By Angie Peltier and Anibal Cerrudo
On January 29, 2025, Seth Naeve, University of Minnesota Extension soybean agronomist and Anibal Cerrudo, a visiting professor for the last 3 years in the Naeve lab joined UMN Extension crops educator Angie Peltier for a discussion about how using crop modeling can help soybean producers in Minnesota gain important insights into how their management choices could impact yield. This was the fourth weekly episode of the 2025 Strategic Farming: Let’s talk crops! webinars. The series runs through March.
To watch this and other episodes: http://z.umn.edu/StrategicFarmingRecordings
The Naeve Lab is working on two types of modeling
Crop modeling incorporates weather, soil condition, crop genetics and management practice data together to estimate crop yield with the end goal of helping farmers make data-driven decisions. Once enough applied, field-scale data has been compiled and fit to a model, it can run various crop production scenarios virtually. While some of the larger ag companies have developed paid, proprietary tools based on crop modeling of how management conditions can impact yield of their own varieties, Cerrudo and Naeve’s crop modeling work isn’t designed to work for a specific set of soybean varieties, but is rather designed to provide an open-source set of tools to aid in farmer decision-making based on Minnesota conditions. As additional data informs a model, the model’s predictive ability improves.
The Naeve Lab is mostly using mechanistic models. Mechanistic models are process-based, relying on a working understanding of the many processes that drive crop growth and development. This knowledge is used to develop a series of equations to explain and then simulate these processes. One of the strengths of this type of model is that because agronomists understand these processes and relationships, the results are easily explained and interpreted; the results of mechanistic modeling also tend to be very accurate. The need for both modeling and agronomic expertise are two weaknesses associated with mechanistic models.
On the other hand, data-driven models use artificial intelligence (AI) to better understand patterns and relationships between related factors (historical crop performance, weather conditions, soil conditions), management practices and yield. AI is actually ‘learning’ from the data by looking for relationships without any pre-existing understanding of the interrelatedness of the processes involved or about how any of the factors may impact crop growth and development or yield. This sort of model may help researchers to choose topics for future study or observe some previously uncharacterized cause and effect relationship between crop production factors. One significant weakness in AI models is that there is generally a lack of transparency regarding how the results are derived. This can limit their generalization and lead an agronomist to an erroneous conclusion.
As these models are not based on an understanding of relationships between multiple factors, there may appear to be a relationship where none exists. For example, one may find a correlation between the stork population density and the number of babies being born over time, when a cause and effect relationship does not exist.
Cerrudo explained a mechanistic model that they are adjusting to describe crop growth and development and yield. Photoperiod and temperature drive crop development and the sun’s radiation drives crop growth. All three impact crop biomass and partitioning of photosynthesis-derived resources to the crops’ roots, stems, leaves, flowers, pods and seeds throughout the growing season. Each aspect of the model needs to (nearly perfectly) describe the relationship between every other aspect of the model. For example, if leaf area is underestimated, this impacts assumptions regarding the amount of radiation that the leaves are able to intercept and the sugars that the crop is able to produce per day.
What modeling can tell ss about planting date
Models designed to predict the relationship between planting date and yield will help producers understand when it is best to plant to balance risks of stand losses due to planting too early into cold, wet soils or having the crop lost to a frost and risks of losing out on some of the crop’s yield potential by delaying planting. To illustrate how planting date impacts yield, Cerrudo described the results of one planting date experiment in St. Paul in 2023 in which every other factor other than planting date was controlled so that there would not be another yield-limiting factor in the mix. In this single trial from a single year, on average 0.6 bushel per acre per day was lost by delaying planting through the end of May; on average 1.5 bushels per acre per day was lost when planting was delayed until after the end of May.
While it is impossible to extrapolate beyond this single experiment, we actually can extrapolate beyond this 2023 experiment using models that link the processes about which we have a great deal of knowledge. For example, we know that the Earth is tilted on its axis and the different orientation of the Midwest as it rotates around the sun causes Minnesota to experience differences in solar radiation and temperature leading to our four seasons. When we delay planting date, we are shortening the critical stages for yield determination because of a shorter photoperiod. We are also pushing these stages into autumn when the crop will experience less solar radiation and lower temperatures, resulting in less growth and therefore less yield. We have long-term weather records that can tell us a lot about the variability in temperature and radiation observed on each day of the growing season for a specific location to put numbers on these processes by using modeling.
Agronomists can experiment to build large data sets over many years, locations or location-years. Recently retired Extension IPM specialist Bruce Potter conducted soybean planting date experiments at the Southwest Research & Outreach Center in Lamberton over a period of 25 years, determining that when there are no other limiting factors, delaying planting until June 2 results in 0.3% less yield per day, and delaying until after June 8 results in 1.3% less yield per day. It is data sets like this that have been fed into the CROPGRO model that the Naeve Lab is using and used to validate and calibrate the model. By running simulations of the various processes related to yield, crop models can simulate many more realistic scenarios looking at the relationship between planting date and yield than would be possible to conduct in real life by a university agronomy lab running on-farm small plot or strip trials. This sort of better understanding of the primary drivers of soybean yield potential for a particular location can result in researchers being able to provide probabilities of the likelihood of specific management practices to impact yield.
Ideally, we would be able to answer every agronomic question on every Minnesota farm field over a period of years to provide field-specific management recommendations. However as this would be cost and time prohibitive, provided that the model can be validated and calibrated using data from a location near to your farm, it can be used to better predict the relationships between (for example) row spacing or maturity group and yield on specific fields. This has the potential to help people to model potential outcomes of changing one or more soybean production practices on their own farms.
Yield Gap Approach
This approach begins with potential yield, or the yield the crop can achieve if only solar radiation, temperature and variety are the limiting factors and everything else is as ideal as it can be. Attainable yield is essentially the same as potential yield except the crop is rainfed and so soil moisture may be yield-limiting. Actual yield is just that, the yield that one obtains in most Minnesota fields. The yield gap is the difference in yield between the attainable and actual yield. The yield gap means that there is still room to improve yield.
In 2023 and 2024, Cerrudo and the Naeve Lab worked to determine the attainable yield in Minnesota soybean fields. After 19 farmers did everything they had planned on doing in a field to raise a soybean crop, the team went to the field to add additional management above and beyond the farmers’ standard practices. One treatment added all of the macro and micronutrients that the crop may use to ensure that there were nutrients were not a yield-limiting factor. Another treatment was to add to the farmer practices and additional fertility, crop protection products to ensure that neither diseases, weeds nor pests were limiting yield. The experiments were very important to help to train the model because the team scouted the field all season long and so could ensure that all of the data going into the model wasn’t compromised by some unknown crop production issue.
The team was surprised to find that in only 1 of the 19 fields were nutrients limiting attainable yield, concluding that if these fields are representative of all Minnesota fields there is little that can be done to increase yields. Perhaps if these fields are representative, the best strategy would be to improve efficiency so that one can produce the same yield with less expense, rather than chasing yield increases that are unlikely to occur. Now, the team is also able to use their model to estimate attainable yield and extend and multiply this yield gap approach to different situations (fields) along the state and continue to assess the potential for intensification in our production systems.
Source : umn.edu