From Urban Heat to Climate‐Resilient Planning: An Interpretable Machine Learning Assessment of Green Infrastructure Priorities in Ikorodu, Lagos, Nigeria

Rapid peri-urban expansion in Ikorodu Local Government Area (LGA), Lagos State, Nigeria, has intensified local surface thermal conditions. Yet, the area lacks a dedicated, quantitatively grounded framework linking urban heat to green infrastructure (GI) planning. This study analyses spatio-temporal Surface Urban Heat Island (SUHI) dynamics in Ikorodu LGA across four epochs (1990, 2002, 2013, and 2025) using multi-temporal Landsat imagery (TM, ETM+, OLI/TIRS, OLI-2/TIRS-2) processed in Google Earth Engine, ArcGIS Pro, and Python, and translates the results into a spatially explicit GI priority framework. Built-up cover expanded from 8.31% to 46.26% of the LGA between 1990 and 2025, while vegetation declined from 78.39% to 37.27%; mean land surface temperature (LST) rose from 33.24 °C to 37.95 °C, and SUHI intensity from 2.98 °C to 5.86 °C. Random Forest land-cover classification achieved 96.8-99.5% overall accuracy across epochs. Interpretable machine learning (XGBoost regression with SHAP explanation) on a 468-cell fishnet grid explained 81-97% of spatial LST variance and revealed a shift in the dominant thermal driver, from NDBI (spectral built-up intensity) in 1990-2002 to BuiltupDensity (spatial built-up density) in 2013-2025, reaching a mean absolute SHAP value of 3.24 °C by 2025. Overlaying 2025 heat-exposure and vegetation-deficit classifications identified 18.45 km² (4.47% of the LGA) as priority zones for GI intervention, concentrated in the urban core. The findings provide a replicable, multi-decadal thermal baseline and an evidence-based, spatially targeted GI prioritization tool directly applicable to the Ikorodu Sub-Region Master Plan (2016-2036) and to comparable rapidly urbanizing West African peri-urban LGAs.