Canadian Forest Service Publications

A Machine Learning Approach to Waterbody Segmentation in Thermal Infrared Imagery in Support of Tactical Wildfire Mapping. Oliver, J.A., Pivot, F.C., Tan, Q., Cantin, A.S., Wooster, M.J., Johnston, J.M., Remote Sens. (2022) 14, 2262.

Year: 2022

Issued by: Great Lakes Forestry Centre

Catalog ID: 40707

Language: English

Availability: PDF (download)

Available from the Journal's Web site.
DOI: https://doi.org/10.3390/ rs14092262

† This site may require a fee

Mark record

Plain Language Summary

When we map wildfires with infrared (IR) imagery using aircraft overnight flights, water bodies are actually warmer than the land. This causes issues when trying to detect smoldering combustion and requires manual screening of the data by a human. The objective of this study was to detect and map water bodies in over night airborne IR data using artificial intelligence to improve the accuracy and efficiency of tactical mapping operations. In this study we used airborne IR data collected by CFS and UK researchers in NW Ontario of wildfires, and trained a series of AI models to recognize water bodies and map them out. We found several approaches reliable, with a preference for the random forest model. The findings are significant in terms of refining the quality, efficiency, and human resources of the Torchlight service. There are potentially other implications with regards to use of IR for precision mapping of water bodies for other applications (e.g. water levels and/or flood mapping).