Crop-yield models based on vegetation indices such as the normalized difference vegetation index (NDVI) have been developed to monitor crop yield at higher spatial and temporal resolutions compared to agricultural statistical data. We evaluated the model performance of NDVI-based random forest models for sugar beet and potato farm yields in northern Belgium during 2016–2018. We also evaluated whether weather variables and root-zone soil water depletion during the growing season improved the model performance. The NDVI integral did not explain early and late potato yield variability and only partly explained sugar-beet yield variability. The NDVI series of early and late potato crops were not sensitive enough to yield affecting weather and soil water conditions. We found that water-saturated conditions early in the growing season and elevated temperatures late in the growing season explained a large part of the sugar-beet and late-potato yield variability. The NDVI integral in combination with monthly precipitation, maximum temperature, and root-zone soil water depletion during the growing season explained farm-scale sugar beet (R2 = 0.84, MSE = 48.8) and late potato (R2 = 0.56, MSE = 57.3) yield variability well from 2016 to 2018 in northern Belgium.