A simulation study comparing statistical approaches for estimating extreme quantile regression with an application to forecasting of fire risk

TL;DR

This paper evaluates various statistical methods for estimating extreme quantile regression, using simulations and real fire risk data, to improve forecasting accuracy of rare but impactful events like wildfires.

Contribution

It introduces a simulation framework for comparing extreme quantile regression methods and advocates for a hybrid approach to enhance fire risk forecasting.

Findings

Certain methods outperform others in extreme quantile estimation

A positive correlation between temperature and fire frequency is confirmed

Hybrid strategies improve forecast accuracy and interpretability

Abstract

This simulation study compares statistical approaches for estimating extreme quantile regression, with a specific application to fire risk forecasting. A simulation-based framework is designed to evaluate the effectiveness of different methods in capturing extreme dependence structures and accurately predicting extreme quantiles. These approaches are applied to fire occurrence data from the Fez-Meknes region, where a positive relationship is observed between increasing maximum temperatures and fire frequency. The study highlights the comparative performance of each technique and advocates for a hybrid strategy that combines their complementary strengths to enhance both the accuracy and interpretability of forecasts for extreme events.