Advancing Wildfire Risk Prediction via Morphology-Aware Curriculum Contrastive Learning

TL;DR

This paper addresses the challenge of predicting wildfire risk from imbalanced, high-dimensional spatio-temporal data by introducing a morphology-aware curriculum contrastive learning (CL) framework. It combines a two-stage training pipeline with two sampling strategies—historical and curriculum sampling—that leverage temporal dynamics and morphological similarity to produce more informative dynamic-feature representations, while allowing the use of smaller patch sizes to reduce computation. The approach integrates a standard cross-entropy loss with a contrastive term, either via triplet loss or supervised contrastive loss, and explores two training paradigms (fine-tuning vs end-to-end CL). Experiments on Greece (FireCube) and Calabria demonstrate that curriculum sampling yields the strongest performance gains, with substantial computational savings and robust generalization across patch scales, suggesting practical value for timely wildfire risk forecasts and resource allocation.

Abstract

Wildfires significantly impact natural ecosystems and human health, leading to biodiversity loss, increased hydrogeological risks, and elevated emissions of toxic substances. Climate change exacerbates these effects, particularly in regions with rising temperatures and prolonged dry periods, such as the Mediterranean. This requires the development of advanced risk management strategies that utilize state-of-the-art technologies. However, in this context, the data show a bias toward an imbalanced setting, where the incidence of wildfire events is significantly lower than typical situations. This imbalance, coupled with the inherent complexity of high-dimensional spatio-temporal data, poses significant challenges for training deep learning architectures. Moreover, since precise wildfire predictions depend mainly on weather data, finding a way to reduce computational costs to enable more frequent updates using the latest weather forecasts would be beneficial. This paper investigates how adopting a contrastive framework can address these challenges through enhanced latent representations for the patch's dynamic features. We thus introduce a new morphology-based curriculum contrastive learning that mitigates issues associated with diverse regional characteristics and enables the use of smaller patch sizes without compromising performance. An experimental analysis is performed to validate the effectiveness of the proposed modeling strategies.

Figures (5)

• Figure 1: Wildfire burnt areas, accoring to the datasets used in the experimentation and referring to the Greece and the southern Italy.
• Figure 2: Box plot of the dynamic features from the balanced version of the Greece dataset.
• Figure 3: Box plot of the dynamic features from the Calabria dataset.
• Figure 4: Illustration of the three sampling strategies used in this work.
• Figure 5: Illustration of the two training strategies used in this work.