Temperature datasets

A reanalysis dataset of 2-m air temperatures is obtained based on the ERA5-Land monthly averaged data from 1989–2018. ERA5-Land is a global atmospheric reanalysis product produced by the ECMWF (European Centre for Medium–Range Weather Forecasts)⁶⁰. A series of improvements to the ERA5 climate reanalysis product have been added to the land component database. These improvements have increased the accuracy of the data for all types of land applications¹⁶. For example, ERA5-Land operates at an enhanced resolution (9 vs 31 km in ERA5). The trends of the reanalysis data combined with global observations are consistent with the general trends of other physical datasets. We can present an accurate description of the historical climate in ERA5-Land data from the past several decades. In addition, the global (land and sea) surface temperature in Fig. 4 is obtained from the monthly data of ERA5. ERA5 is a global atmospheric reanalysis product produced by the ECMWF with a spatial resolution of 0.25°⁶¹. We define spring as March–May, summer as June–August, autumn as September–November, and winter as December–February.

The autumn LST from 2004 to 2018 is calculated based on the monthly LST of the Surface Energy Balance Based Global Land Evapotranspiration database^23,62. The monthly LST is produced from MODIS emissivity data (MOD11C3 V5 and MYD11C3 V5)^23,63,64. MOD11C3 V5 and MYD11C3 V5 data have been verified under a range of representative conditions, with an average deviation of less than 1 K⁶⁵. The monthly LST has a 0.05° grid size, without gaps, and covers the period from March 2000 to 2018; the data include monthly night-time and daytime LST values^23,66. The monthly LST is calculated by averaging the daytime and night-time values of MOD11C3 and MYD11C3.

The autumn air temperatures at weather stations from 2004–2018 are calculated based on the GSOD database processed by the US National Climatic Data Center (ftp://ftp.ncdc.noaa.gov/pub/data/gsod). The database is obtained from the United States Air Force DATSAV3 surface dataset and the Federal Climate Complex Integrated Surface Hourly dataset. The analysis is based on data exchanged through the World Meteorological Organization (WMO) World Weather Watch Programme according to WMO Resolution 40 (Cg-XII)^17,67. More than 400 algorithms are used for the quality control of temperature data from weather stations (see www.ncdc.noaa.gov/isd for details). The weather station data are available on a daily time step. We utilize the average monthly air temperature from stations that collect data for 15 days or more. Next, we calculate the autumn average temperature based on the monthly temperature for all stations (n = 642). Finally, one station with autumn temperature data missing for more than 5 years between 2004 and 2018 is removed. We select 474 stations to represent the station temperatures in the study area (for the distribution, see the dots in Fig. 2a). The quality of the autumn weather station data is shown in Supplementary Fig. ². Seventy-seven percent of stations have no missing autumn air temperature data.