rsofun_doc

This repository contains the code for producing results published in Paredes et al. (2025). This study documents the rsofun package and reports results from a calibration of the P-model to GPP and traits data. A first version of the paper is published as a pre-print (see reference below). The study is currently under review at Geoscientific Model Development. Code contained in this repository serves for the revised version of the study.

Repository contents

• analysis/ contains scripts for site selection, model calibration, calibration diagnostics, and model evaluation.
• fig/ contains publication figures
• data/ contains data objects created by scripts in data-raw/.
• analysis-output/ contains data objects created by scripts in analysis/.

References

Paredes, J. A., Hufkens, K., Marcadella, M., Bernhard, F., and Stocker, B. D.: rsofun v5.0: A model-data integration framework for simulating ecosystem processes, https://doi.org/10.1101/2023.11.24.568574, 8 February 2025.

Automated workflow

setwd("../rsofun_doc")
renv::restore()        # instantiate the correct environment

# process data
dir.create("data")
renv::run("data-raw/bigD13C_data.R",        name = "bigD13C_data")
renv::run("data-raw/vj_data.R",             name = "vj_data")
renv::run("data-raw/gpp_data.R",            name = "gpp_data")
# wait for the earlier scripts to finish, then run:
renv::run("data-raw/append_climate_data.R", name = "append_climate_data")

# setup data split into training and testing
renv::run("analysis/01_subset_test_train_sites.R", name = "subset_test_train_sites")

Run MCMC sampling of various calibration setups (setups are independent and can run in parallel):

# From RStudio:
#   renv::run("analysis/02_start_mcmc_bayesian_calibration_DREAMzs.R",    name = "bayesian_calibration 220", args = c(220,0,50,8))
# or alternatively (running outside of RStudio directly from tmux-shell can be more robust):
#   cd GitHub/geco-bern/rsofun_doc/
#   Rscript -e 'renv::run("analysis/02_start_mcmc_bayesian_calibration_DREAMzs.R", project = "../rsofun_doc", args = c(231,0,50,8))'
#   etc...

# Eventually, these codes were run on UBELIX with SLURM batch scripts:
sbatch ~/GitHub/geco-bern/rsofun_doc/analysis/run_mcmc.sh # specify the scenario throuh: "--array=220-223,226-231"

Sample posterior and predict train and test sets with various calibration setups (setups are independent and can run in parallel):

~/GitHub/geco-bern/rsofun_doc/analysis/run_predictions.sh 228 100000 30000 20 3 "_continued.rds"
~/GitHub/geco-bern/rsofun_doc/analysis/run_predictions.sh 227 100000 30000 20 3 "_continued.rds"
~/GitHub/geco-bern/rsofun_doc/analysis/run_predictions.sh 226 100000 30000 20 3 "_continued.rds"
~/GitHub/geco-bern/rsofun_doc/analysis/run_predictions.sh 222 100000 30000 20 3 "_continued.rds"
~/GitHub/geco-bern/rsofun_doc/analysis/run_predictions.sh 223 100000 30000 20 3 "_continued.rds"
~/GitHub/geco-bern/rsofun_doc/analysis/run_predictions.sh 231 100000 30000 20 3 "_continued.rds"

Generate figures for manuscript:

renv::run("analysis/05_make_figures.R", project = "../rsofun_doc")

Archive results from the scratch filesystem to a permanent one, e.g. with:

rsync -i --info=progress2   -avz --no-owner --omit-dir-times   /scratch/network/giub_geco/fbernhard/rsofun_doc_outputs /storage/capacity/occr_geco/data_2/archive_projects/PRJ_2025_fbernhard_rsofunDoc/