Capturing Unseen Spatial Heat Extremes Through Dependence-Aware Generative Modeling

TL;DR

This paper introduces DeepX-GAN, a generative model that captures spatial dependence of climate extremes, enabling simulation of unseen events beyond historical records to improve risk assessment.

Contribution

The paper presents a novel dependence-aware generative model that can simulate plausible unseen climate extremes, validated against climate model simulations.

Findings

DeepX-GAN effectively captures spatial structure of rare extremes.

Unseen heat extremes pose significant risks to vulnerable regions.

Future warming will increase and shift heat extreme hotspots.

Abstract

Observed records of climate extremes provide an incomplete view of risk, missing "unseen" events beyond historical experience. Ignoring spatial dependence further underestimates hazards striking multiple locations simultaneously. We introduce DeepX-GAN (Dependence-Enhanced Embedding for Physical eXtremes - Generative Adversarial Network), a deep generative model that explicitly captures the spatial structure of rare extremes. Its zero-shot generalizability enables simulation of statistically plausible extremes beyond the observed record, validated against long climate model large-ensemble simulations. We define two unseen types: direct-hit extremes that affect the target and near-miss extremes that narrowly miss. These unrealized events reveal hidden risks and can either prompt proactive adaptation or reinforce a sense of false resilience. Applying DeepX-GAN to the Middle East and North Africa shows that unseen heat extremes disproportionately threaten countries with high vulnerability and low socioeconomic readiness. Future warming is projected to expand and shift these extremes, creating persistent hotspots in Northwest Africa and the Arabian Peninsula, and new hotspots in Central Africa, necessitating spatially adaptive risk planning.