Uncertainty

Sensitivity analyses of the SIPNET ecosystem model used in the MAGiC inventory runs:

• a local (one-at-a-time) analysis that ranks parameters by their share of the parameter variance of modeled soil carbon and nitrous oxide flux, and
• a global (Sobol) analysis that partitions the variance of those outputs among parameters, initial conditions, meteorology, and management.

Reports and training: https://ccmmf.github.io/uncertainty/

Requirements

• the pecan-all conda environment (R, PEcAn, yq, python3)
• a SIPNET binary, such as the sipnet.git built by the ensemble workflow
• the inputs of a prepared ensemble run: site list, ERA5 meteorology, initial conditions, management events, and PFT priors

Model runs are submitted to Slurm or SGE as array jobs, or run with GNU parallel in the current allocation, as set by pecan_parallelism_mode in the configuration.

Running

Copy example_user_config.yaml, point external_paths at the ensemble run and the SIPNET binary, then run from the repository root:

./magic-uncertainty prepare            --config config.yaml
./magic-uncertainty local-sensitivity  --config config.yaml
./magic-uncertainty global-sensitivity --config config.yaml

./magic-uncertainty help lists the commands. --verbose prints each script call. Each command writes a log to run_dir. The training pages walk through a demo at two design points.

Outputs

Path in run_dir	Content
local/local_sensitivity.csv	CV, elasticity, and variance of every parameter at every design point
local/figures/	local analysis figures and parameter_shares.csv
global/global_sensitivity.csv	first-order and total-order indices with bootstrap intervals
global/figures/	global analysis figures and summary tables

Repository layout

magic-uncertainty            command line interface
uncertainty_manifest.yaml    steps, their inputs and outputs, and fixed settings
example_user_config.yaml     settings for one run
scripts/                     one script per step (001, 011-013, 021-023) and launchers
analysis/                    reports, their figure scripts, and figures
docs/training/               training pages
data_raw/                    settings templates and climate region boundaries
R/                           functions used by the scripts

Site

The site is built with Quarto from _quarto.yml:

quarto render
quarto publish gh-pages