An application of History Matching in ORCHIDEE Land Surface Model
This repository uses the R front end ExeterUQ_MOGP for mogp_emulator (released in Python). You must install both repositories (be sure to install moqp_emulator V0.5).
We recommend that you install and use the conda environment provided in this repository.
$ conda install -c conda-forge --name HM --file requirements.txt
$ conda activate HM
Please provide the path of mopg_emulator and ExeterUQ_MOGP on lines 2 and 3 of the info.csv file in the data directory.
This repository provides an R script that executes a wave of 'History Matching'. This script can be used for any dataset as long as the format respects the same convention as the example provided in the Data directory. The data provided here are output metrics calculated from ORCHIDEE land surface model simulations using six different parameters.
To launch a Wave of History Matching please run the following line:
$ Rscript R/HistoryMatching.R /$Ablosulte_path$/Data/$Directory_name$/
The info.csv file provides options for the R script:
- l4 Wave number
- l5 Sample size to be generated with emulator
- l6 cut-off
- l7 tau
- l8 option to use a previous sample
- l9 option to create a new design for the next wave (to be used when NROY is too small)
- l10 option to keep good points in order to have a more real simulation point for the next wave
- l11 keeps good points only on the border (must activate l10 to use this option)
'Exploring the Potential of History Matching for Land Surface Model Calibration' Raoult et al. 2023 DOI : http://dx.doi.org/10.5194/egusphere-2023-2996 The code used for this article is HighTune We then recode the R script to adapt it for further studies.
Simon Beylat, Nina Raoult This script is an adaptation of the script used in HighTune available on svn.