Alfalfa yield estimation using the combination of Sentinel-2 and meteorological data

Context: Alfalfa (Medicago sativa L.) is one of the world's most important forages for livestock feeding. Timely yield estimates could provide information to guide management decisions to improve production. Since alfalfa crops typically undergo multiple harvests in a year and demonstrate rapid regrowth, satellite remote sensing techniques present a promising solution for alfalfa monitoring. Objective: To generate alfalfa yield estimation models at three phenological stages (early vegetative, late vegetative, and budding stages) using vegetation indices (VIs) derived from satellite Sentinel-2 images and their combination with meteorological data. Methods: We analyzed fields located in Navarre (northern Spain) over two consecutive seasons (2020 and 2021). To generate the yield estimation models, we applied a conventional multilinear regression and two machine learning algorithms (Least Absolute Shrinkage and Selection Operator - LASSO and Random Forest - RF). Results: Regardless of the statistical approach, the three phenological stages were not optimal when either VIs or meteorological data were used singularly as the predictor. However, the combination of VIs and meteorological data significantly improved the yield estimations, and in the case of LASSO model reached percentages of variance explained (R2) and normalized root mean square error (nRMSE) of R2= 0.61, nRMSE= 0.16 at the budding stage, but RF reached a R2= 0.44, nRMSE= 0.22 at the late vegetative stage, and R2= 0.36, nRMSE= 0.24 at the early vegetative stage. The most suitable variables identified were the minimum temperature, accumulated precipitation, the renormalized difference vegetation index (RDVI) and the normalized difference water index (NDWI). The RF model achieved more accurate yield estimations in early and late vegetative stages, but LASSO at bud stage. Conclusion: These models could be used for alfalfa yield estimations at the three phenological stages prior to harvest. The results provide an approach to remotely monitor alfalfa fields and can guide effective management strategies from the early development stages.