Estimation of canopy water content in maize using machine learning and multispectral vegetation indices: comparison of Adaboost regression and other methods

Maize plays an important role in global agriculture, with significant productive value in various regions. Effective water management is of paramount importance in this crop as water stress at specific vegetative stages (e.g., V6) can have a detrimental impact on its development and productivity. Accordingly, we aim to compare various regression models with the aim of indirectly estimating the value of canopy water content (CWC) and equivalent water thickness (EWT) in maize. To this end, vegetative indices were calculated and related to CWC and EWT using different wavelengths [visible, red, red-edge (RE), and near infrared] obtained from Sentinel satellite images. The indices calculated using the red edge band demonstrated the strongest correlation with water content. Four regression models were constructed using different indices as input variables. The regression models utilized in this study were multiple linear regression (MLR), ridge regression (RR), random forest regression (RFR), and AdaBoost regression (AR). Ensemble models, such as the AR, are seldom cited in the literature (there are no studies in the literature on measuring CWC and EWT for maize) and may serve as alternatives for measuring different crop development parameters. This hypothesis was validated in this study by demonstrating the superior performance of this model compared with the others used here. In conclusion, the AR model demonstrated superior performance compared with MLR, RR, and RFR, with an R2 of 0.991 for CWC and 0.972 for EWT. RFR exhibited a value of 0.982 and 0.948 for these two water content indices. Therefore, the methodology employed proved effective in obtaining both CWC and EWT over the analyzed period.