topopyscale 0.1.5

A Python package to perform climate downscaling at the hillslope scale

TopoPyScale

Pure Python version of Toposcale packaged as a Pypi library. Toposcale is an original idea of Joel Fiddes to perform topography-based downscaling of climate data to the hillslope scale.

Documentation avalaible (in construction): https://topopyscale.readthedocs.io

References:

• Fiddes, J. and Gruber, S.: TopoSCALE v.1.0: downscaling gridded climate data in complex terrain, Geosci. Model Dev., 7, 387–405, https://doi.org/10.5194/gmd-7-387-2014, 2014.
• Fiddes, J. and Gruber, S.: TopoSUB: a tool for efficient large area numerical modelling in complex topography at sub-grid scales, Geosci. Model Dev., 5, 1245–1257, https://doi.org/10.5194/gmd-5-1245-2012, 2012.

Kristoffer Aalstad has a Matlab implementation: https://github.com/krisaalstad/TopoLAB

Contribution Workflow

Please follow these simple rules:

1. a bug -> fix it!
2. an idea or a bug you cannot fix? -> create a new issue if none doesn't already exist. If one exist, then add material to tit.
3. wanna develop a new feature/idea? -> create a new branch. Do the development. Merge with main branch when accomplished.
4. Create release version when significant improvements and bug fixes have been done. Coordinate with others

Create a new release: Follow procedure and conventions described in: https://www.youtube.com/watch?v=Ob9llA_QhQY

And check out our Slack: tscaleworkspace.slack.com

Contributors to the current version (2021) are:

• Simon Filhol
• Joel Fiddes
• Kristoffer Aalstad

TODO

• class topoclass:
  • implement routine to fecth DEM automatically from ArcticDEM/ASTER
  • add plotting routines.
• topo_sub.py:
  • look into alternative clustering method DBscan
  • write download routine for CORDEX climate projection
• topo_scale.py:
  • try to rewrite function to skip for loop using the power of zarray
  • check against previous implementation of TopoScale
• Documentation:
  • describe config.ini variables
  • short description of topoclass structure
  • example of downscaling without topoclass
  • write about about available forcing datasets

Design

1. Inputs
   • Climate data from reanalysis (ERA5, etc)
   • Climate data from future projections (CORDEX) (not avail.)
   • DEM from local source, or fetch from public repository: SRTM, ArcticDEM, ASTER
2. Run TopoScale
   • compute derived values (from DEM)
   • toposcale (k-mean clustering)
   • interpolation (bilinear, inverse square dist.)
3. Output
   • Cryogrid format
   • FSM format
   • CROCUS format
   • Snowmodel format
   • basic netcfd
   • For each method, have the choice to output either the abstract cluster points, or the gridded product after interpolation
4. Validation toolset
   • validation to local observation timeseries
   • plotting
5. Gap filling algorithm
   • random forest temporal gap filling

Validation (4) and Gap filling (4) are future implementation.

Installation

conda install mamba -n base -c conda-forge
mamba create -n downscaling ipython numpy pandas xarray matplotlib netcdf4 ipykernel scikit-learn rasterio gdal pyproj
conda activate downscaling
pip install cdsapi
pip install h5netcdf
pip install topocalc
pip install pvlib
pip install elevation
pip install configobj
pip install lazydocs

cd github  # navigate to where you want to clone TopoPyScale
git clone [e-mail]:ArcticSnow/TopoPyScale.git
pip install -e TopoPyScale    #install a development version

#----------------------------------------------------------
#            OPTIONAL: if using jupyter lab
# add this new Python kernel to your jupyter lab PATH
python -m ipykernel install --user --name downscaling

Then you need to setup your cdsapi with the Copernicus API key system. Follow this tutorial after creating an account with Copernicus. On Linux, create a file nano ~/.cdsapirc with inside:

url: https://cds.climate.copernicus.eu/api/v2
key: {uid}:{api-key}

Basic usage

1. Setup your Python environment
2. Create your project directory
3. Configure the file config.ini to fit your problem (see config.ini for an example)
4. Run TopoPyScale

import pandas as pd
from TopoPyScale import topoclass as tc
from matplotlib import pyplot as plt

# ========= STEP 1 ==========
# Load Configuration
config_file = './config.ini'
mp = tc.Topoclass(config_file)
# Compute parameters of the DEM (slope, aspect, sky view factor)
mp.compute_dem_param()

# ========== STEP 2 ===========
# Extract DEM parameters for points of interest (centroids or physical points)

mp.extract_topo_param()

# ----- Option 1:
# Compute clustering of the input DEM and extract cluster centroids
#mp.extract_dem_cluster_param()
# plot clusters
#mp.toposub.plot_clusters_map()
# plot sky view factor
#mp.toposub.plot_clusters_map(var='svf', cmap=plt.cm.viridis)

# ------ Option 2:
# inidicate in the config file the .csv file containing a list of point coordinates (!!! must same coordinate system as DEM !!!)
#mp.extract_pts_param(method='linear',index_col=0)

# ========= STEP 3 ==========
# compute solar geometry and horizon angles
mp.compute_solar_geometry()
mp.compute_horizon()

# ========= STEP 4 ==========
# Perform the downscaling
mp.downscale_climate()

# ========= STEP 5 ==========
# explore the downscaled dataset. For instance the temperature difference between each point and the first one
(mp.downscaled_pts.t-mp.downscaled_pts.t.isel(point_id=0)).plot()
plt.show()

# ========= STEP 6 ==========
# Export output to desired format
mp.to_netcdf()

TopoClass will create a file structure in the project folder (see below). TopoPyScale assumes you have a DEM in GeoTiFF, and a set of climate data in netcdf (following ERA5 variable conventions). TopoPyScale can easier segment the DEM using clustering (e.g. K-mean), or a list of predefined point coordinates in pts_list.csv can be provided. Make sure all parameters in config.ini are correct.

my_project/
    ├── inputs/
        ├── dem/ 
            ├── my_dem.tif
            └── pts_list.csv  (optional)
        └── climate/
            ├── PLEV*.nc
            └── SURF*.nc
    ├── outputs/
    └── config.ini

Principle of TopoPyScale