Representativeness Analysis of Meteorological Stations Based on Temperature Estimated from MODIS Data
Long time series Land Surface Temperature (LST) products can be used to quantify representativeness of meteorological stations. However, spatial heterogeneity of LST is dramatically higher than that of air temperature, which is prone to an under estimation of the representativeness. To obtain more accurate results, this paper proposes a new procedure in which spatially continuous air temperature is first estimated from LST and other parameters, and then used for estimating representativeness of meteorological stations. A case study focusing on Linzhi Meteorological Station located in southeastern Tibetan Plateau of China is presented. First, 8 d averaged maximum temperature (T air) and the corresponding MODIS LST product (MOD11A2,8 d composited with 1 km spatial resolution) of meteorological stations during 2000-2011 are extracted. The correlation coefficients between T air and LST are analyzed under different conditions. Then, a Cubist regression tree model for T air estimation is developed using LST, Julian day, and the number of clear skies as predictors (RMSE=1.4718℃, r2=0.95). Finally, the model is applied to the region around Linzhi Station to estimate T air and representativeness. As a result, it is found that the new procedure can produce more reasonable results: 15 km×15 km can be well represented by Linzhi Station, rather than 3 km×3 km when LTS data was used.