Project Description:
As climate change progresses, wildfires become a more prevalent issue. Different environmental conditions can affect how fires spread. This is an educational tool to showcase how wildfires spread under certain weather conditions. The wildfire simulator uses two complementary approaches: an image-based model that leverages satellite data (such as VIIRS bands and land cover maps) with a CNN and spatial transformer to generate next-day fire probability maps, and a non-image model that uses engineered weather and geographic features in a random forest regression to estimate total burned acreage. The two models come together in the format of an interactive website where the user can input weather conditions and draw a burn area. Then, the two models work to give an estimate of the progressing shape of the fire and the total acres burned.
