20010822.doc
PAMS department, University of Canterbury, Christchurch, New Zealand
Forest fragmentation near Tai National Park, Ivory Coast
Abstract: Tropical deforestation and the consequent destruction and fragmentation of habitat are the primary threats to biodiversity in developing counties today. Satellite imagery is used to study the extent of deforestation in an area centered on the Guiglo-Taï road in southwest Ivory Coast, to the northwest of the Taï National Park. Features including primary forest, secondary forest, bare ground and watercourses are classified. The study area partly encompasses several protected areas. The protected areas are found to have mostly escaped deforestation, however forests outside the protected areas are severely fragmented and degraded.
Keywords: Remote sensing, fragmentation, Ivory Coast, Cote d'Ivoire
Introduction
Tropical deforestation and the consequent destruction and fragmentation of habitat are the primary threats to biodiversity in developing counties today. The Ivory Coast in West Africa reached an annual deforestation rate of 6.5 percent in the 1980's, one of the highest rates in the world (Chatelain et al. 1996). The Ivory Coast also has some of the most important refuges for biodiversity. The Taï National park contains the largest remaining tract of Guinean rainforest and was identified as one of sixteen world ‘hotspots' of biodiversity by Myers (1990).
Remote sensing is the only feasible way to map tropical forest fragmentation at regional and global scales (Foody et al. 1997). Improvements in technology and availability of imagery are rapidly increasing the importance of the field in many areas including forest ecosystem monitoring (Lambin & Ehrlich 1997).
This study aims to use remote sensed imagery to assess the extent of deforestation in a region in southwest Ivory Coast. The first goal is to monitor how effectively protected areas and reserves in Ivory Coast are halting the loss of forest in the region. The second goal is to assess the extent and nature of fragmentation in the study area.
Methods
A description of the techniques employed in this study follows. The location of material produced is listed in Appendix 1.
Study Area
The study area is 70km wide and 50km high and is located between 6º08'N and 6º35'N and between 7º00'W and 7º48'W in southwest Ivory Coast. The area includes part of Lake Buyu, the villages of Guiglo and Zagné, the northwest section of the national park of Taï, the N'Zo game reserve and the forest reserves of Goin-Debe and Cavally.
The region is particularly suited to studying deforestation in Ivory Coast because it includes a representative sample of protected and unprotected areas as well as pristine and fragmented forest. The Guiglo-Taï road traverses the area and deforestation has proceeded from this access route. Additionally a study by (Chatelain et al. 1996) covers a similar region and allows for comparison of results.
Data
The study area falls within the Landsat worldwide reference system path and row 198/56. A Landsat TM image acquired on September 3, 1988 (10:12am) was obtained for these coordinates. The scene was geo-referenced to the UTM projection and the WGS84 datum. Plate 1 shows the RGB (visible) bands from the whole Landsat image with the study area highlighted.
Plate 1. The study area, from the 1988 Landsat image. [$home/tai/tai_mini.img]
Classification procedure
During the classification procedure the spectral information of each pixel within the scene was assessed and used to group similar pixels into the classes. Pixels were assigned to seven classes; water, bare ground, clay/urban, marshland, open canopy vegetation, closed canopy forest and unassigned.
The first step in the classification process was to execute a neighbourhood function using the magnify algorithm and a 3x3 pixel window. This process averages groups of nine pixels so that mixed pixels become less important and the classification procedure is more accurate.
The next step was to perform an unsupervised ISODATA clustering algorithm on the scene. The ISODATA algorithm identifies groups of pixels with similar spectral signatures and clusters these groups. The operator must identify the number of anticipated classes, the quantity of clusters, within the image. At this stage two cluster quantities were tested, eight clusters and sixteen clusters. The cluster algorithm was executed on all seven Landsat bands and repeated for sixteen iterations.
The actual scene classification involved comparing the output classes from the unsupervised ISODATA algorithm with visual identification of the terrain and knowledge of the type of land use in the area from Chatelain (1996).
The final step in classification was to use the map composer in Erdas Imagine to create Plate 2, the presentation image, complete with scales and legend.
Fragmentation analysis
The fragmentation analysis attempted to assess the nature and amount of fragmentation in the area surrounding the Guiglo-Taï road. Primary deforestation spread out from this road and it is important to have a general idea of the impact of environmental change on local species.
