Improving regional simulations of processing tomato using remote sensing

Over the past fifty years, tomato has become one of the most extensively cultivated horticultural crops in the Mediterranean region. Climate projections for Italy indicate that temperature increases and rainfall changes will cause a 15% yield reduction in processing tomatoes, requiring an additional 85-110 mm of irrigation and 20-30 kg N ha^-1 to partially offset negative impacts. Mediterranean agriculture is particularly threatened by projected climate changes in temperature and precipitation patterns. Region-specific crop models, validated against local field data, are therefore critical tools for assessing yield risks and identifying effective agronomic adaptations. Conventional process-based crop models often rely on fixed transplanting or sowing dates and harvesting dates, which fail to reflect spatiotemporal variability in management practices. Such assumptions can lead to systematic biases in regional simulations and environmental assessments. Yet phenological observations (e.g., flowering, fruit set, harvest dates) are essential for parameterizing crop models, available data typically represent point locations or experimental stations rather than the field to regional scale resolution needed for spatially explicit modelling. Sentinel-2’s high temporal frequency and spatial resolution allow tracking of within-season crop development at field scale. This study aims to: (a) compare model performance using broad agricultural land masks versus pixel-level tomato identification; and (b) evaluate whether incorporating satellite-observed canopy development dynamics (greenness trajectories, growth stage timing) reduces uncertainty in simulated crop growth, water use, and nitrogen cycling processes including nitrate leaching risk.

We propose a simulation framework that combines the process-based model MONICA (Model for Nitrogen and Carbon dynamics in Agro-ecosystems) with earth observation data for processing tomatoes in the Emilia Romagna region, a major tomato production area in Italy. MONICA was calibrated and validated using four years field trials and two years on-farm data from 49 fields. We integrated two remote sensing inputs: (i) field scale processed tomato masks, and (ii) dynamic transplant and harvest dates extracted from Sentinel-2 EVI time series (validated against on-farm data, R²=0.90). We conducted regional simulations (2007-2023) comparing four model set-ups: fixed transplant and harvest dates with basic cropland mask, fixed dates with tomato masks, dynamic dates with tomato masks, and modified dynamic dates with tomato masks for sensitivity tests on transplanting date.

Our research results indicate that employing specific tomato field maps combined with dynamically determined growing periods significantly improved yield simulation accuracy compared to basic cropland mask (reducing RMSE by 24%) and specific maps without consideration of remotely sensed growing season dynamics (reducing RMSE by 10%). Incorporating remote sensing data and tomato maps into the MONICA crop model also improved the model’s ability to capture yield anomalies as an indicator of its sensitivity to climatic signals, with a 24% reduction in RMSE. Integrating remote sensing-derived growing periods into crop models resulted in a wider range of simulated values, enhancing the model’s capacity to simulate nitrate leaching under real-world conditions.

This study demonstrates that using remote sensing data to inform crop models significantly enhances the understanding of dynamic growth patterns, thereby supporting regional yield estimation and nitrate leaching simulations, while providing crucial insights for agricultural resource management.