Impact of urban sprawl on land surface temperature in the Mashhad City, Iran: A deep learning and cloudbased remote sensing analysis

The evolution of land use patterns and the emergence of urban heat islands(UHI) over time are critical issues in city development strategies. This study aims to establish a model that maps the correlation between changes in land use and land surface temperature(LST) in the Mashhad City,northeastern Iran. Employing the Google Earth Engine(GEE) platform, we calculated the LST and extracted land use maps from 1985 to 2020. The convolutional neural network(CNN) approach was utilized to deeply explore the relationship between the LST and land use. The obtained results were compared with the standard machine learning(ML) methods such as support vector machine(SVM),random forest(RF), and linear regression. The results revealed a 1.00°C–2.00°C increase in the LST across various land use categories. This variation in temperature increases across different land use types suggested that, in addition to global warming and climatic changes, temperature rise was strongly influenced by land use changes. The LST surge in built-up lands in the Mashhad City was estimated to be 1.75°C, while forest lands experienced the smallest increase of 1.19°C. The developed CNN demonstrated an overall prediction accuracy of 91.60%, significantly outperforming linear regression and standard ML methods, due to the ability to extract higher level features. Furthermore, the deep neural network(DNN)modeling indicated that the urban lands, comprising 69.57% and 71.34% of the studied area, were projected to experience extreme temperatures above 41.00°C and 42.00°C in the years 2025 and 2030,respectively. In conclusion, the LST predictioin framework, combining the GEE platform and CNN method, provided an effective approach to inform urban planning and to mitigate the impacts of UHI.