Cellular Automata–Artificial Neural Network Approach to Dynamically Model Past and Future Surface Temperature Changes: A Case of a Rapidly Urbanizing Island Area, Indonesia

In 2024, significant increases in surface temperature were recorded in Batam City and Bintan Regency, marking the highest levels observed in regional climate monitoring. The rapid conversion of vegetated land into residential and industrial areas has been identified as a major contributor to the acceleration of local climate warming. Climatological analysis also revealed extreme temperature fluctuations, underscoring the urgent need to understand spatial patterns of temperature distribution in response to climate change and weather variability. This research uses a Cellular Automata–Artificial Neural Network (CA−ANN) approach to model spatial and temporal changes in land surface temperature across the Riau Islands. To overcome the limitations of single-model predictions in a geographically diverse and unevenly developed region, Landsat satellite imagery from 2014, 2019, and 2024 was analyzed. Surface temperature data were extracted using the Brightness Temperature Transformation method. The CA−ANN model, implemented via the MOLUSCE platform in QGIS, incorporated additional environmental variables, such as rainfall distribution, vegetation density, and drought indices, to simulate future climate scenarios. Model validation yielded a Kappa accuracy coefficient of 0.72 for the 2029 projection, demonstrating reliable performance in capturing complex climate–environment interactions. The projection results indicate a continued upward trend in surface temperatures, emphasizing the urgent need for effective mitigation strategies. The findings highlight the essential role of remote sensing and spatial modeling in climate monitoring and policy formulation, especially for small island regions susceptible to microclimatic changes. Despite the strengths of the CA−ANN modeling framework, several inherent limitations constrain its application, particularly in the complex and heterogeneous context of tropical island environments. Notably, the accuracy of model predictions can be limited by the spatial resolution of satellite imagery and the quality of auxiliary environmental data, which may not fully capture fine-scale microclimatic variations.