ASPRS/ACSM (1994), copyright ASPRS/ACSM
Landsat-thematic mapper images recorded April 1986 and 1990 were obtained for northern Guatemala. NDVI image differencing and post classification change detection were performed. /Spatial characteristics of forest clearing, fallow clearing and regrowth patches were compared between two study sites. The results revealed that there were only half as many new forest clearings in 1990 along the Guatemala border as in the vicinity of Carmelita; they were, however, twice as large (x(mean) = 3.98 ha) as the forest clearings at Carmelita (x(mean) = 2.04 ha). The mean size of regrowth patches along the border were less than half the size of new clearings; however, the clearing and regrowth patches at Carmelita were approximately equal in size and number. The ratio of forest clearing to regrowth area was one to one at Carmelita but nearly 3 to 1 along the border. These results indicate that /Spatial characteristics of tropical forest clearing can be monitored by multidate Landsat imagery and suggest differences in socio-economic factors operating at the two study sites.
The clearing of tropical forests is a subject in the forefront of global environmental issues. Loss of biodiversity, migratory bird habitat, global climate change and sustainable development are just a few of the concerns related to tropical deforestation (Woodwell et al., 1983; Robbins et al., 1989; Wilson, 1988).
With so much media attention focused on deforestation in the Amazon, few people realize that the Central American region has one of the highest deforestation rates, on a percentage basis, in the world (FAO, 1991). Satellite imagery provides the primary source of quantifiable data about forest and land cover changes in tropical America. Several studies of deforestation and vegetation mapping have been conducted using a combination of remote sensing and geographic information system approaches (Sader and Joyce, 1988; Sader et al., 1990).
The Maya Biosphere Reserve was established in 1990 through an agreement between three neighboring countries, Mexico, Guatemala and Belize. The reserve represents the largest contiguous tract of tropical forest remaining in Central America. The area contains some of the most historically significant archaeological sites in Latin America. Much attention is now focused on the Biosphere Reserve as several international organizations and government agencies have developed cooperative projects to foster sustainable development and conservation plans. Unfortunately, this new international cooperation
[End Page 551]
and program development has not been able to eliminate deforestation that has been encroaching on the reserve boundaries (Stuart, 1992).
Sader et al. (in press) investigated forest clearing patterns for the northern Peten region of Guatemala. A 100 km^2 grid network was established to report the location of forest change for 141 grid cell in northeast Guatemala corresponding to the cloud-free Landsat TM coverage between 1986 and 1990. These authors reported that twenty-six percent of the 141 cells had no detectable forest clearing and 29% had minor clearing between 0.l and 0.5 percent over the four year period. Only 14 out of 141 cells indicated regrowth at 0.5% or greater over the four-year period. The percentage of regrowth did not exceed 1.2% in any grid cell. Regrowth, as described here, represents primarily recent fallow milpa (corn fields) in early successional stages composed of pioneer tree species, dense shrubs and vines. Interior roadless regions within the Maya Biosphere Reserve indicated little to no regrowth or clearing activity between 1986-90. There are few permanent residents in these areas. However, the southern portion of their study area represented the agricultural frontier where human population in-migration and associated land clearing was moving from south to north. This area experienced forest clearing rates greater than 4.0% (Sader et al., in press).
Remote sensing and GIS investigations that focus on forest change detection usually emphasize only forest clearing. Estimates of vegetation regrowth and land use conversion trends are rarely reported (Sader et al., 1991). Remotely sensed measurements at more than one time interval may be a useful tool for environmental monitoring of the tropical landscape where human impact may be exceeding the carrying capacity of the land. This paper will describe an interactive digital analysis approach for measuring /Spatial characteristics of tropical forest clearing, fallow clearing and vegetation regrowth using Landsat-TM imagery.
The difference in land use along the southern Mexico (state of Tabasco) and northwest Guatemala (Peten region) border is very distinct on satellite images. Land use on the Mexico side is mostly pasture and annual crops. Only remnants of disturbed forest remain. The Guatemala side of the border was almost totally forested in 1986; however, some deforestation has occurred within the last few years. Subsistence farmers (milperos) follow logging activity and newly cleared land is planted in annual crops such as corn or beans. Further east from the border region in the interior of the Peten, near the small town of Carmelita, shifting cultivation is practiced by local people. The Carmelita area has lower human populations and is more remote compared to the border site which is experiencing in-migration of people from other regions. The two study sites provide an interesting comparison of forest conversion patterns and human impact influenced by socio-economic factors (Figure 1).
[End Page 552]
The normalized difference vegetation index (NDVI) was computed for both dates and image differencing was performed by subtracting the 1990 NDVI from the 1986. The NDVI difference image was thresholded into three gray level ranges representing biomass increase (lower tail), no change and biomass decrease (upper tail). The thresholds were determined interactively by overlaying and comparing the NDVI difference values with 1986 and 1990 false color composites. Unsupervised clustering was performed on TM channel 3, 4 and 5 subsets for each study site, and based on these statistics, a maximum likelihood classification algorithm was employed.
Binary images were prepared of the NDVI biomass increase and biomass decrease pixels by recoding the threshold image. A majority filter (3x3) was passed through these images to reduce isolated pixels. Each binary mask was multiplied by the 1986 and 1990 land cover images to isolate the land cover types within the biomass increase and decrease zones. The 1986 and 1990 land cover classes were combined (recoded) to reveal the temporal (2 date) land cover classes in the "change" areas.
[End Page 553]
For an equal size area studied (approximately 23600 ha), there were over twice as many new forest clearings (greater than one hectare) around Carmelita compared to the Mexico/Guatemala border site (Table 1). However, the average size of a clearing along the border was approximately twice as large as the Carmelita clearings.
