Evaluating the relationships between wildfires and drought using machine learning

Abstract

In recent years, many destructive wildfires have plagued California. Extreme fire conditions, such as drought, have been taking place simultaneously with many of these wildfires. In this study, the relationship was quantified between the self-calibrated Palmer Drought Severity Index (sc-PDSI) and wildfire burn area (BA) in California during the time of 1984–2018, and results indicate that the drought is a significant driver of wildfire BA in California. The methods of wavelet transform coherence, cross wavelet transform, and continuous wavelet transform were used in conjunction with machine learning algorithms to analyse and establish the relationship between sc-PDSI and wildfire BA. This study concludes that there was a statistically significant relationship between wildfire BA and sc-PDSI in 6–8-, 5–6-, and 2–3-year bands during the study period, during which sc-PDSI was one of the main drivers for wildfire BA. In addition, machine learning was utilised in conjunction with the Quantile Regression Model (QRM) in order to quantify the relationship between sc-PDSI and wildfire BA in California. The findings provide a promising direction to improved prediction of wildfire BA which is significant in the aid of damage control of wildfires in California, potentially leading to less burned area, less economic damage, and fewer casualties.