We collected data from experiments conducted in Waseca, Le Sueur, St. Paul, Grand Rapids, and Crookston, during the 2023 and 2024 growing seasons. These experiments covered a broad range of environmental conditions and MG’s ranging from 0.05 to 2.5, resulting in soybean yields ranging from less than 25 to more than 100 bushels per acre.

After calibration, the model predicted the timing of flowering (R1) and physiological maturity (R7) with high accuracy. The mean absolute error was less than four days for R1 and less than five days for R7, indicating a strong model performance in capturing crop development across a wide range of locations, planting dates, genotypes and water regimes.

Yield prediction was also quite accurate, with a mean absolute error just below 7 bushels per acre – a strong result given the yield range explored in the dataset (Figure 1). There was good agreement between observed and predicted values in a wide range of varieties and conditions. This level of accuracy indicates that the model can serve as a valuable decision-support tool for evaluating management impacts across diverse environments.

This tool would enable us to begin answering key questions such as: What is the potential and attainable yield for a specific location? or, How the planting date interacts with MG for a specific location? (See example for Crookston in Figure 2). In addition to estimating average outcomes, the model can also assess variability over time and estimate the probability of achieving a given yield level across years.

This project successfully calibrated and validated a soybean crop model useful to Minnesota’s diverse growing environments and management. By integrating detailed soil, weather, management, and genetic data into the DSSAT-CROPGRO simulation model, we are generating a reliable decision-support tool capable of simulating yield outcomes with relatively high and useful accuracy. In this way the calibrated model’s ability to predict phenology and yield across years and locations – while accounting for interactions between management and environmental condition – makes it a valuable resource for guiding agronomic decisions, supporting research, and informing policy discussions around soybean production. Importantly, this tool is not static. As more high-quality field data continue to be generated across Minnesota through research trials and collaborations with farmers, the model can be further refined, improved and expanded. This ongoing feedback loop of calibration and validation will help enhance its accuracy and applicability, supporting increasingly precise and site-specific recommendations over time.

Figure 1: Simulated vs Observed yield for soybean growing in different locations, years, planting dates, water management conditions and cultivars differing in (MR). The dashed line is the 1:1 line. RMSE: root means square error. MAE: Mean absolute error. n=146

Figure 2.