With rigorous economic research and practical policy solutions, we focus on the issues and institutions that are critical to global development. Explore our core themes and topics to learn more about our work.
In timely and incisive analysis, our experts parse the latest development news and devise practical solutions to new and emerging challenges. Our events convene the top thinkers and doers in global development.
The FORMA filter is technically a restricted and semi-trained learning algorithm, which relies on thermal anomaly and NDVI time-series data from MODIS for each pixel. The analysis proceeds in three broad stages: (1) At the pixel level, we collect relevant indicators from the time-series data. This includes both short- and long-term drops in the NDVI, as well as count and persistence of forest fires. (2) We then “train” the filter based on the work of Hansen, et al. (2008) to statistically match observed forest clearing with its proximate determinants. (3) Finally, we use the estimated parameters from the second stage to appropriately weight the streaming data on NDVI and fires toward likely forest clearing. That is, we use information on how the signals (NDVI drops, fires, etc.) have historically been matched with forest clearing to weight current, streaming signals.