Explainable machine learning analysis of urban heat island in Türkiye: SHAP-based meteorological threshold assessment (1965–2024)

The urban heat island effect (UHI) is a major environmental phenomenon that arises with urbanization, leading to higher temperatures in urban areas than rural areas. This study investigates the long-term variation of the UHI effect in seven major cities (Istanbul, Bursa, Ankara, Izmir, Gaziantep, Mersin, and Kayseri) located in different climate zones of Türkiye from 1965 to 2024. Urban and rural meteorological stations were paired in each city. Hourly and daily temperature data, along with meteorological data (Cloud cover, wind speed, sunshine duration, and precipitation), were obtained from these stations. Also, long-term trends were statistically evaluated using the Mann-Kendall and Sen's slope trend test. The results show that statistically significant (p < 0.05) increasing trends are observed at 15 of 18 stations. Strong increasing trends were detected, particularly in Florya (Z = 6.92, Sen slope = 0.0445 °C/year), Bursa (Z = 5.76, Sen slope = 0.0361 °C/year), and Mersin (Z = 9.32, Sen slope = 0.0871 °C/year). Sequential Mann-Kendall analyses showed that the significant increase in UHI intensity began in the mid-1990s. The SHapley Additive exPlanations (SHAP) methodology was applied using the Random Forest (RF) model. SHAP analyses showed that sunshine duration had the strongest positive effect on UHI, while wind speed had a negative effect. Finally, based on meteorological threshold values, the average UHI intensity was 1.80 ± 2.22 °C under calm conditions and 1.96 ± 2.14 °C under windy conditions. The increase in Türkiye's urban population rate from 24% in 1927 to 93% in 2024 is a critical factor in the intensification of the UHI.