The 30-year average climate dataset (1981-2010) comes from two sources, namely the China Meteorological Data Service Center (CMD: http://data.cma.cn ）2160 meteorological stations and 25 meteorological stations of the Central Weather Bureau (www.cwb. gov.tw). The monthly ground observation dataset of 756 meteorological stations from 1952 to 2019 is from the China Meteorological Administration http://data.cma.cn .

    The TRMM3B43 product was used in the study, with a spatial resolution of 0.25 ° and a latitude range of 50 ° S to 50 ° N. follow https://mirador.gsfc.nasa.gov I downloaded the monthly data of TRMM3B43 7th edition in NetCDF format.

    The Land Surface Temperature (LST) is compiled by the Medium Resolution Imaging Spectroradiometer (MODIS). The day night LST average values from 2001 to 2019 were extracted from MOD11A2 images with a resolution of 1 kilometer, and averaged by month and year.

    Interpolate the monthly average precipitation and temperature values for multiple years (1980-2010) using thin plate spline curves (TPS) from the R software package "fields". Specifically, the process of generating ChinaClim_baseline based on the ten fold spatial stratification cross validation method can be described as follows: In order to ensure that each climate region has sufficient training and testing data to construct and validate the model, while reducing spatial autocorrelation, the meteorological stations in each climate region are divided into ten fold intervals. In addition to longitude, latitude, and altitude, we also obtained the distance to the nearest coast based on the geographic coordinates of each weather station, as well as satellite driven variables (TRMM and LST). We calculated the difference values between satellite driven variables (TRMM and LST) and field observations, and set a threshold (± 3 SD represents temperature; ± 4 SD represents precipitation) to detect outliers. Similarly, we also examined the correspondence between the elevation reported by the station and the elevation obtained from the elevation grid data with a spatial resolution of 1 km. It is worth noting that according to Hutchinson's (1995) proportional suggestion, the elevation (m) is divided by 1000, and the precipitation is square root transformed according to Hutchinson and Xu's (2013) suggestion before fitting. We randomly selected 9 folded weather stations from each climate region and integrated them into a new training dataset. The remaining datasets were merged into a test dataset to validate the performance of the model. Attempt to construct a TPS model using 11 models per month in each climate region using different combinations of variables (as described in Table S1 for longitude, latitude, altitude, distance to nearest coast, and model formulas for TRMM and LST). By identifying the TPS model with the minimum root mean square error (RMSE), select the best TPS model for each climate zone each month, and use the entire dataset to generate the final surface for each of the four climate zones. Finally, these surfaces are combined together using the inverse distance weighting method.