Comparison of the machine learning and AquaCrop models for quinoa crops

Abstract

One of the main causes of having low crop efficiency in Peru is the poor management of water resources; which is why the main objective of this article is to estimate the amount of irrigation water required in quinoa crops through a comparison between the machine learning and AquaCrop models. For the development of this study, meteorological data from the province of Jauja and descriptive data of quinoa crops were processed and a simulation period was established from June to December 2020. From the simulation carried out, it was determined that the best model to predict the required irrigation water is the Adaptive Boosting (AdaBoost) model in which it was observed that the mean and standard deviation of the AdaBoost models (mean = 19.681 and SD = 4.665) behave similarly to AquaCrop (mean = 19.838 and SD = 5.04). In addition, the result of ANOVA was that the AdaBoost model has the best P-value indicator with a value of 0.962 and a smaller margin of error in relation to the mean absolute error (MAE) indicator with a value of 0.629. Likewise, it was identified that, for the simulation period of 190 days, 472.35 mm of water was required to carry out the irrigation process in red quinoa crops.