Aquacrop model and machine learning algorithms for sugarcane yield prediction: A performance evaluation in dry zone of Sri Lanka.

Abstract

AquaCrop, a leading crop simulation model, improves agricultural practices by simulating an extensive tapestry of diverse crop conditions. Nonetheless, the model's strong demand for inputs, which includes climate, agricultural, and management data, causes issues in many developing nations. As a result, the development of an exceptionally accurate model with minimum essential inputs becomes an absolute need. The purpose of this study was to (a) deploy the AquaCrop model to simulate sugarcane yield under various weather scenarios in Sri Lanka's dry zone, and (b) create and evaluate the performance accuracy of an intelligent model for predicting sugarcane yield using climatological inputs using Machine Learning (ML) algorithms. Observational datasets comprising four growth seasons were used to calibrate and validate AquaCrop default parameters for sugarcane farming in the research location. At the Sugarcane Research Institute in Udawalawa, one hundred scenarios were simulated on a field plot, and the AquaCrop model provided the corresponding yield values. For the prediction model, this study used three ML algorithms: Random Forest (RF), Support Vector Regression, and Gradient Boosting Regressor (GBR). These algorithms were tested in a variety of scenarios, using input variables such as minimum and maximum temperatures, sunshine hours, average relative humidity, rainfall, and wind velocity. The results revealed that the AquaCrop model was successfully validated in the specified study area, with an R2 of 0.86. In the comparison of predicted and observed values, the GBR had an R2 of 0.79 and an RMSE of 3.19 among the ML models. Similarly, the RF model generated an R2 score of 0.74, indicating a good relationship between the projected and actual sugarcane yield data. As a result, it is evident that the GBR and RF algorithms are the best ML models for predicting the yield of sugarcane based on the research location's specific climatological variables. Additional research, extensive model validations, and integration into real agricultural systems are required to assure the effective use and acceptance of these models by sugarcane growers.