Mapping drivers of tropical forest loss with satellite image time series and machine learning

Abstract

Abstract The rates of tropical deforestation remain high, resulting in carbon emissions, biodiversity loss, and impacts on local communities. To design effective policies to tackle this, it is necessary to know what the drivers behind deforestation are. Since these drivers vary in space and time, producing accurate spatially explicit maps with regular temporal updates is essential. Drivers can be recognized from satellite imagery but the scale of tropical deforestation makes it unfeasible to do so manually. Machine learning opens up possibilities for automating and scaling up this process. In this study, we have developed and trained a deep learning model to classify the drivers of any forest loss - including deforestation as well as temporary disturbances - from satellite image time series. The results show that time series bring a significant improvement over using single images. We have designed the model architecture to allow understanding of how the model uses the input time series to make a prediction. We analyzed these data and showed how the model learns different patterns for recognizing each driver. Finally, we used our model to classify over $588'000$ sites of recent forest loss to produce a map detailing the driving forces behind forest loss across the tropics.