High resolution spatiotemporal modelling of outdoor temperature for assessment of heat-related health inequalities in Accra, Ghana

Urban heat exposure is unevenly distributed within cities, often disproportionately affecting socially and environmentally vulnerable populations. In Sub-Saharan Africa(SSA), where warming is occurring faster than the global average, the lack of high-resolution temperature data constrain efforts to characterize exposure patterns and inform epidemiological research and urban heat adaptation. We developed the first ground-based spatiotemporal outdoor temperature surfaces for a major SSA city using an 18-month measurement campaign across the Greater Accra Metropolitan Area (GAMA), Ghana. Temperature data were collected from 143 sites using iButton sensors. The field measurement data were combined with satellite-derived and land use predictors within a mixed-effects land use regression framework to model daily mean, maximum, and minimum temperature at 24-hour, daytime(6:00-17:59), and nighttime(18:00-5:59) periods. Models were developed using forward feature selection and evaluated through cross-validation and external validation using data from six fixed-site meteorological stations. Predicted temperature surfaces were generated at a 50x50m spatial resolution. Daytime temperatures were primarily predicted by solar geometry, whereas nighttime temperatures were more strongly influenced by land use characteristics and coastal proximity. Models performed well across temperature metrics and temporal periods with cross-validated R2 ranging from 0.73-0.87. Distinct seasonal patterns were observed, with higher temperatures in Harmattan(mean = 28.6 ○ C; maximum = 35.4 ○ C) and lower, less variable temperatures in non-Harmattan season(mean = 26.4 ○ C; maximum = 32.0 ○ C). Peri-urban areas were generally cooler than urban and commercial areas. Day-to-night cooling varied spatially and seasonally, with an average reduction in mean temperature of 4.6 ○ C. Out-of-sample predictions for 2024-2025 showed strong agreement with external meteorological station data(correlation: 0.69 – 0.94). These high-resolution temperature estimates provide an important resource for epidemiologic studies and can be integrated with health and sociodemographic data to better characterize heat-related vulnerabilities and inequalities. Our framework offers a scalable approach for outdoor heat assessment in rapidly urbanizing cities across SSA.