A Machine Learning Approach for Assessing Wildfire Vulnerability in the Colorado River Basin
Abstract
Wildfire is becoming more frequent and severe in both the western United States and globally. Accurate prediction of wildfires would be beneficial because of its impact on the environment that crosses multiple disciplines such as agricultural systems, hydrology, and environmental studies. Therefore, it is essential to identify key drivers of wildfires and predict when and where wildfires are likely to occur. With the advancement of computational algorithms, machine learning has proven sufficient potential in dealing with challenging data-driven problems that consider a complex relationship between a large amount of information. Thus, this study builds a machine learning-based approach for identifying key drivers of wildfires and assessing the vulnerability of wildfires in the Colorado River basin. Various predictors of local hydrological and meteorological variables and topographic characteristics from remote sensing-based observation systems were evaluated to predict wildfire vulnerability. The results are validated with historical wildfire events using various evaluation metrics: overall classification accuracy, precision, and confusion matrix values. It is expected that the machine learning-based approach has sufficient power for wildfire prediction, mitigation, and management. In addition, the wildfire vulnerability map produced by the proposed approach can be valuable information for developing effective strategies for wildfire risk management in the Colorado River basin.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMNH35F..01H