Multi-scale spatial analysis of satellite-retrieved surface evapotranspiration in Beijing, a rapidly urbanizing region under continental monsoon climate.

Abstract

As one of the important components of hydrological cycle, evapotranspiration spatial distribution is of great significance to regional water resources planning and rational utilization. This research used Surface Energy Balance System model to estimate the daily evapotranspiration (ET) in Beijing based on Landsat 8 observations. Results showed that the daily ET in Beijing ranged from 3.469 to 5.474mm/day. ET is known to decrease with the increase of land surface temperature (LST) and to increase with the increase of Normalized Difference Vegetation Index (NDVI). NDVI primarily decreased from the northwest to the southeast. When the NDVI value was 0.4-0.6, the average ET peaked at 4.88mm/day, and then slightly decreased by 3.7%. The coefficient of determination of NDVI (0.95) was much greater than that of LST (0.30) upon linear fitting, showing LST was not the main factor controlling ET in Beijing. In contrast to the linear fitting results, the spatial correlation between LST and ET is more significant than that between NDVI and ET in the global bivariate spatial analysis, where the absolute value of global bivariate Moran's I of LST (0.51) was higher than that of NDVI (0.21) at a resolution of 150m. And the univariate spatial autocorrelation indices of LST, ET, and NDVI equaled 0.84, 0.65, and 0.51, respectively. Furthermore, the complex spatial distribution pattern of variables could significantly affect the correlation analysis results. Local bivariate spatial analysis showed that over 60% of the Beijing area had a significant correlation, of which the negative correlation area of LST accounted for about 85%, and the positive correlation area of NDVI accounted for 74%. By improving the correlation analysis accuracy, the regional conditions for the establishment of correlation analysis results were clarified from the overall correlation analysis results.