Coupling hydrological models with plant physiology is crucial to capture the feedback mechanisms occurring within the soil–plant–atmosphere continuum. However, the ability of such models to describe the spatial variability of plant responses to different environmental factors remains to be proven, especially at large scales (field or watershed). We used an innovative three-dimensional soil–plant model to quantify temporal and spatial variability of crop productivity at the field scale, and we assessed simulation accuracy by comparison with spatially distributed crop yield measurements. A 25-ha field located in the Venice coastland, Italy, cultivated with a maize (Zea mays L.) crop and characterized by a highly heterogeneous soil subject to salt contamination, has been extensively studied by soil sampling, geophysical surveys, and hydrological monitoring. Based on these observations, field-scale simulations of soil moisture dynamics coupled with plant transpiration, photosynthesis, and growth were run and compared with crop yield maps of different growing seasons. The model captured the observed crop productivity (grain yield varying between 2 and 15 Mg ha−1), but the accuracy of the predicted spatial patterns was limited by the available information on soil heterogeneities. Further model uncertainties are related to the characterization of the rooting systems and their responses to environmental factors (soil characteristics, precipitation) that were shown to be crucial to describe the effect of drought conditions on growth processes. These results demonstrate that large-scale mechanistic simulations of soil–plant systems require a trade-off between site characterization, model processes, and computational efficiency, offering an open challenge for future ecohydrological research.