New Algorithms for Detecting Forest Fires on a Global Scale 1 From MODIS Time Series Analysis

Mapping forest fires globally is an important task for supporting climate and 18 carbon cycle studies. There are two primary approaches to fire mapping: field-and aerial-based 19 surveys, which are costly and limited in their extent; and satellite remote sensing-based 20 approaches, which are more cost-effective but pose several interesting methodological and 21 algorithmic challenges. In this paper, we describe evaluate a new algorithm framework for 22 mapping forest fires based on satellite observations from NASA's Moderate Resolution Imaging 23 Spectroradiometer (MODIS) instrument. A systematic comparison and validation against ground 24 truth sources with alternate approaches across diverse geographic regions demonstrates that our 25 algorithmic paradigm is able to overcome many of the limitations in both data and methods 26 employed by prior efforts. We quantitatively show that the new framework out-performs the 27 well-known MODIS Burned Area (BA) framework in the states of California (US), Georgia 28 (US), Yukon, (Canada), and Victoria (Australia). Results demonstrate that our new framework is 29 highly robust to noise in one of its primary inputs, MODIS Active Fires (AF), which is known to 30 have low precision. 31 32 33