A dynamic-thermodynamic sea ice model with 50-km spatial and 24-hour temporal resolution was used to investigate the spatial and temporal variability of the sea ice cover and the surface energy exchange in the Arctic Basin. Daily surface level air temperature and pressure data from National Centers for Environmental Prediction for 1958--1997 and climatic data for cloudiness, relative humidity, snow precipitation, and the heat flux from the deep ocean in the Greenland, Barents, and Bering Seas were used as external forcing. The model is integrated using the method of large particles <Belotserkovskii, 1984>. The model satisfactorily reproduces the seasonal and interannual variability of the main characteristics of the sea ice, the sea ice extent and exchange through the main straits, and the surface heat exchange for different parts of the Arctic Basin. In particular, estimates of the year-to-year difference between mean sea ice thickness in September show good agreement with the essential thinning of sea ice in the Canadian Basin that Rothrock et al. <1999> found. Most of the decrease in sea ice thickness is caused by a decrease in ridge concentration and an increase in the area covered by undeformed ice. Numerical experiments show that only 20% of the ice decrease can be explained by an increase in surface level air temperature. Model results for the Canadian and Eurasian Basins allow us to determine the main trends in the temporal variability of ice thickness in these regions. Our model results suggest that the recent thinning of the sea ice correlates with a decrease in the concentration of ridged ice. In turn, the concentration of ridged ice and the total sea ice volume in the Canadian Basin correlate with cyclonic and anticyclonic regimes of the wind-driven ice motion. The sharp decrease in modeled ice volume in the Canadian Basin after 1987 also coincides with a significant increase in the mean atmospheric vorticity index for the central Arctic Ocean.