Research On Spring Maize Evapotranspiration And Soil Moisture Models Based On Machine Learning Algorithm

This study is based on the results of spring maize experimental study carried out in the Mu Us sandy land.The dual crop coefficient approach is used to estimate crop evapotranspiration.Based on the measured meteorological and crop growth characteristics data,the basal crop coefficient and soil evaporation coefficient of the dual crop coefficient method can be calibrated.The dual crop coefficient method is used to estimate and distinguish the evapotranspiration of the spring maize field in the Mu Us sandy land.The study explore the factors that can affect the water transport in the farmland,Based on BP neural network,we can establish a soil moisture prediction model with influence factors as input variables and multi-layer soil moisture as output variables.the k-nearest neighbor algorithm,the BP neural network and the extreme learning machine are used to construct models for predicting crop evapotranspiration,and the simulation performance of models is investigated under different input scenarios.Under the condition of reducing the meteorological characteristics,we can determine the optimal model.Based on the above research,the following main conclusions are obtained:(1)Under fully irrigated conditions,the evapotranspiration values of T5 treatment were estimated to be 1.83,3.34,4.73,2.70,and 3.21 mm/d of the initial,development,middle,late and full growth periods of 2018 using the dual crop coefficient method,respectively;in 2019 They are1.77,2.98,5.29,3.13 and 3.54mm/d,The variation trend is consistent with the measured values in the monitoring period.The estimated soil evapotranspiration during the whole growth period in2018 and 2019 accounted for 24.17%and 16.15%of the evapotranspiration,respectively;the estimated crop evapotranspiration accounted for 75.83%and 83.85%of the evapotranspiration,respectively.Combined with the actual situation of spring maize planting in the Mu Us sand land,the yield reaches the optimal state.This study recommends that 1.0 times the crop water requirement is the optimal irrigation water amount for local drip irrigation spring maize.(2)Based on the BP neural network,we can construct a soil moisture prediction model that predicts soil moisture content at 8 different depths,which can predict the soil moisture content after a time at 8 different depth.The data shows that the simulated soil moisture content at 10 and20 cm soil depths,R~2 is greater than 0.8,and at 30,40,50,60,70 and 80 cm soil depths,R~2 is greater than 0.9.The multi-depth soil moisture content prediction model based on BP neural network provides a solution for mastering the distribution characteristics of soil moisture content.(3)The k-nearest neighbor algorithm,BP neural network and extreme learning machine model can use meteorological factors and crop coefficient data to simulate the nonlinear process of the crop evapotranspiration.with low cost and simple data processing.Compared with traditional monitoring methods and Penman-Monteith formula,these models have advantages.Evapotranspiration simulation can also be performed when there is only one data input,which can provide a simple solution for nonlinear multivariate functions.The three models can simulate spring maize evapotranspiration well,and the BP neural network has the highest accuracy,which can be recommended as a simulation model of spring maize evapotranspiration.When the weather input features are reduced,The three methods can also maintain high precision and can complement each other in practical application.