Evaluation of Eurasian Snow Cover Fraction Prediction Based on BCC-CSM1.1m

Abstract

The model ability to predict Eurasian snow cover fraction (SCF) is evaluated by using the hindcast data during 1984-2019 from the Beijing Climate Center (BCC) Climate Prediction System version 2 (CPSv2), developed based on Climate System Model BCC-CSM1.1m. The SCF reanalysis data from National Snow and Ice Data Center (NSIDC) and other common variables reanalysis datasets are also used against the model forecasts. The prediction skills of Eurasian SCF in January and April are investigated, which separately represent the snow cover situation of winter and spring. The possible causes of model prediction errors are also discussed partly using the simulation data of two BCC climate models, BCC-CSM1.1m and BCC-CSM2-MR, respectively participating the phase 5 of Coupled Model Intercomparison Project (CMIP5) and phase 6 (CMIP6). Empirical orthogonal function (EOF), spatial and temporal correlation analysis, statistical test and other common methods are also adopted. The results show that, BCC-CSM1.1m is capable of forecasting the SCF in Eurasia two months ahead. However, the prediction skill varies both in space and time. In comparison with January, the model shows a better prediction skill both in climatology and interannual variability of Eurasian SCF in April. The prediction skill is highest in western Europe in January and in western Siberia in April. Lower-than-observed SCF are found in most areas of Eurasia except Tibetan Plateau in the predictions for LM0 (0 lead month). This coherent negative biases hardly varies with longer lead time in January, while the biases in key area of April reverse to positive and gradually increase. Analysis indicates that the SCF biases in January and April are positively related with those of precipitation and negatively related with those of surface temperature in the model. Moreover, since the corelated region between the precipitation biases and SCF biases reduces to some small areas in contrast with the surface temperature, the biases of SCF in the model exhibit closer relationship with surface temperature biases. In addition, comparing simulations from two BCC models, it's also found that the systematic biases originated from model resolution, parameterization scheme, etc. are also fundamental factors, which can explain the obvious underestimation of SCF in high latitude where observed SCF is nearly 100%.

Keywords

• prediction skill
• evaluation