks and other land cover types), and degradation (selective logging of woody species for economic purposes that affects the forest canopy and the biodiversity) are often included in estimating deforestation. The characterization of forest into one of these categories depends on the temporal and spatial scale of observation. A small farmer's plot may be seen as deforestation in local scale but a forest fragmentation in regional scale. At the same time, if the density of crops is low enough to keep the forest ecological function intact and the return period is long enough to allow regeneration, then shifting cultivation may not imply deforestation. The subjective meaning of the term deforestation is thus not only linked to a value system but to the nature of the measurement designed to assess it (Poor, 1976; Mayaux and Malingreau, 1996). Adopting different perspectives of deforestation in data analysis have caused considerable variations in estimation of the area of forest cleared.
The tropical forest of Africa is 18 per cent of the world total and covers over 20 million km2 of land in West and Central Africa. This region has been facing deforestation with various degrees of intensity throughout this century. The actual pace of deforestation varies from one country to another and accurate data do not yet exist. Recent estimates indicate that the annual rate of deforestation in the region can vary from 15,000 hectares in Gabon to 290,000 ha in Cote d'Ivoire (FAO, as of 1980). Remaining tropical forest still cover significant areas in Central Africa but are reduced to patches in West Africa. Table 1 and 2 show the latest estimates of forest area and forest loss rates for Central African countries for which data are available (FAO/UNEP, 1988).
Table 1. Distribution of closed broad-leaved forests and fallow (deforested) in Central Africa at the end of 1980 (FAO/UNEP, 1988)
Table 2. Distribution of closed broad-leaved forests and fallow (deforested) in Central Africa at the end of 1985 (FAO/UNEP, 1988)
In general, estimates of deforestation in African rain forests suffer from lack of accurate maps and up-to-date survey data. In many of the countries in the region, the rate and patterns of clearance vary rapidly depending on several socio-economic factors. The time scales used in measurements and the availability of accurate baseline surveys in some countries against which to compare recent changes are important for assessing the deforestation on a regional scale. It is possible to obtain deforestation estimates using tabular summaries from standard government census sources. However, these data do not exist in all countries in the region and if available, they do not directly report the area deforested, but instead they provide estimates of land in various forms of permanent and temporary agriculture and pasture. These data can be used as a proxy for deforestation but cannot provide a direct measure.
Remote sensing holds much promise for revealing exactly where and when deforestation is taking place and for providing baseline information for assessing land-use changes. In recent years, the use of satellite imagery has shown that earlier statistics are not reliable and are open to many errors and distortions. Analysis of remote sensing data suggest that the overall rate of loss of rain forests (204,000 km2) is over 80 percent higher than the estimates by FAO in 1980 (114,000 km2) (United Nations Environment Program, 1987). There have been a number of applications of remote sensing to estimate deforestation rates and total area cleared over large regions (Tucker, 1984; Nelson and Holben, 1986; Malingreau and Tucker, 1988; Skole and Tucker, 1993). Although initial utilization of regional estimates diverged widely, they converged in recent years as a result of a wider utilization of high-resolution (< 100 m) imagery (Townshend and Justice, 1988). Deforestation has been quantified using Landsat TM imagery over the entire legal Amazon (Sokle and Tucker, 1993) and in southeast Asia through NASA's Landsat Pathfinder project (Chomentowski et al., 1994). Over Africa, JRC TREES (Tropical Ecosystem Environmental Observation by Satellites) project has been using a multi-satellite observation to map land-cover and deforestation in Central Africa (Mayaux and Malingreau, 1996; TREES, 1995). Radar remote sensing technology has also been used in tropical forest studies primarily for their mapping purposes (TREES, 1995). Imaging radar instruments can acquire continental-scale data sets without being constrained by cloud cover. In addition, they can provide information about the density, structure, biomass, and understory vegetation in forests and the type of clearing practices because radar signal probe the larger size constituents of the forest and penetrate deep into the forest canopy. Results from the analysis of Shuttle Imaging Radar SIR-C/X-SAR, JERS-1 and ERS-1 data indicate that the imaging radars can map deforestation in the tropical rain forest environment and when combined with optical data can improve the detection of secondary growth and plantations(Saatchi et al., 1996; Rignot et al., 1996; TREES, 1995; Sader, 1987).
JERS-1 images acquired during the GRFM phase of the satellite are being processed and will be delivered to JPL and JRC as the PIs of the NASDA project over African rain forest. The entire African rain forest was mapped by about 2000 scenes between January and March of 1996. Another data acquisition is scheduled for October-November 1996 to map the high flood area of the Congo basin in order to provide a seasonal variability of flooded forests. Landsat images are available for 1970s, 1980s, and 1990s years. In areas that the 1990 Landsat images are not available, JERS-1 will help to fill the gaps, and in other areas it can improve the classification of Landsat data for generating regional forest map. In addition, the Africa mosaic of ERS-1 data will be available through institutional agreements. We will investigate the use of this data set in interpreting the JERS-1 data or improving the classification accuracy. A brief outline of our approach follows.
