Crop models allow any stakeholder in agriculture to explore potential management practices, crops, yields, and more before moving forward with expensive and time-consuming field choices. A Crop Model is the set of mathematical equations describing how crop genetics, crop management practices, and the environment interact together to determine crop growth. It is in essence a mathematical representation of a cropping system.
According to the USDA, “Crop simulators are computer programs that mimic the growth and development of crops. Data on weather, soil, and crop management are processed to predict crop yield, maturity date, the efficiency of fertilizers and other elements of crop production. The calculations in the crop models are based on the existing knowledge of the physics, physiology, and ecology of crop responses to the environment.” By using a crop model, farmers can experiment with many different types of trials and treatments on a computer.
Being able to run various simulations is a vital tool in decision-making to assess the impacts of climate change or variability and management practices on productivity and environmental performance of alternative cropping systems, to promote better and sustainable agriculture. Crop models are quicker and less expensive ways to investigate different management changes on the final yields. These tools are helpful decision support systems to assess the risk and economic impacts of management strategies in agriculture.
Different Types of Crop Models
Simple Models
Crop models have traditionally been constructed for academic purposes and therefore tend to focus narrowly on the main focus area of the investigator (i.e. photosynthesis, soil hydrology, etc). As a result, these types of crop models tend to oversimplify relationships between a limited number of variables, and therefore are only applicable to specific geographical or use cases. Another challenge of these simple crop models is that any modification of them to incorporate new or missing variables or interaction effects of old variables with any new ones requires a complete overhaul of the model itself.
Statistical Models
With extensive investments in agricultural data collection – whether through IoT-enabled planters and harvesters on the field, weather stations, satellites, and/or soil readings – the world has an opportunity to develop better models for agricultural ecosystems. Some companies are collecting this data and applying statistical data science approaches that leverage black-box machine learning techniques to extract statistical markers as predictors of crop yields and other outcomes. In this type of modeling system, the feature that has the most leverage on the outcome in many cases does not reveal what the underlying process is that is producing the outcome. Therefore, the model can easily be overfit to statistical markets that are not related to the true underlying mechanism.
Mechanistic Models
Mechanistic models are a better way to parameterize natural processes and their interactions and impact on agricultural ecosystems, while also allowing for greater accuracy, model deployment, and cycling of improvements. Mechanistic models permit inspection of daily intermediate variables, and when outcomes are counterintuitive, one can drill down into the time series of plant and soil state over a season or over several seasons and determine, in a causal way defined by the various functions, what led to a particular outcome. Similarly, this type of model can be used to understand new scenarios where data is scarce or missing – for example, with a structured representation of the role of the plant’s germplasm in field trials outcomes, its behavior in new , real-world environments can be represented.
The CIBO Model
The CIBO model began with a mechanistic crop model (SALUS: Systems Approach to Land Use Sustainability) capable of simulating the performance of various crops at many scales. The initial version of SALUS was developed at Michigan State University and has been the subject of 20 years of testing across hundreds of fields in 46 countries, more than 25 Ph.D. dissertations, over 230 peer-reviewed journal articles, and thousands of academic citations. CIBO has an exclusive license to SALUS which has continuously been improving, expanding to new crops and scaling to address the practical needs of agriculture from subfield to continental scales.
CIBO’s crop model is part of a larger simulation engine which uses a combination of farmer reported data, government and academic statistics, and remote sensing to build a detailed scenario that describes genotypic (crop parameters representing genotypic potentials of the crop), environmental (weather, soil physical and chemical properties), and management (planting date, fertilizer application dates and amounts, tillage date, depth, and material, etc.) conditions of the growing crop. Based on these inputs, the crop model calculates the plant growth stage, plant leaf area, solar energy absorbed through the leaves, biomass accumulated in different plant tissues, and water and nutrient uptake by the roots, and saves outputs for that day. These variables are calculated at every time step until the crop matures.
Some of the benefits of CIBO’s Differentiated Mechanistic Model:
- Effectively utilize information garnered from precision agriculture approaches via a combination of data, crop, and environmental sciences.
- Account for complex, interacting biological systems in a diversity of geographies and crops, even in data scarce situations.
- Understand plant physiological processes in order to produce daily intermediate variables that contribute to yield and other outcomes.
- Carry out retroactive analysis, counterfactual scenarios, and forecasting for scenario planning.
How Crop Models are Validated
CIBO is perpetually seeking new opportunities to validate and build out the robust modeling capability. Through continuous validation, CIBO’s insights accurately identify potential limitations and assumptions asserted in customer “what if” questions to ensure the model takes into account all types of potential situations and conditions. The data science team is regularly validating all the model components, including input, processing, and reporting to ensure the model has the most accurate inferences.
Innovation Research Partnerships
To help test and improve the accuracy of the model, the company partners with progressive farming leaders to run the various scenarios on fields to validate CIBO’s software. Partners are chosen for their openness to innovative ideas and commitment to the collection and analysis of data to guide strategic managerial decisions. The growers share their detailed data collection and analysis of the strategic decisions for their farming operations.
Validation with USDA Data
The CIBO model not only can predict the performance of a single plot but can be scaled to various geographies and thousands of fields. Using the monthly reports from the United States Department of Agriculture (USDA), the team compares CIBO’s simulated numbers in order to test and validate the model. By comparing state-wide numbers over many years with the model outputs, the team can ensure outputs are applicable across counties and not just on a specific farm.
University Research Studies
To validate the model and give CIBO confidence that it behaves well across a wide range of management scenarios and weather conditions, simulations are run compared to global scientific research. Using data from the scientific literature, CIBO’s scientists evaluate and validate the mechanistic components of the model. These assessments ensure that the model responds correctly to things like variable nitrogen additions and a range of rainfall and irrigation scenarios. The opportunity to continuously validate the model components helps to improve the validation assessment to help increase reliability.
