MAGiC downscaling training

This is the downscaling component of the MAGiC system. It uses output from magic-ensemble: the ensemble workflow runs SIPNET at a set of design points, and this workflow turns that into carbon estimates for fields with the required inputs, then aggregates to county and state.

The design points reduce the number of SIPNET runs needed and are chosen to span the state’s covariate space, a Random Forest learns covariates to model output at those points, and applies it everywhere else. One forest is fit per ensemble member, so the ensemble’s parameter uncertainty propagates to the field level rather than collapsing to a single number.

Each notebook is a self-study demo to work through beforehand, then discuss in the live session.

Topic	Audience	Objectives	Notes
Environment setup	Everyone	Conda, AWS, and repository setup	Needed before the inventory notebook
Inventory downscaling	MAGiC users	Turn an ensemble run into field, county and state carbon	Demo dataset covers four Delta counties
Scenario downscaling	MAGiC users	Practice-change arms and scenario deltas	Deferred; same four commands with a longer scenario list
Design point selection	Data managers	How the design points are chosen	Scripts 010-022, upstream of the ensemble and not a CLI command

Commands

The extract, downscale, and analyze commands run the corresponding workflow steps.

Command	What it does
get-demo-data	Downloads and extracts the demo data package. Run once, before prepare.
prepare	Stages ensemble output and data layers into run_dir.
extract	Reads SIPNET NetCDF and writes monthly means in EFI long format.
downscale	Random Forest downscaling to all fields, then county and state aggregation.
analyze	Diagnostics, uncertainty, maps.

Data flow

flowchart LR
  ENS["magic-ensemble<br/>SIPNET at design points"] --> PREP["prepare"]
  PREP --> EXT["extract"]
  EXT --> DS["downscale"]
  DS --> AN["analyze"]

Audiences

Following the same roles as the MAGiC training index:

• Everyone – environment setup.
• MAGiC users – running the workflow and reading its output. Assumes the setup notebook is done and you are comfortable on the command line.
• Data managers – regenerating the design points or the covariate layers.