Skip to content

installation.md

Latest commit

 

History

History
196 lines (160 loc) · 10.9 KB

installation.md

File metadata and controls

196 lines (160 loc) · 10.9 KB

Installation

1. Get the OSVAS code

mkdir OSVAS
cd OSVAS
git clone https://github.com/ACCORD-NWP/OSVAS.git .

2. Setup OSVAS installation (clone HARPSCRIPTS)

After cloning the OSVAS repository, run the installation setup script to automatically clone the HARP verification scripts:

cd OSVAS
./scripts/bash_scripts/setup_osvas_installation.sh

This script will:

  1. Clone the HARPSCRIPTS repository from git@github.com:harphub/oper-harp-verif.git into $OSVAS/HARPSCRIPTS/
  2. Apply OSVAS compatibility patches to the cloned HARPSCRIPTS files
  3. Display environment variable configuration for your shell (.bashrc or .bash_profile)
  4. Verify the setup by checking that all directories and expected files are present

Typical usage:

# Basic usage (auto-detects OSVAS from script location)
./scripts/bash_scripts/setup_osvas_installation.sh

# Or specify OSVAS directory explicitly
./scripts/bash_scripts/setup_osvas_installation.sh --osvas /path/to/OSVAS

After running the script, you can set the environment variables in your shell by copying the suggested export commands, or by adding them to your ~/.bashrc file.

Note: The launcher scripts automatically detect the OSVAS root directory from their script location, so you typically only need to set the environment variables if you plan to call the scripts from a different directory.

2. Install OSVAS dependencies within a conda Environment, and HARP packages within an isolated Renv setup.

On ATOS (ECMWF HPC), conda is available via a module:

module load conda/24.11.3-2

If conda is not installed on your system, install Miniconda:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

After installing Miniconda, you will be prompted to initialize conda for your shell. We recommend accepting this. To avoid auto-activating the base environment when opening a new terminal, you can disable it:

conda config --set auto_activate_base false

After conda is available, next step is to create the OSVAS conda environment OSVASENV, including the installation of HARP's R libraries inside an Renv. This is done by the bash script create_conda_and_R_envs.sh which, after installing the condas environment (called OSVHARP), runs the script ( renv_{atos,ubuntu}/renv_setup.R) for generating the Renv and installing all the harp packages. This installation can take about 30-60 minutes ( it actually compiles HARP packages; in the future this will be avoided by installing pre-compiled packages). Since there is a strict rate limit on "anonymous" installs of R libraries from CRAN mirrors, renv_setup.R must be edited to add your personal github PAT:

# (more info here https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/
# /managing-your-personal-access-tokens
# ---------------------------------------------------------------------------
 Sys.setenv(GITHUB_PAT = "Put_your_github_personal_access_token_here")

If you don't have a Personal Access Token, you can either create one in your github (settings-developer_settings-personal_access_tokens, borrow one from a colleague or repeat the next step a few times after it fails due to this rate limit (waiting for some time between tries to complete the compilation of R libraries).

Now simply run the script to complete the installation with conda & R environments:

cd scripts/bash_scripts
./create_conda_and_R_envs.sh

This will:

  1. Create an OSVASHARP conda environment with Python 3.11 and required packages
  2. Install Python dependencies from requirements.txt
  3. Set up an isolated R environment (renv) for HARP libraries (tested on ATOS and UBUNTU)

The conda environment includes:

  • Python data packages: pandas, numpy, scipy, matplotlib, xarray, netCDF4, cftime
  • ICOS API: icoscp (for accessing ICOS datasets)
  • Notebook tools: notebook, nbconvert, jupyterlab (for Jupyter notebook support)

Python launcher scripts and Jupyter notebooks

The OSVAS workflow is orchestrated by two Python launcher scripts in scripts/python_scripts/:

  • surfex_OSVAS_run_linux.py — Runs on local Linux systems
  • surfex_OSVAS_run_atos.py — Runs on ATOS HPC

These scripts execute Jupyter notebooks for data processing steps:

  • scripts/notebooks/WRITE_Station_forcing.ipynb — Generates forcing from ICOS data
  • scripts/notebooks/Flux_downloader.ipynb — Downloads and processes validation data & performs optional soil initialization

Notebook execution modes:

  • Via nbconvert (default): The launcher converts notebooks to Python scripts using jupyter nbconvert and executes them. This supports headless environments and doesn't require a Jupyter server.
  • Direct Python execution: Pre-converted .py versions are available (WRITE_Station_forcing.py, Flux_downloader.py) and can be executed standalone. This is activated by setting jupyter = True in the relevant line of the Python launcher scripts.
  • Interactive Jupyter: Notebooks can also be run manually with jupyter notebook or jupyter lab for development and debugging.

HARP and verification setup

HARP installation:

  • ATOS, UBUNTU: HARP and R dependencies are automatically installed by the setup script in renv_atos/ or renv_ubuntu/.
  • On other systems: The ubuntu Renv installation might work for your linux system. If not, follow the HARP installation guide: https://harphub.github.io/harp_training_2024/get-started.html#installation and make sure that R packages can be loaded from the R terminal after the installation.

