Research On Crop Water Requirement Models In Middle Reaches Of Heihe River Basin Based On Machine Learning

Crop water requirement,which is also called as crop evapotranspiration,is mainly used for guiding farm irrigation and agricultural water schedulement.It is one of the most important parameters in Soil-Plant-Atmosphere Continuum(SPAC).Crop water requirement can be estimated by crop coefficient method based on reference crop evapotranspiration and crop coefficient.Among them,reference crop evapotranspiration depends on meteorological factors,and crop coefficient is determined by crop factors,crop management and environmental conditions.However,the experience models based on air temperature and radiation is of limited accuracy and poor adaptability,and FAO Penman-Monteith model need multiple meteorological factors which is difficult for agricultural application.Meanwhile,without agricultural knowledge,the estimating models developed by machine learning is difficult to provide the accurate assessment of water consumption.In view of the three problems,we explored to develop the estimation models of reference crop evapotranspiration by taking subtraction of meteorological factors,improvement of estimation accuracy,reduction of training time and introduction of agricultural knowledge characteristics into account.Water resources of the middle reaches of Heihe River in Gansu province is of high shortage,and agricultural irrigation occupied 90%of water consumption.Firstly,the multiple time scale reference crop evapotranspiration models were developed with meteorological data of different varieties in the middle reaches of Heihe River.Based on the obtained model,the crop water requirements of wheat and corn of growth period were assessed using single crop coefficient method and introduction of the local environmental factor as crops and soil.For construction of crop evapotranspiration models,the following works were carried out in our study.1.Study on monthly reference crop evapotranspiration models based on air temperature.Air temperature,evapotranspiration mechanism and agricultural knowledge were combined for model construction using divide-and-conquer algorithm(DC),back-propagation neural network(BP-NN)and correlation between monthly ordinal and reference crop evapotranspiration.The prediction performance gain improvement.2.Study on daily reference crop evapotranspiration models based on air temperature and radiation.As data amount of day scale atmosphere is highly large than the month scale data,computation speed of BP-NN is slow,and generalization ability of BP-NN is insufficient,and the final results depend on the initial values of parameters and weights.So divide-and-conquer algorithm(DC),evapotranspiration mechanism and extreme learning machine(ELM)algorithm were used to develop DC-ELM model.Results shows the accuracy can be maintained with shorter training time.3.Study on daily reference crop evapotranspiration models based on multiple meteorological feature.Relationship between the ordinal number of ten days in farming season knowledge and reference crop evapotranspiration models were used to evaluation model with random forest(RF)algorithm and different meteorological feature.And application of ensemble learning in reference crop evapotranspiration models was also explored.4.Analog computation of crop water requirement using scale reference crop evapotranspiration models and single crop coefficient method.Crop water requirements of spring corn and spring wheat of different growth period in Ganzhou of Zhangye and Suzhou of Jiuquan were computed during 2016,and characteristic of water requirements was also obtained,which can provide the reference for agricultural irrigation.In this paper,we have studied the reference crop evapotranspiration models for reducing meteorological factors,improving evaluation accuracy and reducing training time based on agricultural knowledge and machine learning methods.We also simulated the water requirements of wheat and corn in growing period in the middle reaches of Heihe River Basin and provided the reference frame for agricultural irrigation forecast and agricultural water regulation.