Seasonal variability of a satellite-derived Environmental Stress Index and road-network morphology in Makkah, Saudi Arabia

The process of rapid urbanization in arid megacities may result in heterogeneous thermal and atmospheric environmental conditions, which may be influenced by the road-network morphology. This study provides a spatially explicit framework to investigate the seasonal relationship between road-network morphology and environmental stress in Makkah, Saudi Arabia. Seasonal daytime land surface temperature (LST) derived from Moderate Resolution Imaging Spectroradiometer (MODIS) Terra MOD11A1.061 and Aqua MYD11A1.061, along with four Sentinel-5 Precursor (Sentinel-5P) atmospheric constituents, namely nitrogen dioxide (NO 2 ), carbon monoxide (CO), sulphur dioxide (SO 2 ) and formaldehyde (HCHO), were used for four seasons of 2024. The year of 2024 was selected based on the assessment of the representativeness of data of 2019–2025 and harmonized into a 1-km common analytical grid. Environmental Stress Index (ESI) was calculated based on the LST and pollutants data of four seasons of 2024 using the equal-weight approach where each of the five input variables has a weight of 0.20. The principal component analysis (PCA) was applied to diagnose the multivariate structure and assess its sensitivity rather than determining the ESI weights. The obtained index demonstrated similar coverage in all seasons with 1,454–1,458 valid grid cells and the seasonal standard deviation from 0.396 to 0.552. Graph-theoretical indices describing road-network morphology were assessed using seven regression algorithms and a blocked nested cross-validation scheme. Models considering only road networks lacked spatial generalizability, with pooled out-of-fold R 2 varying between − 0.25 and − 0.03 among road types. Inclusion of the spatial context calculated based on a regular grid resulted in an increase in predictive performance of the models for several road types, with pooled R 2 for Living Streets, Primary roads, Secondary roads, and Unclassified roads equal to 0.53, 0.46, 0.35 and 0.20 respectively. Predictive performance of the models remained heterogeneous among seasons, with both positive and negative predictive associations were observed across road types and seasons. SHapley Additive exPlanations (SHAP) analysis was applied to assess the importance of road-network variables included in the models, with an attempt to separate predictive associations from causal effects. The study covers a single arid city and one year, so the findings are case-specific and require testing in other cities.