The summer of 2024 has seen India gutted by forest fires in Uttarakhand, Himachal Pradesh, and Jammu and Kashmir. Wildfires are a growing problem for the country, with climate change likely to boost fire activity.
Researchers have now developed a machine learning algorithm to better predict when and where those fires could spark, potentially bolstering district-level disaster preparedness.
The model analyzed patterns across forests in the states of Chhattisgarh and Madhya Pradesh (MP), an area spanning more than 150,000 square kilometers (57,915 square miles).
To map fire behavior between 2001 and 2020, the algorithm collected input from 26,000 instances of forest fire and incorporated it with data from climatic, topographical, and ecological parameters such as temperature, precipitation, soil moisture, and plant litter. These data sets were then overlaid across the forested land to accurately estimate the probability of forests at risk.
“The random forest model hierarchizes the parameters and assigns them a number based on how important they are for the prediction,” said Sumedha Surbhi Singh, a doctoral student at Birla Institute of Technology (BIT) Mesra and coauthor of the study. By thus quantifying the impact of terrain parameters on the forest fire occurrences, the new model was able to predict the likelihood of fire occurrence in a location.
“A combination of high temperature and low rainfall instigated a higher number of fire events.”
Singh said rainfall and the normalized difference vegetation index emerged as the most important factors when it came to predicting fire occurrence. “A combination of high temperature and low rainfall instigated a higher number of fire events in both states,” the authors mentioned.
The model, said Singh and her colleagues, validated their hypotheses, but only to an extent. “In some of the areas facing less rainfall,” she explained, “the number of forest fires…was higher. But when we studied Chhattisgarh, which gets higher rainfall than MP, the ratio of forest fires was higher.”
The authors believe the discrepancy may be due to increased human intervention in the form of local communities collecting forest products such as food and medicinal plants through traditional methods, which involve burning. “They clear out patches of underlying grass and burn the place. If the rainfall is low and the temperature is high, then there is a higher risk of a larger forest area burning,” Singh explained.
“We can predict nature depending on the rainfall, temperature, soil moisture, etc., but human behavior cannot be predicted,” said C. Jeganathan, a study coauthor and dean of the Department of Remote Sensing, BIT Mesra.
More Frequent Fires
“With increasing temperatures, the incidents of forest fire are also increasing.”
Efficient prediction models are crucial for addressing the growing vulnerability of India’s forested landscape.
“With increasing temperatures, the incidents of forest fire are also increasing,” said Ayushman Malakar, a senior research fellow at the Indian Council of Forestry Research and Education who is not associated with the new research.
The new model’s ability to incorporate localized characteristics may make it particularly valuable, said Jeganathan. Researchers are confident that the model can be trained and deployed across various forest patches such as the Western Ghats and the Himalayas with region-specific data.
“We can train this model for other regions as well, where it can be part of the early warning systems to ascertain vulnerable areas and prioritize the deployment of more firefighters in the sensitive region,” said Singh.
—Pragathi Ravi (@pragathi_r24), Science Writer
Citation: Ravi, P. (2024), New model can better predict areas vulnerable to forest fires in India, Eos, 105, https://doi.org/10.1029/2024EO240307. Published on 24 July 2024.
Text © 2024. The authors. CC BY-NC-ND 3.0Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.
Related