SuperMaaS Example Containers for DSSAT-pythia

These are the Expert Modeler supplied files and the Dockerfiles used to run DSSAT-pythia.

Prerequisites

• Download and unpack the large data repositories.
  • Base: global soil databases, other generic shipped data.
  • Ethiopia: shape files, raster files, and templates for Ethiopia.
  • Weather: DSSAT-formatted weather files for Ethiopia.
• Docker

Structure

This repository contains what could be the results of multiple model registrations. The crop for the base model is Maize.

• baseline - The baseline model to which all other models in this example set should be compared against. Contains 4 management practices (irrigated/high nitrogen inputs, rainfed high N, rainred low N, rainfed 0N).
• subsidy - An example of multiple fertilizer subsidies, which is additional fertilizer compared to the baseline. values are 10,25,50 kg/ha.
• forecast - An example of running a forecast (in this case a backcast based on 2018 data with proxy years from 1984-2017).
• suitability - An example of planting Teff where Maize is grown. Uses the (unregistered) Teff model base and replaces the rasters for harvest area to the Maize areas.

The includes directory is custom cultivar files which would be provided by the Expert Modeler.

Files under the above directories are mounted under the /userdata/ directory inside the image.

Assumed directory structure

For the purposes of this documentation, the following directory structure is assumed.

.
├── baseline
├── data
│   ├── base
│   ├── ethiopia
│   └── weather
├── forecast
├── includes
├── outputs
├── subsidy
└── suitability

Customizing the JSON files

Because we don't know the parameters of the system running DSSAT-pythia, the following should be modified inside the JSON file prior to building the image.

• cores - Number of cores to use
• threads - Number of threads to use (Some testing has shown that using 1/2 cores for threads works most efficiently)
• sample - Number of valid pixels to run per management. (This can be removed to run the whole country)

Building the images

For these examples, we can execute docker build from within the above directory structure.

docker build -t supermaas-dssat-eth-baseline -f baseline/Dockerfile .

Executing the images

docker run --rm  -it -v ${PWD}/data:/data -v ${PWD}/outputs:/outputs supermaas-dssat-eth-baseline --all /userdata/eth_maize_baseline.json

Outputs

All outputs are generated in the outputs/ directory. Each image has a different output directory underneath outputs/ in this example. The csv file at the outputs/<image>/ directory is a trimmed and aggregated file of all the CSV files underneath each directory.

Future Thoughts

• In order to simplify the execution of the model, the JSON files could technically be named the same in each directory.
• There needs to be a dynamic way to give the desired cores and threads.
• This assumes that the Expert Modeler is setting up the baseline conditions prior to indicating which procedures (subsidy, forecasting, etc.) should be shimmed into the JSON file.
• Currently, these are all just JSON files with minor differences between them. There should be a way to take the differences and abstract them into the procedures.