WildFire Modelling with Satellite Image Stream
As wildfires are expected to become more frequent and severe, improved prediction models are vital to mitigating risk and allocating resources. With remote sensing data, valuable spatiotemporal statistical models can be created and used for resource management practices. In this paper, we create a dynamic model for future wildfire predictions of five locations within the western United States through a deep neural network via historical burned area and climate data. The proposed model has distinct features that address the characteristic need in prediction evaluations, including dynamic online estimation and time-series modeling. Between locations, local fire event triggers are not isolated, and there are confounding factors when local data is analyzed due to incomplete state observations. When compared to existing approaches that do not account for incomplete state observation within wildfire time-series data, on average, we are able to achieve higher prediction performances.
Dataset from the ten areas.
Selected rectangular grids map for data set generation from FIRMS. The rectangles are centered at Portland, Medford, Reno, Denver, Salt Lake City, etc.
Comupational Networks
Dynamic auto-encoder and fire map prediction network. The dynamic auto-encoder is trained to predict the observation at k+1 using the observation at k and the state of RNN. And the state of RNN is also used to predict fire map at T time steps ahead.
Online Prediction After Training
Prediction of fire risk for selected seven train datasets (seven columns). And the top row shows the ground truth fire gridmap at the current time. The other rows show the predicted fire grid map. The 2nd row is the prediction for k+1. The 3rd row is the prediction for k+2 and so on. In this prediction, the step size is a week. The fire looks spreading from below to the top because the predictions are working.
Relevant Paper:
- Hyung-Jin Yoon and Petros Voulgaris. “Multi-time Predictions of Wildfire Grid Map using Remote Sensing Local Data,” IEEE International Conference on Knowledge Graph (2022 ICKG)