A deep learning approach to identify smoke plumes in satellite imagery
On this page:
This research was completed during the authors' previous position at Duke University prior to August 2020. Exposure to pollution from wildland fires (wildfires; bushfires) contributes significantly to poor air quality, a leading risk factor for premature death. The frequency and intensity of wildfires are expected to increase, making it vital to improve tools for estimating exposure to fire smoke. New generation satellite-based sensors produce spectral data at high spatial and temporal resolution that plays an increasingly important role in providing real-time information of surface features during wildfire episodes. Because of the vast quantities of data, new generation satellite data products require new automated methods for processing information. We present a deep fully convolutional neural network (FCN) capable of predicting fire smoke presence in high-resolution satellite imagery in near real time (NRT). The FCN identifies smoke plumes by referring to the output of currently operational methods for smoke identification as the training data. In this way, our approach leverages validated smoke identification products in a format that can be operationalized in NRT. We demonstrate this concept for a fire episode in the Northern Territory of Australia, but the algorithm is applicable to any geographic region. Our results indicate the algorithm has high classification accuracy (99.5\% of pixels correctly classified on average) and precision (average intersection over union = 57.6\%). These results demonstrate that the FCN algorithm has high potential as an NRT exposure assessment tool, capable of providing critical information to fire managers, health and environmental agencies and the general public to prevent the health risks associated with exposure to hazardous smoke from wildland fires.