Once the scene had been classified into multiple classes as described above a model was created using the Imagine Spatial Modeling Language (SML) and used to separate the closed canopy forest class from the rest of the identified classes (see Appendix 2). In addition to including the closed canopy forest class in the output a one-pixel buffer zone of open canopy forest was included around the primary forest, if it existed. This was necessary to ensure mixed pixels were included in the forest and to more accurately reflect the way organisms perceive forest edge as a gradient rather than a sharp boundary. The output from this process was the binary mask, Plate 3, showing the remaining areas of relatively undisturbed forest and isolated forest fragments.
Results
The unsupervised classification into eight clusters proved to be unsatisfactory. There were too few classes to separate features adequately. The classification into sixteen classes was far more acceptable. The closed canopy forest was separated into five classes so these were merged to form a single class primary forest class. The successional forest was separated into three very similar classes and these were also merged into a single class. Water features produced two classes that were merged into a single class.
Plate 2. The final classification of the study area. [$home/tai/tai.map]
The forest/non-forest binary mask in Plate 3 shows fragmentation in the study area. It is worth noting that quantitative measures of fragmentation are scale dependent. The amount of fragmentation measured will vary depending on the scale at which you measure (Riitters & Wickham 2000; Wiens 1989).
Plate 3. The binary mask contrasts forest and non-forest pixels. [$home/tai/tai_frags.img]
Discussion
In their study, Chatelain (1996) demonstrated that ground-truthing is essential to differentiate between young, early successional forests and degraded primary forests. In this study these two classes have been classified to together and degraded forest. As such the classification map will overestimate the amount of usable habitat for some species that are not able to survive in early successional forest (Lovejoy et al. 1984).
It is known that there are significant quantities of cocoa and coffee plantations in this area. At this stage it is unclear which class these crops fall under. The regular pattern of the feature in the bottom-left of the classified map suggests that this may be a cash crop plantation.
Clearly the protection status of the four reserves in the study area has contributed greatly to their protection from deforestation and fragmentation. The forest of Goin-Debe on the left of the study area and the Tai National Park on the right are very clearly delineated. Protected areas, when sufficiently supported by local education programs and political will, have been clearly shown to be of value to conservation.
Acknowledgments
The University of Maryland Global Land Cover Facility provided the Landsat imagery used in this study.
Appendix 1
Appendix 2
The following code was used to produce the forest/non-forest classification:
# set cell size for the model
#
SET CELLSIZE MIN;
#
# set window for the model
#
SET WINDOW UNION;
#
# set area of interest for the model
#
SET AOI NONE;
#
# declarations
#
Integer RASTER n1_tai_themes_simple FILE OLD NEAREST NEIGHBOR AOI NONE "/home/students04/djr58/tai/tai_themes_simple.img";
Binary RASTER n3_tai_forestfrags FILE OLD NEAREST NEIGHBOR AOI NONE "/home/students04/djr58/tai/tai_forestfrags.img";
#
# function definitions
#
n3_tai_forestfrags = EITHER 1 IF (
($n1_tai_themes_simple == 7) OR
($n1_tai_themes_simple == 6 AND SEARCH ( $n1_tai_themes_simple , 1 , 7) <= 1)
) OR 0 OTHERWISE;
QUIT;
References
Chatelain, C., L. Gautier, and R. Spichiger. 1996. A recent history of forest fragmentation in southwestern Ivory Coast. Biodiversity and Conservation 5:37-53.
Foody, G. M., R. M. Lucas, P. J. Curran, and M. Honzak. 1997. Mapping tropical forest fractional cover from coarse spatial resolution remote sensing imagery. Plant Ecology 131:143-154.
Lambin, E. F., and D. Ehrlich. 1997. The identification of tropical deforestation fronts at broad spatial scales. International Journal of Remote Sensing 18:3551-3568.
Lovejoy, T. E., J. M. Rankin, R. O. Bierregaard, K. S. Brown, L. H. Emmons, and M. van der Voort. 1984. Ecosystem decay of Amazon forest remnants. Pages 295-325 in M. H. Nitecki, editor. Extinctions. Univ. of Chicago Press.
Myers, N. 1990. The biodiversity challenge:expanded hotspot analysis. Environmentalist 10:243.
Riitters, K., and J. Wickham. 2000. Global-scale patterns of forest fragmentation. Conservation Ecology 4:27-56.
Wiens, J. A. 1989. Spatial scaling in ecology. Functional ecology 3:385-397.
