Method of regional crop yield estimation based on remote sensing evapotranspiration model

Crop yield estimation over irrigation district is important for evaluation of water use efficiency and agricultural water management. The development of remote sensing technology provides an effective way to estimate crop yield at regional scale. In this paper, Hetao Irrigation District in Inner Mongolia, a representative irrigation district in arid region, was taken as the study region. Three counties (Linhe, Wuyuan and Hangjinhouqi) in Hetao Irrigation District were selected as the study area, where occupied most farmland in Hetao Irrigation District. Maize was the major crops in the study area. Maize daily evapotranspiration during growing period over Hetao Irrigation District was obtained based on remote sensed evapotranspiration model (HTEM) and remote sensed crop classification model fed with MODIS data from 2003-2012. The HTEM model was established from hybrid dual-source scheme and trapezoid framework and crop classification model was based on NDVI time series and phenology. These two models had been tested with experimental data and statistical data. On these bases, three water production functions, Jensen function, Blank function and Stewart function, were selected for the establishment of maize yield estimation model. Jensen function is a product model, while Blank function and Stewart function are both summation models. The parameters and applicability of the yield estimation models were also analyzed in this study. Results showed that, the HTEM model was capable of estimating evapotranspiration in this region with relative error of 7.02% and root mean square error (RMSE) of 0.52 mm/day at site scale, respectively. The relative error and RMSE based on water balance model at region scale were 5.28% and 26.21 mm, respectively. The annual change of the maize actual evapotranspiration was in single-peak type, and the peak value occurred on late July with daily evapotranspiration of approximately 5 mm. The growth period of maize was 160-170 days. The annual average evapotranspiration of maize during growth period was approximately 526 mm. The results also showed that three water production functions had good performance in maize yield estimation with high accuracy during 2003-2012. The Stewart function had the highest accuracy, with relative error of 4.30% and correlation coefficient of 0.75. The relative error of Jensen function and Blank function were 4.47% and 4.36%, and the correlation coefficient of Jensen function and Blank function were 0.74 and 0.75, respectively. The average maize yield from 2003-2012 estimated by three water production functions were 10 185.82 kg/hm2(Jensen function), 10 176.58 kg/hm2 (Blank function) and 10 176.00 kg/hm2 (Stewart function), respectively. As a result, the Stewart function had the best performance in Hetao Irrigation District. The three parameters of Stewart function were also fitted with K=1.1, B=2.76 and n=5.0. The spatial distribution of maize yield estimated by Stewart function showed that the northern part of study area have the lowest maize yield and the highest maize yield occurred in southern part of study area. The interannual variation of maize yield indicated that the lowest yield and highest yield during the study period occurred in 2007 and 2003, respectively. Moreover, the study indicated that remote sensing data and remote sensed evapotranspiration model and remote sensed crop classification model were feasible to estimate maize yield over Hetao Irrigation District. © 2019, Editorial Department of the Transactions of the Chinese Society of Agricultural Engineering. All right reserved.