Data Processing and Management
The analysis of the JERS-1 SAR data requires a number of interrelated processing tools that can be applied to a large volume of data in order to make feasible the extraction of thematic information from the signal. These tools function at various levels. The first layer processes data to correct the geometry and radiometry, reduce spatial sampling and increase the signal to noise ratio. The image files received from NASDA are processed to 12.5 m pixel spacing and will go through the following chain of processing tools:
The choice of land-cover classes depends on two factors: (1) the importance of the
classes in the global change and land-use change studies, 2) the characteristics
and the dynamic range of the JERS-1 SAR backscatter data. The general categories
of the land-cover types for global change ecosystem process studies have been suggested by
the core project of the International Geosphere Biosphere Program (IGBP, 1990).
We propose to use these categories as our guideline to study class reparability in
JERS-1 images. The use
We propose to implement a supervised MAP (Maximum a posteriori Baysian) classifier based on the a priori knowledge of tonal and texture features of the radar backscatter. The training areas will be selected among various types of land cover visually separated in JERS-1 backscatter images over test sites. Labeling of training sites into land cover types will be based on the classes identified in Landsat TM images, existing maps and field data over test sites, and the knowledge of the radar scattering from vegetated terrain. For example, the areas of lowest radar cross section are related to either the deforested areas or grassland savanna, or high radar cross section may correspond to inundated/swamp forests and urban areas. Texture information extracted over these areas can help to identify them more accurately. The classifier has been tested on JERS-1 mosaic backscatter image acquired over the Amazon rain forest (Figure 4). In this study, we propose to apply the classifier on individual images and generate 100 m resolution land cover maps. These maps will be georeferenced, projected and used to form a land cover mosaic map over the entire Central African rain forest.
We have chosen test sites to verify the accuracy of land cover maps. These sites cover a wide range of land cover types: areas representing a cross-section of geographic locations, deforestation and land use practices, undisturbed regions of ecological importance, forest fragments under biodiversity and conservation studies (protected areas and wildlife reserves), areas within the flooded forest of the Congo basin, and areas overlapping the IGBP transects and NASA Landsat pathfinder test sites. We have established the initial contacts and intend to work closely with the IGBP Core projects, NASA Landsat pathfinder, TREES projects, The World Bank environmental and policy departments, and several conservation societies in order to valid the local and regional vegetation maps. The test sites will be chosen from areas where Landsat TM date are available. Depending on the time of data acquisition and intensity of land-use changes, TM images can be used either in the classifier or as for validation purpose. The list of potential sites for validation of the regional land cover map and the correspoding collaborating individuals and agencies are given in Appendix A.
On the regional scale, we will compare and verify the land cover map with the AVHRR derived vegetation map of Central Africa produced by TREES project (Laporte, 1995; Mayaux and Malingreau, 1996). The comparison will allow us to produce a more accurate map of the region where ambiguities in forest-savanna and secondary growth cover types in AVHRR vegetation map are potentially removed.
Socio-economic Analysis of Deforestation in Africa
A. Analytic Context and Approach
Deforestation is driven by social and economic dynamics. The prevailing models rely largely on demographic and market factors to explain clearing, in part because these data are collected by governments, and seem to correlate with land use change. These models tend to view the evolution of clearing patterns as straightforward linear processes that diffuse through the landscape. Recent scholarship on tropical deforestation has, however, brought more complex, socially nuanced analyses to bear. This research has focuses on how social factors organize land use through mediation of markets, institutions, communities and the political economy of local, national and international actors. These approaches, which now fall under the rubric of the political economy of natural resources, or political ecology, have strong explanatory powers and thus are far more useful for understanding the outcome and options of policy and the processes of tropical development. Correlative demographic models while powerful statistically, reveal little about dynamics, equity issues or potential sustainability of land uses. In the main such models generate unsophisticated policy tools, since in many cases the drivers underlying deforestation may have little to do with population (Hecht, l989; Angelson, l995; Repetto l985; Berry, l993; Bassett and Crummey, l993)
Recent studies examine the logic and dynamics of deforestation at several different levels and introduce many innovations. At the international level, recent scholarship has emphasized the role of structural adjustment and the need for international currencies, debt dynamics, and the increasing internationalization of markets to understand how these potentially effect patterns of deforestation. (Lean et al l990). The surge of deforestation in Cameroon largely reflects the need for debt repayment, while the demand for rubber has actually intensified the maintenance of forest cover in Indonesian small holder settlements (Angelsen l995).
At the national level, research emphasizes the role of macroeconomics policy ( and its inherent political interests) in stimulating deforestation by its ability to distort market functions. Logging in Indonesia, livestock in Amazonia, Cacao in the Ivory Coast are classic examples of policy distortions driving clearing (with very inequitable social as well as environmental consequences) (See Repetto and Gillis l988, Hecht and Cockburn l989, Peluso l993) Another dimension is the state's role in guiding the colonization into humid tropical zones for various reasons, including revenues, geopolitical concerns, as a means of defusing political conflict, and as a counter insurgency tactic (Chazen l988, Bates l981, Hart l982). National kleptocracies may find it very useful to permit predatory exploitation of tropical resources by political cronies, a problem often observed in undemocratic regimes such as that of Nigeria and Zaire (MacGaffey, l987, l991).Environmental policy is also a factor, but an equivocal one since the way it is implemented can trigger resistance which can result in sharp clearing and resource poaching (Western et al l995).
