A new approach to accurate validation of remote sensing retrieval of evapotranspiration based on data fusion

Abstract. The study presented a new method of validating the remote-sensing (RS) retrieval of evapotranspiration (ET) under the support of a distributed hydrological model: Soil and Water Assessment Tool (SWAT). In this method, the output runoff data based on a fusion of ET data, meteorological data and rainfall data, etc. were compared with the observed runoff data, so as to carry out validation analysis. A new pattern of validating the ET data obtained from RS retrieval, which was more appropriate than the conventional means of observing the ET at several limited stations based on eddy covariance, was proposed. It has integrated the advantage of high requirement of ET with high spatial resolution in the distributed hydrological model and that of the capacity of providing ET with high spatial resolution in RS methods.

First, the ET data in five years (2000–2004) were retrieved with RS according to the principle of energy balance. The temporal/spatial ditribution of monthly ET data and related causes were analyzed in the year of 2000, and the monthly ET in the five years was calculated according to the PM model. Subsequently, the results of the RS retrieval of ET and the PM-based ET calculation were compared and validated. Finnaly, the ET data obtained from RS retrieval was evaluated with the new method, under the support of SWAT, meteorologic data, Digital Elevation Model (DEM), landuse data and soil data, etc. as the input, being compared with the PM-based ET.

According to the ET data analysis, it can be inferred that the ET obtained from RS retrieval was more continuous and stable with less saltation, while the PM-based ET presented saltation, especially in the year of 2000 and 2001. The correlation coefficient between the monthly ET in two methods reaches 0.8914, which could be explained by the influence from clouds and the inadequate representativeness of the meteorologic stations. Moreover, the PM-based ET was smaller than the ET obtained from RS retrieval, which was in accordance with previous studies (Jamieson, 1982; Dugas and Ainsworth, 1985; Benson et al., 1992; Pereira and Nova, 1992).

After the data fusion, the correlation (R²=0.8516) between the monthly runoff obtained from the simulation based on ET retrieval and the observed data was higher than that (R²=0.8411) between the data obtained from the PM-based ET simulation and the observed data. As for the RMSE, the result (RMSE=26.0860) between the simulated runoff based on ET retrieval and the observed data was also superior to the result (RMSE=35.71904) between the simulated runoff obtained with PM-based ET and the observed data. As for the MBE parameter, the result (MBE=−8.6578) for the RS retrieval method was obviously better than that (MBE=−22.7313) for the PM-based method. The comparison of them showed that the RS retrieval had better adaptivity and higher accuracy than the PM-based method, and the new approach based on data fusion and the distributed hydrological model was feasible, reliable and worth being studied further.