A machine learning framework to estimate crop coefficient dynamics of citrus orchards

Abstract

Accurate estimations of vegetation biophysical parameters such as crop coefficient (Kc) are necessary to estimate crop water requirements, optimize water use efficiency, and implement smart irrigation strategies, especially in Mediterranean climatic zones, where the effects of climate change on water resource availability are evident. Therefore, obtaining accurate Kc values is challenging due to the lack of spatial and temporal resolution in certain agro-meteorological data, such as actual evapotranspiration or vegetation indices, and the high costs of acquiring such data in the field. This study aims to develop a framework that harnesses the potential of machine learning (ML) algorithms in conjunction with the Seasonal-Trend decomposition (STD) algorithm to effectively estimate the dynamics of Kc time series. First, three different ML algorithms are employed, namely Multi-Layer Perceptron (MLP), Random Forest (RF), and k-Nearest Neighbors (kNN), to predict actual evapotranspiration (ETa) in Mediterranean citrus orchards. Then, after predicting ETa, crop coefficient values are derived using FAO-56 guidelines, and their accurate dynamics are estimated using anomaly detection algorithms, such as Isolation Forest and STD. The proposed framework was validated using four baseline models: field measurements, two theoretical models (FAO-56 and its most recent revision of Kc values), and an empirical model. The RF algorithm provided the best performance, achieving a root-mean-square error (RMSE) of 0.13 using simple agro-meteorological variables acquired by low-cost instruments, such as global solar radiation, maximum and minimum air temperature, maximum and minimum relative air humidity, and wind speed measured at 2 m from the ground. The proposed framework does not employ satellite images and requires less data and computational capacity compared to traditional methods, making it an excellent solution for implementing Kc estimation in smart agriculture.