At the local level several dynamics affect deforestation, including those linked to institutions, resource control and acquisition, as well as to subsistence. In cases of unclear property rights in contexts of limited local sanctions and powers, deforestation to claim land and its ancillaries (timber, minerals etc.) often occurs. (Berry l988). But, it is important to also assert that local populations may inhibit such incursions by invoking the claims of traditional rights. The arena of property rights is thought to be significant in deforestation processes, although how property regimes unfold in clearing remains contradictory (Berry, 1988).
Widely cited as the dominant driver of tropical deforestation, (Bandy et al, l994) in many contexts this system of successional management may actually be among the most important means of intensifying tropical agriculture in humid tropical settings. A great deal of data from Asia, Africa and many Latin American countries points to spontaneous intensification of land uses by shifting cultivators responding to market opportunities for forest products (Hecht et al l996, Angelsen l995, Guyer, and Lambin l993) Dvorak l991.
The complexity of deforestation processes is further compounded by the fact that several mechanisms may be operating at the same time. Given the relative paucity of detailed studies of the dynamics of deforestation in Africa, (especially when compared with Latin America and Asia), the research we outline has the potential :
Step 1: Historical Patterns of Deforestation
The countries of central and West Africa that are the subject of this research are extremely heterogeneous in their ecology, ethnic diversity, colonial history, political structures, economic configurations, institutional arrangements and development approaches to mention only a few of the variables that play out in the region. Any meaningful analysis of such diversity requires both inductive and deductive techniques. This involves explaining deforestation patterns as they have developed using the international, state and local approaches outlined above to generate a recent historical, policy, institutional, and economic analysis for each relevant area to construct a general environmental history of regional forests. Archives and current remote sensing data can provide more precise data on land use and land cover change. This exercise provides the background for more detailed hypothesis development, which will be established in consultation with the entire research team in such a way that social analysis, the remote sensing archives and current images are used to their maximum advantage. Finally policy analysis and suggestions can be from these efforts.
Step 2. Hypothesis development and evaluation.
This section will be developed with a concern for understanding economic, social and environmental policy effects on deforestation patterns in the region. While the exact questions cannot be finalized at this time, certain kinds of questions already suggest themselves because they reflect policy changes that should be detectable in changes in land use.
The combination of analytic models with the current and archival empirical data generated from remote sensing should provide some exciting insights into policy performance. Since this work will not be uniquely "at a distance", and will incorporate a complex analysis of policy and its context, we are confident that this study can make major contributions to the study of the technical dimensions of deforestation monitoring, the understanding of the social dynamics of clearing, and the performance of policy in mediating the patterns of cutting we observe
The proposed work will have three phases to complete: 1) post-processing the data and generating a geocoded mosaic of 100 m resolution image data, development of classification algorithms, and accumulation of ancillary data for land use and socio-economic studies, 2) implementing the classifier over the images and validation of maps by incorporating field data maps acquired by our collaborators in Central African countries, 3) geocoding the land cover map and determining the general patterns and trends of deforestation and land use over the entire region. The project will require three years, starting January 1997, to complete. The socio-economic research will be performed on local and regional scales as maps and information over test sites and region become available. A portion of the data required for socio-economic research will be provided by World Bank, international environmental and Central African national institutions (see Appendix A).
Travel costs are for trips to Washington for science team meeting and the use of libraries and research facilities at the World Bank. If necessary, we will have site visits to coordinate the proposed study with our collaborators, acquire field data, maps and sociological and land-use information from national institutions, and provide educational material.
S. Saatchi-(PI) Task management, development of classification algorithm, validation, geocoding and producing regional and national land cover maps.
E. Njoku-(Co-I) Data analysis and validation.
S. Hecht-(CO-I) Socio-economic analysis.
F. De Grandi-(Co-I) Post-processing and generating geocoded JERS-1 mosaic image data, classification algorithm.
J-P Malingreau-(Co-I) Validation, and coordination with IGBP and TRESS projects, comparison with regional AVHRR-based vegetation map.
M. Massart-(Co-I) Validation of results using Landsat TM and field data, coordination with NASA Landsat Pathfinder/Central Africa program.
This work has not been funded by NASA or any other institutions. The NASDA GRFM project will provide JERS-1 data without any costs to investigators (JPL and JRC). The level of effort will be divided among JPL, UCLA, and UMD (partially funded by NASA Landsat Pathfinder Program). The effort of JRC will be funded by TREES project. Here, a summary of the required budget is outlined for each year. The fringe benefits and burdens are included in the costs. The detailed cost plan for each institution is provided in Appendix B.
Cost Plan Summary