cornXplain

Inspiration

Many good tools for crop modeling and explainability are available. However, they usually require coding skills to use. We want to reduce the entry barrier to explainable crop modeling, and create an interactive, browser-like environment that strips away the complexity of a GIS or programming language. By being accessible and responsive, exploring crop models can be fun.

What it does

We create a three-component dashboard linking around a speedy crop model. The three components allow the user to explore the input data, the model performance and variable importances, and counterfactual scenarios.

Data Explorer

Allows the user to visualize the data that goes into the model.

Model Explorer

A quick overview of the models structure with key performance metrics. This gives the user at-a-glance information about how the model functions, how good it is, and which variables it uses.

Scenario Explorer

Here, the user can change the environmental variables that are input to the model - within reason - to explore possible scenarios and how the model responds to changes in certain variables.

How we built it

Using R shiny and the many packages for spatial processing, data visualisation, and machine learning.

Challenges we ran into

The data were provided in numpy format, but we needed them in R. This made some preprocessing necessary.

Accomplishments that we're proud of

We learned a lot in just two days! From the team, the mentors, the R community, and from each other.

What we learned

How to build interactive apps, combine crop models, and make them explainable, all with free and openly available tools.

What's next for cornXplain

The dashboard can be deployed on a server to be accessible to anyone on the internet. There is potential to improve the design and interactivity of the dashboard. The built-in crop model could be improved by tuning and additional inputs.

Built With

• r