Quantifying the spatial and temporal relationship between air and land surface temperatures of different land-cover types in Southeastern China

Near surface air temperature (T_a) is an important variable utilized for various fields of research such as agriculture, epidemic forecasting, and climate change. It is usually obtained from a limited number of weather stations that are unevenly distributed within a region. Thus, the spatial information of T_a is rarely provided especially in complex topographical and underdeveloped regions. To compensate for this shortcoming, T_a can be estimated by using the land surface temperature (T_s) from satellite data because remote sensing has an advantage of describing the spatial heterogeneity over a large geographical area. In previous research, authors focused on analysing the effect of limited factors on T_a observations by using a small number of weather stations. However, this study explores a thorough sensitivity analysis of several factors on the relationship between T_a (obtained from a dense network of automatic weather stations) and T_s of different land-cover types using statistical analysis. This article discovered the relationships between T_a and T_s. First, comparisons between daytime and night-time Moderate Resolution Imaging Spectroradiometer (MODIS) T_s data with T_a data showed that a better agreement was achieved during the night than the day. Second, when comparing the results of different land-cover types, the correlation coefficient of the vegetated area was higher than that of water and impervious surface. Third, a comparison of statistical results for different seasons indicated that the correlation between T_a and T_s was weakest in a hot season. Fourth, altitude seemed to have no significant effect on the T_a–T_s relationship. Finally, the relationship between T_a and T_s strengthened with increasing window size except for the vegetated area but tended to saturate as the T_s window increased to the optimal size. These results will be helpful for building an appropriate model to derive air temperature directly from the remotely sensed data in the future.