Using Cross-Scale Data to Constrain an Agro-Ecosystem Model to Produce Estimates of Miscanthus Production at a Field-Scale

Over the last decade, 400 hectares of miscanthus in eastern Iowa has been planted to produce several bioproducts and bioenergy, and due to this miscanthus production, precision yield monitor data has been collected from these fields which show high intra-field yield variability due to patchy rhizome death. However, modeled interannual and intra-field yield variability is poorly constrained. Fine (101 m2) scale data from research plots are often outside the range of precision yield monitor data while coarse (104 m2) scale estimates from regional models may have similar average yield compared to precision yield monitor data but the variability within and between fields is not well captured. This requires resolving environment and management interactions and representing those interactions in a framework capable of resolving variations at scales useful to miscanthus producers. Therefore, the objective of this project is to compare commercial field-scale chopper miscanthus yields across eastern Iowa to modelled yields from a mechanistic agro-ecosystem model (Agro-IBIS). Agro-IBIS was calibrated using leaf-level gas exchange, biomass, and leaf area data that were collected from a miscanthus field experiment in central Iowa. Using the gridded soil survey geographic database and a daily high-spatial resolution surface meteorological dataset as inputs into Agro-IBIS, we simulated yield at several commercial miscanthus fields in eastern Iowa. Averaged over 20 site-years, the model overestimated yields by 37% with individual modeled site-years ranging from -23.5% to 291% of the observed yield. We hypothesize that the model overestimation is due to model poorly representing areas in the field where plants were killed. To test this hypothesis, Sentinel-2 bottom-of-atmosphere reflectance is being used train a machine learning model that detects areas in fields where there is an absence of miscanthus. This binary land classification of miscanthus versus no miscanthus is used to update the model predictions. This work will evaluate the model's ability to simulate the potential yield of the field assuming there are no plants killed and allow for more strategic placement of this bioenergy crop candidate that would optimize profit for the producer while reducing nitrate leaching and sequestering carbon.