Abstract: Time series of MODIS land surface temperature (LST) data, together with meteorological data of 137 stations and ASTER GDEM data for 2001-2007, were used to estimate and map the spatial distribution of monthly mean air temperatures of the Tibetan Plateau and neighboring areas. Time series and regression analyses of monthly mean land surface temperature (Ts) and monthly mean air temperature (Ta) were conducted using both ordinary linear regression (OLS) and geographical weighted regression (GWR) methods. Analysis shows that recorded Ta is rather closely related to Ts, and that the GWR method has a much better result (adjusted R² > 0.91, root mean square error (RMSE)=1.16-1.58℃) for estimating Ta than OLS. The GWR model, with MODIS Ts and altitude as independent variables, was thus used to estimate Ta for the Tibetan Plateau. For more than 80% of the stations, the Ta retrieved from Ts had residuals lower than 2℃. Analysis of the spatial pattern of retrieved Ta data showed that the mean Ta of the summer half year was higher than 0℃ even at high altitudes of 5000±600 m of the plateau, especially in the warmest month (July) the Ta in high mountain areas with altitudes of 4000-5500 m could reach as high as 10 ℃. This may help explain why the highest timber line in the northern hemisphere is located on the Tibetan Plateau. According to our results, Ta in July was probably 6-10℃ warmer in the inner plateau than in the outer plateau at any given elevation which resulted from the heating up effect of the Plateau.

Key words: Tibetan Plateau, air temperature estimation, MODIS land surface temperature, geographical weighted regression, heating-up effect