A transferable decision-support framework coupling land-cover and land surface temperature simulation for automated green–blue–grey mitigation allocation in data-scarce tropical watersheds

Most land-cover simulations end at a predictive map and leave the engineering question -what mitigation to build, and where -unanswered, an omission that is most acute in data-scarce tropical watersheds lacking dense climatic or hydrometric instrumentation.This study develops and demonstrates a reproducible, four-module decision-support framework, termed coupled thermal-land-cover mitigation allocation (CTMA), that chains landcover and land surface temperature (LST) forecasting to an automated green-blue-grey mitigation prescription using only openly available Earth-observation data.The framework was demonstrated on the Wanggu watershed, Southeast Sulawesi, Indonesia, using multitemporal Landsat 8 OLI/TIRS imagery (2015, 2020, 2025), a cellular automata-Markov chain (CA-Markov) model coupled to an artificial neural network multi-layer perceptron (ANN-MLP) transition sub-model, and a transition-specific empirical LST-warming model requiring no external climate forcing.Classification reached a kappa accuracy of 86.59%, and the coupled projection was validated by hindcast against the observed 2025 map (Kno = 0.9128, Klocation = 0.9501, K-standard = 0.8940).Rather than inferring causation from correlation, a multiple regression showed that impervious and bare cover independently raised mean LST by 1.24 °C after controlling for the decadal trend (p = 0.036; model R² = 0.88), corroborating the surface-energy-balance mechanism.Under a business-as-usual scenario the LST class above 42 °C is projected to expand by 123% by 2035 and 8,471.18ha to exceed warming thresholds; the framework then translated this forecast into 28 coded green-blue-grey directives spanning the watershed, dominated by downstream infiltration works (22.19%), mid-watershed water-control terracing (21.40%), and upstream gully-plug structures (14.68%).The contribution is a transferable, low-data workflow that converts environmental simulation into an actionable ecological-engineering prescription.