TS-SatFire: A Multi-Task Satellite Image Time-Series Dataset for Wildfire Detection and Prediction

TL;DR

This work presents a comprehensive multi-temporal remote sensing dataset for active fire detection, daily wildfire monitoring, and next-day wildfire prediction, covering wildfire events in the contiguous U.S. from January 2017 to October 2021.

Abstract

Wildfire monitoring and prediction are essential for understanding wildfire behaviour. With extensive Earth observation data, these tasks can be integrated and enhanced through multi-task deep learning models. We present a comprehensive multi-temporal remote sensing dataset for active fire detection, daily wildfire monitoring, and next-day wildfire prediction. Covering wildfire events in the contiguous U.S. from January 2017 to October 2021, the dataset includes 3552 surface reflectance images and auxiliary data such as weather, topography, land cover, and fuel information, totalling 71 GB. The lifecycle of each wildfire is documented, with labels for active fires (AF) and burned areas (BA), supported by manual quality assurance of AF and BA test labels. The dataset supports three tasks: a) active fire detection, b) daily burned area mapping, and c) wildfire progression prediction. Detection tasks use pixel-wise classification of multi-spectral, multi-temporal images, while prediction tasks integrate satellite and auxiliary data to model fire dynamics. This dataset and its benchmarks provide a foundation for advancing wildfire research using deep learning.

Figures (8)

• Figure 1: Major components of the TS-SatFire dataset and three distinct tasks.
• Figure 2: Overview of locations and land covers of fires in an example split into train/test/validation set. The legend in the test set does not cover any fires.
• Figure 3: Percentage of missing values: Rates of missing values of each spectral band used in active fire detection, burned area mapping and fire progression prediction tasks. All the missing values are replaced with zeros during training and testing.
• Figure 4: Feature Importance of the input bands for BA task (SwinUNETR-3D) and AF task (T4Fire).
• Figure 5: Feature importance for fire prediction task: We investigate a simple measure of feature importance, by setting one feature to zero in all inputs and measuring the resulting test F1 score. Removing important features should reduce the performance while removing unimportant features should have little influence.