HARP verification scripts (oper-harp-verif): The oper-harp-verif repository (https://github.com/harphub/oper-harp-verif.git) is automatically cloned during installation into $OSVAS/HARPSCRIPTS/ by the setup_osvas_installation.sh script, and "patched" to include a few modifications that will make possible to visualize verification results month by month. These scripts include:

  • Point verification scripts (point_verif.R)
  • Visualization apps (Shiny apps for dynamic and static displays)

Automatic path detection: The launcher scripts (surfex_OSVAS_run_linux.py and surfex_OSVAS_run_atos.py) automatically:

  1. Detect the OSVAS root directory from their script location (no need for manual configuration in most cases)
  2. Default HARPSCRIPTS to $OSVAS/HARPSCRIPTS

You can override these defaults by:

  • Setting environment variables: export OSVAS=/path/to/osvas and export HARPSCRIPTS=/path/to/harpscripts
  • Using command line arguments when running the launcher scripts: --osvas and --harpscripts

Extra required R packages for HARP (for manual installation if the Renv way is not used):

pkg_list <- c("here","argparse","yaml","dplyr","tidyr",
              "purrr","forcats","stringr","RColorBrewer","grid",
              "gridExtra","pracma","RSQLite","scales","pals",
              "shiny","shinyWidgets","lubridate","scico","cowplot")
for (pkg in pkg_list) {
  install.packages(pkg)
}

Activate the conda environment

Once setup is complete, activate the environment:

conda activate OSVHARP

How to make your SURFEX installation known to OSVAS:

A functional SURFEX installation is required. The workflow expects:

  • SURFEX binaries compiled and available in the system PATH or via environment modules
  • A profile script (e.g., profile_surfex-LXgfortran-SFX-V8-1-1-NOMPI-OMP-O2-X0) that sets up the compilation environment

Recommended SURFEX versions:

Currently, the PATH to your SURFEX setup, executables, profile names, etc, need to be hardcoded in step3 block of the Python launcher script, like this:

        # SURFEX paths
        surfex_parent = os.path.expanduser('~')
        surfex_ver = 'SURFEX_ACCORD'
        surfex_home = f"{surfex_parent}/{surfex_ver}"
        surfex_profile = 'profile_surfex-LXgfortran-SFX-V8-1-1-NOMPI-OMP-O2-X0'
        surfex_exe = f"{surfex_home}/src/dir_obj-LXgfortran-SFX-V8-1-1-NOMPI-OMP-O2-X0/MASTER/"
        os.environ['PATH'] = f"{surfex_exe}:{os.environ['PATH']}"
        
        paramfiles = f"{surfex_home}/MY_RUN/ECOCLIMAP/"
        dirfiles = f"{os.path.expanduser('~')}/PHYSIO/"

If you plan to use ECOCLIMAP files for any paremeter not resolved by namelist, make sure to have the corresponding version available in $dirfiles

How to set up access to forcing/validation data & Copernicus Global Land Service LAI data

ICOS login & token

To download ICOS data via the Python API, you must:

  1. Create an account at https://cpauth.icos-cp.eu/login/
  2. Log in and retrieve your access token from your user profile
  3. Store the token in $OSVAS/icos_cookie.txt Important: ICOS tokens expire after 27.8 hours (100,000 seconds) and must be renewed regularly.

KNMI login & API_key

  1. Register at https://developer.dataplatform.knmi.nl/register/
  2. Login at https://developer.dataplatform.knmi.nl/login
  3. Get API key at https://developer.dataplatform.knmi.nl/member/ and store at $OSVAS/knmi_apikey.txt

Important: To access the soil moisture datasets for Cabauw https://dataplatform.knmi.nl/dataset/cesar-soil-water-lb1-t10-v1-1, a special request must be addressed to opendata@knmi.nl

Copernicus Global Land Service LAI data

  1. Create an account in https://dataspace.copernicus.eu/
  2. When running Step 2c for generating monthly LAI estimates, you will be prompted to open a link in the browser and log-in; after doing so, the script in the terminal will continue with the data retrieval.

Environment variables and paths

The launcher scripts automatically detect paths based on their script location. However, you can optionally set these environment variables to override the defaults:

export STATION_NAME=Majadas_del_tietar       # Station to process (required)
export OSVAS=$HOME/OSVASgh                    # OSVAS root (auto-detected if not set)
export CONDAENV=OSVASENV                      # Conda environment name (default: OSVHARP)
export HARPSCRIPTS=$OSVAS/HARPSCRIPTS         # HARP scripts path (auto-detected if not set)

Typical usage: In most cases, you only need to set STATION_NAME. The launcher scripts will automatically:

  • Detect OSVAS from their location in scripts/python_scripts/
  • Set HARPSCRIPTS to $OSVAS/HARPSCRIPTS
  • Use OSVHARP as the default conda environment

Command line override: You can also pass these as command line arguments:

python surfex_OSVAS_run_linux.py --stations Cabauw Loobos --osvas /path/to/osvas --harpscripts /path/to/harpscripts