An innovative long short-term memory-based algorithm for constructing vegetation health maps over Ethiopia

Rehaan Ahmad, Brian Yang, California, USA 16-18

This study develops a new method for forecasting vegetation health in Ethiopia. Vegetation health is measured by satellites through an index known as Normalized Difference Vegetation Index (NDVI). Existing methods are only able to forecast 1 pixel (1 sq. km) regions, thus making it difficult to scale the solution over larger regions. In this project, we introduce the Annual Gate ConvLSTM, improving upon previous NDVI forecasting methods to construct NDVI forecasts of regions in sub-Saharan Africa. This network is able to process 10,000 sq. km and produce a 100 pixel by 100 pixel image with state-of-the-art accuracies.

Show more