Gradient boosting with extreme-value theory for wildfire prediction

TL;DR

This paper presents a novel gradient boosting method incorporating extreme-value theory for wildfire prediction, achieving competitive results and emphasizing the importance of spatial cross-validation for accurate performance assessment.

Contribution

It introduces a theoretically justified loss function based on extreme-value theory into gradient boosting for wildfire prediction, with a new spatial cross-validation scheme.

Findings

Achieved second place in the 2021 Extreme Value Analysis data challenge.

The proposed method performs competitively against other boosting approaches.

Spatial cross-validation improves performance estimation accuracy.

Abstract

This paper details the approach of the team $\textit{Kohrrelation}$ in the 2021 Extreme Value Analysis data challenge, dealing with the prediction of wildfire counts and sizes over the contiguous US. Our approach uses ideas from extreme-value theory in a machine learning context with theoretically justified loss functions for gradient boosting. We devise a spatial cross-validation scheme and show that in our setting it provides a better proxy for test set performance than naive cross-validation. The predictions are benchmarked against boosting approaches with different loss functions, and perform competitively in terms of the score criterion, finally placing second in the competition ranking.