Tools
The analyses developed in TROLL version 4 evaluation rely mainly on the development of a new version of rcontroll integrating this new version of TROLL, the encapsulation of this new version with other R pacakges in a singularity image, and the snakemake worklfow to automatically run the virtual experiment in a cluster with slurm. The rest of the analyses are done directly in the quarto book.
rcontroll version 4
A new version of rcontroll (Schmitt et al. 2023) has been developed for TROLL version 4 evaluation available on the corresponding branch on GitHub: https://github.com/sylvainschmitt/rcontroll/tree/TROLLV4 . This version mainly includes the inclusion of the new C++ code, the adaptation of inputs parameters and outputs in rcontroll objects and plots. The download of ERA5-Land data has also been updated. Every improvement since the version published on CRAN are described here: https://github.com/sylvainschmitt/rcontroll/blob/TROLLV4/NEWS.md (starting from rcontroll 0.2.0.9001).
singularity encapsulation
To ease portability, reproducibility and integration into a workflow we encapsulated rcontroll with TROLL version 4 and 3 and other useful R packages for the analyses in two singularity images. The singularity recipes are available in singularity/Singularity with the associated R code. The code is inspired from https://github.com/sylvainschmitt/singularity-troll. Once singularity installed (linux only but WSL can help windows users), you can install the image with the following command:
sudo singularity build troll.sif SingularityThen the snakemake workflow will automatically use the corresponding image. In case you want to test it manually use the following command:
singularity shell -e troll.sif snakemake workflow
To ease portability, reproducibility, and sustainability of the analyses, all virtual experiments have been included in a single snakemake workflow including thus calibration, variation, and evaluation defined in Snakefile. Once snakemake installed you can easily run the workflow locally. However the number of simulations is very high for calibration and spin-up simulations are very long so we recommend that you uses an high performance computing (HPC) platform. Most will include snakemake and singularity, and a submission script is already ready for cluster that uses slurm. In case you want to adjust virtual experiments definitions you can adjust the config/config_*.yml file. To test the worklfow locally you can use a dry run:
snakemake -np -j 1On the HPC you first need to activate the snakemake module, for instance:
module load bioinfo/Snakemake/7.20.0Than used (and adapt) the predefined submission script:
sbatch job.shThere are several configuration files for each step of the workflow (calib_str3, cali_str4, calib_pheno, eval) and site (Paracou, Tapajos).