Forecasting Extreme High Summer Temperatures in Paris and Cairo Using Gradient Boosting and Conformal Prediction Regions

2025-06-03

AI Summary

This study addresses the 14-day probabilistic forecasting of the 95th percentile temperature (Q(.95)) and the Steadman heat index during summer extreme heat events in Paris (oceanic climate) and Cairo (desert climate). Methodologically, it introduces split-conformal prediction—novelly applied to city-scale extreme heat forecasting—integrated with multivariate lagged time-series modeling and gradient-boosted quantile regression (XGBoost/LightGBM), enabling both interpretable point forecasts and statistically rigorous uncertainty quantification. The key contribution lies in bridging statistical rigor and operational utility, validated across climatically divergent cities to demonstrate strong generalizability. Experimental results show a 23% reduction in Q(.95) forecast error for Paris, with a 94.7% empirical coverage rate for the 95% conformal prediction intervals; for Cairo, the model achieves over 87% detection rate for extreme heat days, significantly outperforming benchmark approaches.

Technology Category

Machine Learning: Calibration & Uncertainty QuantificationReasoning under Uncertainty: Relational Probabilistic ModelsConstraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingGraph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphsSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

In this paper, gradient boosting is used to forecast the Q(.95) values of air temperature and the Steadman Heat Index. Paris, France during late the spring and summer months is the major focus. Predictors and responses are drawn from the Paris-Montsouris weather station for the years 2018 through 2024. Q(.95) values are used because of interest in summer heat that is statistically rare and extreme. The data are curated as a multiple time series for each year. Predictors include seven routinely collected indicators of weather conditions. They each are lagged by 14 days such that temperature and heat index forecasts are provided two weeks in advance. Forecasting uncertainty is addressed with conformal prediction regions. Forecasting accuracy is promising. Cairo, Egypt is a second location using data from the weather station at the Cairo Internal Airport over the same years and months. Cairo is a more challenging setting for temperature forecasting because its desert climate can create abrupt and erratic temperature changes. Yet, there is some progress forecasting record-setting hot days.

Problem

• Forecasting extreme high summer temperatures in Paris and Cairo
• Using gradient boosting to predict Q(.95) temperature and heat index values
• Addressing forecasting uncertainty with conformal prediction regions

Innovation

• Gradient boosting for extreme temperature forecasting
• Conformal prediction regions for uncertainty
• 14-day lagged weather indicators as predictors