*2.2. Data*

We selected a gridded evapotranspiration dataset—Derived Optimal Linear Combination Evapotranspiration (DOLCE) [40]. DOLCE is a hybrid evapotranspiration dataset under observational constraints, which merges four available global ET datasets: ERA5- land, FLUXCOM METEO+RS, GLEAM v3.5a, and GLEAM v3.5b. The contribution of each dataset to DOLCE ET is based on its ability to match field observations as well as its dependence to the other parent datasets. The latest version of DOLCE-ET V3.0 also provides time-varying estimates of its uncertainty errors. DOLCE-ET V3.0 has a temporal resolution of months and a spatial resolution of 0.25◦. It spans the years 1980–2018 and covers the global land surface.

We use the China Meteorological Forcing Dataset (CMFD) for precipitation and nearsurface air temperature data [41]. CMFD uses Princeton reanalysis data, GLDAS (global land data assimilation system) data, GEWEX-SRB (the global energy and water exchanges) radiation data, and tropical rainfall measuring mission precipitation data as background fields, and merges the conventional meteorological observation data of the China Meteorological Administration to produce a regional high spatial and temporal resolution dataset. The dataset has a temporal resolution of months and a spatial resolution of 0.1◦, with a spatial range of 60–140◦E and 15–55◦N.

For soil moisture, we selected the Climate Change Initiative (CCI) data set of the European Space Agency (ESA) [42]. The CCI SM v04.7 product consists of three sets of surface soil moisture data: active products, passive products, and combined products. The "active product" and "passive product" are generated by merging the soil moisture outputs of the scatterometer and radiometer, respectively; the "combined product" is a hybrid product based on the first two data sets, and used in this study. The dataset has a temporal resolution of a day and a spatial resolution of 0.25◦, and spans the period from November 1978 to December 2019. Site observations was used to validate CCI data.

Boundary layer heights used ECMWF monthly averaged ERA5 data with a spatial resolution of 0.25◦. All data were time-scaled to months and interpolated to 0.25◦ × 0.25◦ spatial resolution.
