1. C-LSAT2.0 Data Source

    This includes three global data sources (CRUTEM4, GHCN-V3, and BEST), three regional data sources, and eight national in-situ data sources.

    The data comes from https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/

    2. Sea surface temperature: ERSSTv5

    3. Sea ice surface air temperature: International Arctic Buoy Program (IABP)（ http://research.jisao.washington.edu/data_sets/iabppoles/ ;

    Reconstruction of land and ocean components: By summing the low-frequency and high-frequency components, reconstructed land temperature data can be obtained. Finally, observation constraints are applied to the reconstructed data to remove low-quality reconstruction data; Fill the data from 1850-1853 with SST anomaly observed in ICOADS Release 3.0 to form a complete monthly SST anomaly dataset from 1850-2020, and then reconstruct it using Huang et al.'s (2017) EOTs to reduce missing data.

    Arctic ice surface temperature reconstruction: Improved the ST reconstruction method for the Arctic region, representing the ST in the Arctic region as ice surface temperature (considering the similar physical properties of ice and land, sea ice is treated as land). According to data from the National Snow and Ice Data Center in the United States, from 1980 to 2020, the year with the largest sea ice area in March was 1983, and the year with the smallest sea ice area in September was 2012. Therefore, we designed two experiments: (1) CMST 2.0- Imax uses 2-meter temperature to represent the temperature in the 65-90 ∘ north latitude region, simulating the ST of the Arctic sea ice coverage area in March 1983, which is the maximum sea ice extent. (2) CMST 2.0- Imin uses 2-meter temperature to represent the temperature within the 80-90 ∘ N region, representing the ST of the Arctic sea ice coverage area in September 2012, which is the minimum sea ice extent.

    Estimation of Uncertainty in Reconstructed CMST 2.0: The uncertainty in reconstructed CMST 2.0 includes uncertainty in both land and ocean. Ocean uncertainty is the uncertainty of ERSSTv5. Land uncertainty is based on the reconstructed C-LSAT2.0 combination, which is divided into two parts: parameter uncertainty and reconstruction uncertainty. Since our method for reconstructing the temperature of polar sea ice is the same as that for reconstructing LSAT, we calculated the uncertainty in the 65-90 ∘ N (Imax) and 80-90 ∘ N (Imin) regions of CMST 2.0- Imax and CMST 2.0- Imin using the same method for calculating land uncertainty.