Spatial autoregressive regression analysis of surface urban heat island intensity in the tropical industrial city of Rayong, Thailand

In this study, we introduced the spatial autoregressive (SAR) mixed model as a viable alternative for measuring the surface urban heat island intensity (SUHII). The purpose was to showcase how the SAR mixed model can effectively address the issue of biased estimates in conventional simple linear regression (SLR), specifically when analyzing spatial dependence data. Such concerns have been overlooked in the studies on the SUHII. The case study illustrates a multi-urban core in the tropical industrial Rayong, Thailand. Using eight diurnal and seasonal MODIS land surface temperature datasets as a response and the percentage of impervious surface areas as a predictor, we evaluated the performance of three regression models (SLR, SLR with kernel density estimation, and SAR). Our findings indicated that the SAR mixed model outperformed the SLR models, as evidenced by its lowest AIC and Moran's I statistic. Mapping the SLR residuals highlighted model misspecification due to spatial dependence, but this was not the case for SAR residuals. The annual SUHII in this tropical, industrial city was 1.89 °C during the day and significantly dropped to 0.66 °C at night. There was no seasonal variability in daytime SUHII, but it was present at night, especially in the rainy season.