ForestNet: Classifying Drivers of Deforestation in Indonesia using Deep\n Learning on Satellite Imagery

Abstract

Characterizing the processes leading to deforestation is critical to the\ndevelopment and implementation of targeted forest conservation and management\npolicies. In this work, we develop a deep learning model called ForestNet to\nclassify the drivers of primary forest loss in Indonesia, a country with one of\nthe highest deforestation rates in the world. Using satellite imagery,\nForestNet identifies the direct drivers of deforestation in forest loss patches\nof any size. We curate a dataset of Landsat 8 satellite images of known forest\nloss events paired with driver annotations from expert interpreters. We use the\ndataset to train and validate the models and demonstrate that ForestNet\nsubstantially outperforms other standard driver classification approaches. In\norder to support future research on automated approaches to deforestation\ndriver classification, the dataset curated in this study is publicly available\nat https://stanfordmlgroup.github.io/projects/forestnet .\n