Adjustment of the urbanization bias in surface air temperature series based on urban spatial morphologies and using machine learning

Urbanization bias is one of the main systematic biases in surface air temperature (SAT) in climate change monitoring. The traditional linear approach assumes that the impact of urbanization on SAT sequences increases at a uniform rate, but this contradicts the real urbanization processes that occurred around observation stations. This paper puts forward a new adjustment method by using urban spatial morphologies and a random forest (RF) model. Results showed a warming rate of 26.2% (59.1%) during 1991 -2020 (2006 -2020) that was caused by urbanization. The RF model achieved better performance between urbanization bias and urban morphologies around 4 km buffer areas of the national meteorological stations than other buffers. It was noted that the importance analysis illustrated the urban horizontal morphologies had a stronger explanation for urbanization bias than vertical morphologies. Finally, the importance values were introduced into the traditional adjustment method. Taking Hefei as an example, the adjustment results obtained using the new method reveal that the urbanization bias maintains dynamic consistency with the observed changes in the surrounding environment of the target station. The new adjustment method might help improve our understanding of the relationship between human activities and regional climate change and also would provide technical references for the protection of the observation environment and the construction of reference climate data.