Analyzing heatwaves, droughts, and wildfires on the Iberian Peninsula: A comparison of data-driven and threshold-based methods for identifying extreme events

Abstract

Heatwaves, drought and wildfires are devastating extreme events, whose frequency and intensity are expected to increase under climate change. Studying these extreme events and better understanding their relationship is critical to best prepare for their impacts, and starts with their identification. Robust identification methods are crucial for the study of extreme events. However, traditional threshold-based approaches are sensitive to the threshold value and reference period used. Alternative data-driven methods exist, but their application for extreme event identification is in the early stages.

This study analyzes heatwaves, droughts and wildfires on the Iberian Peninsula, a hot spot for climate change, comparing traditional identification methods to the isolation forest method, a data driven approach. Heatwaves and droughts are identified using temperature and precipitation data from 2000 to 2022. The agreement between the two sets of methods is evaluated, and their differences are characterized. The predictive performance of each method is assessed using a Convolutional Neural Network (CNN) model predicting wildfire occurrences based on the identified heatwaves and droughts.

The results show that while both methods identify similar patterns of hot and dry extremes, the identified events show differences in durations and spatial extent, with more pronounced discrepancies for droughts than heatwaves. Despite these differences, both methods perform equally well in predicting wildfire occurrences using a CNN model.

The findings in this study provides valuable insights into the nature of heatwaves and droughts on the Iberian Peninsula, as well as demonstrate how the isolation forest method can be an effective alternative to traditional methods for identifying extreme events. However, further research is required to determine the underlying cause of the differences between the two methods.