The ratio of forest clearing (deforestation) to fallow area clearing was more than four to one at the border site. Most of the fallow land clearing was, however, on the Mexican side of the border. The ratio at the Carmelita site was approximately three to one. The area-to-perimeter ratio of forest clearings was 0.42 along the border compared to 0.29 around Carmelita. The larger area-to-perimeter ratio indicates more angular clearings with less edge compared to Carmelita clearings. The difference in size and shape of clearings are apparent by visual inspection of the images. The larger clearings along the western border are likely the result of illegal logging (Canteo, 1993) and both Mexican and Guatemalan entrepreneurs growing corn for local commercial markets.
The Carmelita study area had more than twice as many regrowth patches compared to the border region (Table 2). The regrowth sites along the border were almost exclusively within Mexico because Guatemala's forest was nearly completely intact prior to 1986. The average size of the regenerating patches (forest fallows) was 2.28 ha at Carmelita compared to 1.46 ha along the border. The average size of new clearings (2.04 ha) at Carmelita were nearly the same as the regrowth patches (2.28 ha) indicating some consistency in the pattern of shifting cultivation.
Perhaps the most interesting and revealing observation is determined by calculating the ratio of new forest clearing to regrowth area for the 1986 to 1990 time frame. This ratio for the Carmelita site was one to one; however, the border site revealed nearly three times the area in new clearings compared to regrowth. These findings suggest that traditional agricultural practices (with relatively low human populations) around Carmelita were in relative balance in terms of new forest clearing compared to cultivated area abandoned or allowed to return to the fallow cycle.
The author visited the Carmelita site in March 1992 and found many 2-3 year old fallows located along the road at sites that were recent forest clearings in 1990. The fallow land was occupied by young successional forest, vines and shrubs.
The relationship between road access and forest clearing activity is particularly strong in the Peten region (both study sites) where there are
[End Page 554]
relatively few roads. Ninety-six percent of the clearings were within 2 km of the border or known roads compared to 81% within 2 km of roads around Carmelita. Almost 90% of the border regrowth sites were within 2 km of the border or a known road compared to 70% for Carmelita. However, 92% of Carmelita regrowth sites were within 3 km of a road. Foot travel or horseback was the normal mode of transportation for farmers at Carmelita. Many corn fields, accessible by an established system of old trails, were not visible on the satellite images and, therefore, were not digitized.
The change detection procedure described in this paper combined image differencing and post-classification change detection. The image difference thresholds selected to develop the biomass increase and decrease strata were set low to include probable commission error. False change classes were reduced through interactive processing and editing of temporal land cover classes. It is believed that omission error (changes due to forest clearing or regrowth not detected) would have been increased by setting the thresholds too high. Excessive omission error would result in underestimation of forest clearing and vegetation regrowth. The NDVI difference technique used alone is useful for stratification of major change categories but cannot effectively separate the difference between clearing of forest fallow and clearing of older forest. Post-classification change detection, within the biomass increase and decrease zones, allowed the distinction between forest and fallow clearing.
Multitemporal remote sensing can be a useful tool for assessing human land use patterns and agricultural intensification in tropical forest frontiers. Limited field observations and qualitative assessments of the Guatemala change detection images suggest that the procedures described provide reliable estimates of "deforestation" by taking into account the areas of fallow clearing. The observation of vegetation recovery and fallow land are rarely considered in digital satellite forest change detection studies.
The author would like to thank Stuart Gardner, Stacie Grove, and James C. Smoot for technical support at different stages in the project. The work was supported by McIntire-Stennis grant ME-0968. Maine Agricultural Experiment Station External Publication 1783.
Canteo, C. 1993. La destruccion de la Frontera Verde. Siglo Veintiuno, Jueves, April 29, 1pp.
Food and Agricultural Organization. 1991. Second Interim Report on the State of Tropical Forests. Forest Resources Assessment 1990 Project. United Nations FAO Rome, Italy. Unpublished, 2pp.
[End Page 555]
Robbins, C.S., J.R. Sauer, R.S. Greenberg, and S. Droege. 1989. Population declines North American birds that migrate to the tropics. In: Proc. of National Academy of Sciences 86(19):7658-62.
Sader, S.A. and A.T. Joyce. 1988. Deforestation rates and trends in Costa Rica, 1940-1983. Biotropica 20(1):11-19.
Sader, S.A., T.A. Stone, and A.T. Joyce. 1990. Remote sensing of tropical forests: an overview of research and applications using non-photographic sensors. Photogrammetric Engineering and Remote Sensing 55(10):1343-1351.
Sader, S.A., G.V.N. Powell, and J.H. Rappole. 1991. Migratory bird habitat monitoring through remote sensing. International Journal of Remote Sensing 12(3):363-372.
Sader, S.A., T. Sever, and J.C. Smoot. Forest change estimates for the northern Peten region of Guatemala - 1986 to 1990. Human Ecology (in press).
Stuart, G.E. 1992. Maya heartland under seige. National Geographic 10:94-107.
Wilson, E.O. 1988. Biodiversity. National Academy Press, Washington, D.C. 52lpp.
Woodwell, G.M., J.E. Hobie, R.A. Houghton, J.M. Mellilo, B.J. Peterson, G.R. Shaver, T.A. Stone, B. Moore, and A.B. Park. 1983. Deforestation measured by Landsat: steps toward a method. DOE/EVI0468-1. NTIS, Springfield, Virginia. 62pp.
[End Page 556]
[End Page 557]