Study on different Machine Learning Techniques (MLT) in forecasting Crop Water Requirement
Loading...
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
DRPCAU, PUSA
Abstract
Crop water requirements must be accurately determined for irrigation scheduling, water resources management, and environmental analysis. There are several locations where different climatic data may not be accessible for its determination. In these circumstances, crop water requirement modelling is a suitable method for predicting crop water requirements. To predict the crop water requirements of rice, wheat, maize, and sugarcane in the Samastipur area of Bihar, Machine Learning Techniques such as Random Forest (RF), Multivariate Adaptive Regression Splines (MARS), Support Vector Machine (SVM) and Multiple Linear Regression (MLR) were utilised in the current study. The Meteorological department of RPCAU, Pusa, Bihar provided the data including relative humidity (maximum and minimum), and temperature (maximum and minimum) while data about solar radiation and wind speed was obtained from Prediction of Worldwide Energy Resources (NASA/POWER) website. For this study, data spanning 20 years (2001–2020) was collected. The FAO-56 Penman-Monteith method and crop coefficient approach were used to determine the water requirement for the selected crops. The Gamma test was utilized for the best input determination. The entire dataset was divided into training (80%) and testing (20%) datasets. The assessment of the models was by the Coefficient of Determination (R2),
Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Nash-Sutcliffe Efficiency (NSE). It can be inferred that the MLR and MARS models predicted rice crop water requirements finer than they did for other crops. The SVM model accurately predicted the water requirement of the wheat crop compared to other crops. More precisely than other crops, the RF model forecasted the water requirement for sugarcane. For all crops, the overall models' performance-wise rankings were RF, MARS, SVM, and MLR. From the results attained, the RF model beat the other three models throughout training and testing for all crops and can be highly recommended for crop water requirement prediction.