Performance of logistic regression model for predicting deforestation, case study: cikepuh wildlife reserve and cibanteng natural reserve
Abstract
Cikepuh Wildlife Reserve and Cibanteng Natural Reserve, since both conservation area was established in 1973 and 1925 have been facing complex problem caused by land use changed, deforestation, illegal hunting, forest fire, and so on. Deforestation itself is a complex socio-economic, cultural, and political event. This thesis focused on what factors affect the rate of deforestation by considering some common driving forces of deforestation and using logistic regression for predicting deforestation. It is clearly important to know where deforestation is likely to occur. The objectives of the thesis are to quantify the contribution of each deforestation driving factor such as distance from center of dweller, aspect, slope, distance from shore line, distance from existing road, and elevation, and to elaborate spatial projection of future trends of deforestation based on possibility of deforestation as the result of logistic regression equation. The methodology is using Stacking Method from CI (Conservation International) CABS (Center for Biodiversity Applied Science) and developed together with WCS IP (Wildlife Conservation Society – Indonesia Program). Two image with different dates or one period was stacked and analyzed by visualization from both images. Signature area was extracted from the stacked-images by using shapefile polygon for forest to forest class, forest to non forest class, non to non forest class, water, cloud and shadow. Signature area should be represented certain spectral characteristic, so for obtaining number of class as many as possible, it could use 16 bit data type indeed 8 bits. Classification method is supervised classification that was done by CART ERDAS Imagine plug in tool and See5, a stand alone decision-tree based classification program. The result of classification is thematic raster image with forest change attribute. Analysis was done in one attribute table of polygon vector cell (PVC), that is created by using Edit Tool Vector Grid, an extension from ArcView 3.3. All attribute of independent variables fill the squared-shaped polygon as called PVC, and the result probability of logistic regression as the result of the calculation as well. Independent variable is divided to two binary category 0 and 1. 1 is a parameter that tends to occur deforestation such as less 1 km distance from road. 0 is stable condition that there is no change from forest to non forest. The result of possibility deforestation occurrence is if the road distance less than 1 km, tends to deforested occurrence 3 times compare the distance greater or equal 1 km. The smallest possibility of deforestation occurrence was contributed by predictor distance 1 km from river, and almost has no effect to deforested occurrence. Regression logistic equation in this thesis can predict deforestation significantly, although some processes of polygon vector cell could not accommodated to assign data from attribute of independent variables to polygon vector cell exactly. Regression logistic model could predict deforestation better if distribution of independent variables that are assumed to tend to deforestation occurrence distribute evenly entire the study area.
Collections
- MT - Forestry [1373]