University of Twente Student Theses


Spatial Modelling and Prediction of Tropical Forest Conversion in the Isiboro Sécure National Park and Indigenous Territory (TIPNIS), Bolivia

Sanabria Siles, Nelson Jery (2009) Spatial Modelling and Prediction of Tropical Forest Conversion in the Isiboro Sécure National Park and Indigenous Territory (TIPNIS), Bolivia.

[img] PDF
Abstract:Forest conversion is occurring in the Isiboro Sécure National Park and Indigenous Territory (TIPNIS), Bolivia. Activities such as agricultural encroachment and forest extraction are leading to a rapid loss of primary forest and also have disturbed the traditional life of the indigenous communities. The present study has integrated the statistical approach of logistic regression and also that of artificial neural networks with GIS in an attempt to analyze and predict forest conversion in the TIPNIS. Based on information obtained from land cover maps and satellite images, forest loss for the years 1976, 1986, 1991, 2001, 2004 and 2006 were calculated. According to the results of the study, during the period 1976 – 2006, 23% of primary forest has been lost in the southern part of the TIPNIS. The deforestation rates presented variations, they rose and fell and then rose again. The rates of deforestation were 0.005%.a −1 until 1986, 1.3%.a −1 until 1991, 0.5%.a−1 until 2001, and 2.3%.a−1 until 2004 and 3.5%.a−1 until 2006. This study revealed that the variations of deforestation rates in the TIPNIS coincide with the degree of control of coca (Erythroxylum coca) cultivation that the Bolivian government has permitted in the Chapare Province. When government controls of coca growing were more lax, the deforestation rates increased. To model forest conversion this study considered the change that has occurred in the forest areas as a categorical dependent variable. The univariate tests of association Cramer’s V was used to test five potential explanatory variables for forest conversion (“Distance from Forest Edge”, “Distance from Roads”, “Distance from Settlements”, “Landscape Position” and “Type of Settlement”). “Type of Settlement” was excluded from the modelling because the data input (map) was too coarse. Logistic regression analysis was used (i) to assess the relative significance of explanatory variables on forest change during the period 2001-2004; and (ii) to predict probability of forest change for the period 2004-2006. “Landscape Position” was the most significant explanatory variable, followed by the explanatory variables “Distance from Forest Edge”, “Distance from Roads”, and “Distance from Settlements”. Logistic regression prediction resulted in an Area Under a ROC Curve (AUC) of 85%. Finally, the study made use of the artificial neural network Multi-Layer Perceptron (MLP) to improve the prediction of probability of forest change for the period 2004 to 2006. The prediction performed used the same data set used by the logistic regression prediction. The AUC obtained by MLP was 92%. The predictive performance of both models proved successful. While MLP produces better prediction results in general, logistic regression analysis is still needed to understand the relative significance of the explanatory variables on the forest change.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page