Monitoring and Managing Spatial-Temporal Variability on Farm: A Crop Modelling Approach for Nitrogen Management

Soil properties within fields can vary as much as they do across regions, causing significant spatial and temporal variation in crop nutrient demand. For wheat, the economic optimal nitrogen (N) application rate has been shown to vary by more than 150 kg N ha-1, with N fertiliser recovery ranging from 30% to 100% within fields. Spatially appropriate N fertiliser applications can therefore enhance nutrient use efficiency, forming a key foundation of precision agriculture. However, current spatial management technologies focus primarily on economic return, with little consideration for environmental consequences, making it challenging to balance these often-conflicting demands. Crop canopy sensors have seen limited adoption and applicability, particularly when co-limitation by other nutrients or water occurs. Cropping system models, which describe system changes in response to weather and management practices, provide a means to quantify interactions among crops, soil, and system losses. Despite this, such models have not been evaluated or applied to UK wheat crops at the sub-field scale. To address this gap, a systematic review was conducted to identify key themes for developing a methodology applicable to UK wheat. High-resolution soil and agronomic data from a long-term experiment were used to parameterise the 'Sirius' crop model. Sirius simulated observed spatial-temporal variation in grain N uptake (RMSE: 21.3 kg N h ha-1, RRMSE: 19.5%) to an acceptable level of accuracy, demonstrating its potential as an on-farm N management tool. To facilitate farm-scale application, a framework was developed using yield map data to reduce spatial data demands for model parameterisation. However, an important finding revealed significant errors in on-farm yield map datasets (RMSE: 1.03 t ha-1), indicating the need for caution when evaluating and calibrating technologies based on such data. The methodologies were then combined, demonstrating how targeted sampling in homogenous management zones could be used to spatially parameterise and apply the Sirius model for informing spatial N management across 30 ha. Simulated optimal N rates across management zones were shown to range from 160 to 216 kg N ha-1. This research has demonstrated the applicability of crop simulation modelling for spatially relevant on-farm N management decisions. Future work could expand this approach by incorporating rotational modelling, validating model-derived inputs, and applying experimental modelling to inform broader soil management decisions.