Session: (Mathematical) Modeling

Modeling is a large topic!

  • What types of models are we talking about?
    • Diagnostic vs prognostic
    • Empirical vs mechanistic
  • Building new models/improving vs how to use existing models

User group:

  • Modeling soil carbon - what exists to use for my tool?
  • Learning the rigor/math to model your own data
  • Hydrology/stormwater
  • Interoperability of models
  • What types of problems do we need models for?
  • How to bring remote sensing/spatial data into models as inputs
  • Bringing accessibility to a broad user base of existing models
  • What information in our models is already there/can be adapted that farmers would be interested in?

Main tasks:

  • Core concepts

    • The guts of DNDC: a series of equations that model the microclimate of soil microbes. The initial parameters of the equations come from empirical research and first principles of science.
    • Multiple resolutions of data quality: models can fill in with good guesses where data aren’t available, or can give better outputs if you have a full inventory of data for parameters. AKA how good is good enough for your inputs?
    • All models are wrong, but some models are useful.
    • Swappable model components: you need to know the interface to do it.
    • Interface design is a computer science specialty, we need to share that knowledge.
    • How do you go from natural language request for information to a logical workflow that can be coded as inputs and outputs?
    • Who are the target audience for models? Farmers, crop consultants, researchers? Mostly researchers so far, but there’s hunger to implement them in consumer-facing tools.
    • How do you expand the inference space of a model? They were designed using empirical data from field sites, but not every possible field situation.
    • Models need to be plugged into decision support tools, so that you can see outcomes before action.
    • We need an infrastructure for plugging them together (farm management platform?)
    • Calibration is a fundamental challenge: It’s possible to get the right outputs for the wrong reason when there are components you’re not simulating. Need to be cognizant of double-dipping data so you’re not introducing confounded bias.
    • IFSM is a “typical” model, where the code is not open source, but the basic equations are presented textbook-style. Hard to reproduce.
    • Who owns a model? A research group? A consortium? A PI at a university or agency? A company?
  • Accessibility

    • What is the role of GOAT in making models extensible and interoperable? A working group?
    • Do you drive accessibility by creating demand? Making models more economic, so farmers will want to use them.
    • What tools are usable to a farmer? They’re familiar with spreadsheets, but they aren’t flexible enough for models.
    • Values: farmers are businesspeople, but they’re also people who care about XYZ environmental/social externalities.
    • Values can also be dollars-and-cents decisions. Consumers may be willing to pay premiums for organic/dietary change for livestock, etc.
    • Understanding year-to-year variability can be a big challenge when looking at model outputs. Short-term development goals can be detrimental to finding realistic values. How does error/uncertainty propagate through a model?
    • What data formats are useful for crosstalk between models? Getting data into a prespecified format can be a challenge for users.

Next steps

  • Forum thread
  • Skills trade: modelers need validation data, and users need model APIs
  • Repository for finding models