Evaluating Heatwaves through Classical Threshold Metrics and AI-Driven Clustering: A Multi-Method Comparative Study in a Changing Climate

Heatwaves are intensifying as one of the deadliest climate hazards, necessitating accurate detection methods. This study compared three heatwave identification approaches for the extended warm season (June–September) ۱۹۹۳–۲۰۲۲ across four synoptic stations with different spatial distribution in Iran (Sanandaj, Iranshahr, Ahvaz, and Bojnurd): a multi-threshold percentile-based method, the Excess Heat Factor (EHF), and a hybrid K-means clustering technique. The multi-threshold approach, due to its very high percentiles and strict duration-continuity criteria, detected the fewest events (۵–۱۲) and cumulative days while severely underestimating actual heatwave burden. EHF identified physiologically significant hotter events (average maximum temperature ۴۷–۵۰ °C) but remained conservative in frequency and total duration (۱۹–۲۷ events). In contrast, the hybrid K-means clustering method exhibited the highest sensitivity and flexibility, detecting the largest number of events (۳۴–۴۹), the greatest cumulative heatwave days (۱۴۱–۲۳۶), and still very high average maximum temperatures (۳۷.۸ – ۴۹.۷ °C), thus providing the most comprehensive and realistic representation of heatwave occurrence in Iran. The results demonstrate that AI-driven clustering outperforms traditional threshold-based indices in accuracy, completeness, and adaptability for heatwave monitoring in a warming climate.

کلیدواژه ها:

Heatwave ، K-means clustering ، Excess Heat Factor (EHF) ، Multi-threshold method ، Iran climate