|KIVI, MARISSA - University Of Illinois|
|BLAKELY, BETHANY - University Of Illinois|
|MASTERS, MICHAEL - University Of Illinois|
|MIGUEZ, FERNANDO - University Of Illinois|
|DOKOOHAKI, HAMZE - University Of Illinois|
Submitted to: Science of the Total Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/12/2022
Publication Date: 5/10/2022
Citation: Kivi, M., Blakely, B., Masters, M., Bernacchi, C.J., Miguez, F., Dokoohaki, H. 2022. Development of a data-assimilation system to forecast agricultural systems: A case study of constraining soil water and soil nitrogen dynamics in the APSIM model. Science of the Total Environment. 820. Article 153192. https://doi.org/10.1016/j.scitotenv.2022.153192.
Interpretive Summary: Forecasting crop growth, yields, and impact on the environment is critial, particularly given the drastic impact that global warming and climate change is having on ecosystems around the world. However, there are limitations to our ability to make accurate crop forecasts that must be overcome. In this study, a modeling framework was designed and tested that takes existing data from experimental plots and uses the data to fine-tune a crop model to improve forecasting. Various important crop growth and ecosystem parameters were extracted from a long-term experiment at the University of Illinois Energy Farm. The new forecasting system that uses real data improved key modeling accuracy for soil moisture, nitrogen, water flow from fields, and nitrate runoff. This work demonstrates the potential benefits of using data from field experiments to fine-turn models.
Technical Abstract: As we face today’s large-scale agricultural issues, the need for robust methods of agricultural forecasting has never been clearer. Yet, the accuracy and precision of our forecasts remains limited by current tools and methods. To overcome the limitations of process-based models and observed data, we iteratively designed and tested a generalizable and robust data-assimilation system that systematically constrains state variables in the APSIM model to improve forecast accuracy and precision. Our final novel system utilizes the Ensemble Kalman Filter to constrain model states and update model parameters at observed time steps and incorporates an algorithm that improves system performance through the joint estimation of system error matrices. We tested this system at the Energy Farm, a well-monitored research site in central Illinois, where we assimilated observed in situ soil moisture at daily time steps for two years and evaluated how assimilation impacted model forecasts of soil moisture, yield, leaf area index, tile flow, and nitrate leaching by comparing estimates with in situ observations. The system improved the accuracy and precision of soil moisture estimates for the assimilation layers by an average of 42% and 48%, respectively, when compared to the free model. Such improvements led to changes in the model’s soil water and nitrogen processes and, on average, increased accuracy in forecasts of annual tile flow by 43% and annual nitrate loads by 10%. Forecasts of aboveground measures did not dramatically change with assimilation, a fact which highlights the limited potential of soil moisture as a constraint for a site with no water stress. Extending the scope of previous work, our results demonstrate the power of data assimilation to constrain important model estimates beyond the assimilated state variable, such as nitrate leaching. Replication of this study is necessary to further define the limitations and opportunities of the developed system.