Research On Water Demand Model For Winter Wheat In Luancheng Based On Machine Learning

Water is the main lifeline for crops and the lack of water resources in the country has greatly exacerbated the difficulties of water availability and has had a major impact on agricultural water use.The study of crop water requirements therefore plays an important role in agricultural irrigation and water availability.In this paper,a reference crop degree evapotranspiration model based on meteorological features was established by combining agricultural theory with machine learning methods.Using temperature,air temperature and radiation,respectively,multiple meteorological features fused data,multiple machine learning methods were applied to simulate the evapotranspiration and transpiration of winter wheat at the daily scale and to analyse its water demand.（1）A temperature-based model for daily-scale evapotranspiration of winter wheat was developed.Simulations were carried out using nine machine learning algorithms and compared with measured values calculated by the ETO_HSempirical model as well as the Penman-Monteith model.The results showed that the predictions of the machine learning algorithm models using a single temperature factor were all significantly better than the empirical model predictions,with the empirical model RMSE value of 1.96 mm/d and all nine machine learning RMSE below 1.5 mm/d.（2）A daily-scale evapotranspiration model for winter wheat based on temperature and radiation was developed.Simulations were performed using eight machine learning algorithms and compared with measured values calculated by three empirical models as well as the Penman-Monteith model.The results showed that the machine learning algorithm predictions using a fusion of temperature and radiation were slightly better than those of the machine learning algorithm model using a single temperature factor and significantly better than the empirical model predictions.The RMSE of the empirical models reached above1.93mm/d,while the RMSE of all eight machine learning models was below 1.2mm/d.（3）A daily-scale evapotranspiration model based on the fusion of multiple meteorological features was developed for winter wheat,and the LSTM and CNN algorithms were used to simulate the predictions and compare with the measured values calculated by the Penman-Monteith model.The results showed that the LSTM model was more accurate and stable in predicting winter wheat evapotranspiration,with RMSEs of 1.131 mm/d and 1.129mm/d for the two validation data sets,and the CNN model had a very high estimate in predicting winter wheat decadal evapotranspiration,with an RMSE of 0.3 mm/d.（4）The water demand of winter wheat was estimated and analysed using data within the reproductive period of winter wheat and the corresponding crop coefficients.The water demand of winter wheat was found to be greater from nodulation to maturity than from greening to initiation than from seedling to overwintering,with the overall water demand being greatest in summer and least in winter.The results of this paper show that machine learning can be used to simulate evapotranspiration from reference crops instead of empirical models,thus simplifying the computational complexity.The integration of the reference crop evapotranspiration model with the crop coefficients also allows the estimation of crop water requirements for agricultural irrigation purposes.