Estimation of crop coefficient and evapotranspiration of summer maize by path analysis combined with BP neural network

Abstract: Accurate estimation of evapotranspiration (ET) is critical for the precise management of farmland. Crop coefficient method is widely used to estimate ET. This study aimed to establish a crop coefficient model by meteorological factors combined with leaf area index so as to accurately estimate ET. A experiment was carried out at Wudaogou Hydrological Experimental Station. Meteorological and soil parameters were measured by local weather station and weighing lysimeters in 2018. These parameters included leaf area index, temperature, relative humidity, solar radiation, net radiation, soil heat flux, wind direction, water surface evaporation, wind speed, atmospheric pressure and the soil moisture content of different layers. The path analysis method was used to screen the key factors affecting crop coefficient. And the model based on BP neural network was established to estimate the crop coefficient and ET of summer maize at different groundwater depths (1 and 3 m) in the rain free period. The results showed that the leaf area index, temperature and wind speed were critical factors affecting the crop coefficient in both at depth of 1 and 3 m. Moreover, in the depth of 3 m, the crop coefficient was significantly affected by the soil moisture content at depth of 130 cm. The model could simulate well crop coefficient. The average absolute error in the whole growth period at depth of 1 m was 0.04 mm/d, and the correlation coefficient was 0.94. The average absolute errors at the initial, developmental, intermediate and late stages were 0.06, 0.09, 0.05 and 0.03 mm/d, respectively. In the depth of 3 m, the average absolute error was 0.08 mm/d, and the correlation coefficient was 0.92. During the four growth periods, the average absolute errors were 0.11, 0.10, 0.07 and 0.03 mm/d, respectively. Therefore, the crop coefficients model considering temperature, wind speed and leaf area index had high accuracy. Then, the model was used to estimate the ET and the results showed that the estimation at the depth of 1 m was better than that at 3 m. The accuracy of estimating ET at both depths was high. In the whole growth period, the average absolute error of ET at depth of 1 m was 0.72 mm/d, and the daily forecast ability at different growth stages was also well. At each growth stage, the average absolute errors of ET were 0.56, 059, 0.66 and 0.45 mm/d, respectively. At depth of 3 m, the average absolute error of ET in the whole growth period was 0.73 mm/d. At each growth stage, the average absolute errors of ET were 0.82, 098, 0.68 and 0.29 mm/d, respectively. At different time scales (1, 3 and 5 d) , the concordance index of estimated ET was about 1.00 and the average absolute error was less than 1.0 mm/d at different depths. The absolute error decreased and the consistency index appeared the opposite as the forecast time scale increased. It indicated that the forecast accuracy and forecast ability increased. Thus, the method above is reliable to estimate the ET of summer maize and it can meet the forecasting requirements of irrigation planning and agricultural water management.