Machine learning based crop water demand forecasting using minimum climatological data

Abstract

Rice is one of the world’s most popular food crops. Since its production is dependent on intensive water use, water management is critical to ensure sustainability of water resource. However, very limited data is available on water use in rice irrigation. In the present study, traditional machine learning methods have been used to predict the irrigation schedule of rice daily. The data of year 2013-2015 is used to train the models and to further optimise it. The data of 2016-2017 is used for testing the models. Correlation thresholds are used for feature selection which helps in reducing the number of input parameters from the initial 26 to final 11. The models estimated the crop water demand as a function of weather parameters. Results show that Adaboost performed consistently well with an average accuracy of 71% as compared to other models for predicting the irrigation schedule.