Title

Simulating national-scale deforestation in the Philippines using land cover change models

Document Type

Article

Publication Date

12-2019

Abstract

After the year 2010, a century of rapid decline in Philippine forest cover was reversed due to a deliberate National Greening Program (NGP). Drivers that can sustain or counter this increase can be better investigated through the help of land change models. However, such models are not yet in mainstream use for planning in the Philippines. Hence, this study used two models – FOREST-SAGE and GEOMOD – to simulate forest conditions based on anthropogenic drivers and to evaluate model applicability in the Philippines. The performance of each model was assessed using the root-meansquare error (RMSE), mean absolute error (MAE), Kappa, national average tree cover percentage, and common deforestation hotspots with reference datasets. Validation with Climate Change Initiative Land Cover dataset (CCI-LC) yielded similar results between the models: 2015 tree cover maps with RMSE of 22–25% tree cover, MAE of 10–12% tree cover, and moderate agreement with reference map based on Kappa (0.4–0.6); and 2010–2015 change maps with RMSE of 8–9% tree cover, MAE of 1–2% tree cover, and agreement due to chance based on Kappa (0.01–0.03). Validation with MODerate Resolution Imaging Spectroradiometer-Vegetation Continuous Field (MODIS-VCF) maps yielded similar MAE results (10% tree cover) between the models. Validating FOREST-SAGE end-time and change maps with MODIS-VCF yielded better RMSE results than GEOMOD (RMSE of 13% tree cover for FOREST-SAGE; 22% tree cover for GEOMOD). However, GEOMOD tree cover maps yielded better Kappa than FOREST-SAGE (0.60 for GEOMOD; 0.01 for FOREST-SAGE). Results suggest that FOREST-SAGE is more applicable in the Philippines for provincial extent studies that aim to quantitively track forest cover change, while GEOMOD is more applicable for national extent studies that use categorical data. Results also suggest that input parameter settings must be improved to simulate spatial distribution of forest cover.

Share

COinS