Mapping Oil Palm Expansion and Environmental Impacts in the Eastern Amazon Using Machine Learning and Satellite Imagery

Abstract

This paper maps palm oil plantations in the Eastern Amazon, Brazil's largest producer, for the years 2014, 2017, and 2020, using machine learning algorithms to understand recent expansion dynamics. To achieve this, we combined optical spectral bands from Landsat-8, radar backscatter values from Sentinel-1, vegetation, and texture indices, and a linear spectral mixing model. The Random Forest algorithm achieved the best classification accuracy, with overall accuracies of 94.53%, 94.28%, and 95.53%, and Kappa coefficients of 0.9075, 0.9031, and 0.9239 for 2014, 2017, and 2020, respectively. Through land use and land cover transition analysis, we identified a significant expansion of oil palm in the region, with the area increasing from 1,074 km² to 1,849 km² between 2014 and 2020—a growth of 72.16%. However, approximately 156.88 km² (20.24%) of this expansion occurred directly over vegetation cover, raising concerns about its environmental sustainability. The mapping and pixel-level analysis of land use and land cover changes presented in this paper are instrumental in identifying areas of high environmental risk. This analysis provides crucial insights that can inform the development and monitoring of public policies aimed at regulating the expansion of oil palm plantations, ensuring that such growth occurs in a sustainable and environmentally responsible manner.

Keywords: Oil Palm, Deforestation, Machine Learning, Remote Sensing.