Quantitative Estimation Research Of Mesoscale Land Surface Evapotranspiration By Remote Sensing

Evapotranspiration(ET) is the combination process of the surface evaporation and plant transpiration which occur simultaneously, is defined as the liquid water contained in land surface being converted to gaseous water in atmosphere. It links the terrestrial water cycles, carbon cycles and energy exchange, and is the main process parameter of water and energy exchange in hydrosphere, atmosphere and biosphere. It is an important factor in heat balance and water balance of land, lake or river surface, and is central to earth system science. Evapotranspiration in ecosystems vary widely due to large regional differences on climate in China, and its trend will have a direct impact on water resources assessment. Therefore, research which carried out on recent decades’ evapotranspiration of temporal-spatial distributions, variation trends and influencing factors is very essential for the climatic and ecological environmental evaluation, and has a very important practical significance for agricultural, hydrological forecasting, weather forecasting and other industries development in our country.At present, a lot of explorations and researches have been carried out about evapotranspiration algorithm and regional application by domestic and foreign scholars. According to the existing researches, most traditional evapotranspiration estimation algorithms are based on physical models, so coefficient calibrations are required in different regions with large regional differences. While there are ample observational, researches are less with uncertain simulation accuracy which based on data corcern on global large-scale and mesoscale land surface evapotranspiration remote sensing estimation and the accuracy of algorithms. Studies that based on spatial and temporal distribution of evapotranspiration on large scale and mesoscale are few. In the context of global change, the impact mechanism of major climatic events on evapotranspiration and their quantitative relationship researches are few. For example, more comprehensive research has not been carried out about the effects of drought on land surface evapotranspiration and their quantitative relationship in China. In this paper, based on global FLUXnet sites observations, generalized regression neural network and support vector machine both machines algorithms, and semi-empirical algorithm were used to estimate global land surface evapotranspiration on these sites and validate and evaluate the accuracy of three algorithms. By using meteorological reanalysis data combined with remote sensing data, Chinese regional land surface evapotranspiration were estimated during 1982 to 2010, and long sequence of temporal-spatial distribution and change trends of the whole China and various natural geographical regions were studied. At last, the influence of drought on land surface evapotranspiration in different natural geographical regions were discussed quantitatively.The results in this paper show that:(1) Three algorithms had different performance in estimating the evapotranspiration for different vegetation types, and the differences were obvious. The results of GRNNs algorithm and SVM algorithm were lower than observations, and the bias of GRNNs algorithm was smaller, while the results of UMD-SEMI algorithm were higher than observations. GRNNs algorithm and SVM algorithm had high precision, and when ten indicators were selected the simulate effect was overall better than when five indicators were selected. The selection of the training data is a key factor which affects the results, when the training data selecting is comprehensive, there will be good results. Compared to support vector machines, generalized regression neural network required longer training and simulation time. Semi-empirical algorithm was relatively steady, and had overall good effect. Semi-empirical algorithm, which had certain physical mechanism and did not require too much data preprocessing, had the optimal speed and was relatively steady, it probably could explain 56%-75% of the land surface evapotranspiration change, and it is easy to estimate the long time series of large-scale regional evapotranspiration. When there are only five indicators(daily average temperature, relative humidity, wind speed, the incident shortwave radiation and NDVI) as the input data of generalized regression neural network algorithm and support vector machine algorithm, which consistent with the semi-empirical algorithm, they probably could explain 71%-86% of the land surface evapotranspiration change.(2) Annual daily average ET values varied from 32.90-36.05W/m2 of different algorithms in china, the regional values varied from 2.15-167.22W/m2, and the overall trend of land surface evapotranspiration decreased from southeast to northwest in china.(3) For the whole China, annual land surface evapotranspiration increased at first and then decreased in China during 1982 to 2010. Annual evapotranspiration increased on average by +0.06W/m2 per year from 1982 to 1998, due to the impact of El Ni?o during 1997 to 1998, then decreased on average by-0.13W/m2 per year from 1998 to 2010.(4) For different natural geographical sub-regions, ET change trends were different. The influence of drought on land surface evapotranspiration varied from region to region. In Northwest, Inner Mongolia and Qinghai-Tibet region, when drought occured, due to the limitation of soil water, it caused water evaporation reduced. In Central area, when drought occured, due to enough heat and good moisture conditions, affected by solar radiation, it caused water evaporation increased. In North China, when drought occured, affected by radiation, the temperature reached at the peak, evapotranspiration was higher than usual. In Northeast and South China, the relationships between drought and evapotranspiration were not obvious.