HRDS 0.1.4

Read in multiple raster data files in a hierarchical fashion

About hrds

hrds is a python package for obtaining points from a set of rasters at different resolutions. You can request a point and hrds will return a value based on the highest resolution dataset (as defined by the user) available at that point, blending datasets in a buffer region to ensure consistency.

Current release:

Paper:

Prerequisites

• python 3+
• numpy
• scipy
• osgeo.gdal (pygdal) to read and write raster data

hrds is available on conda-forge, so you can install easily using:

conda config --add channels conda-forge
conda install hrds

It is possible to list all of the versions of hrds available on your platform with:

conda search hrds --channel conda-forge

On Debian-based Linux you can also install manually. To install pygdal, first install the libgdal-dev packages and binaries:

sudo apt-get install libgdal-dev gdal-bin

To install pygdal, first check which version of gdal is installed:

gdal-config --version

pygdal can be installed using pip, specifying the version obtained from the command above. Note that you may need to increase the minor version number, e.g. from 2.1.3 to 2.1.3.3.

pip install pygdal==2.1.3.3

Replace 2.1.3.3 with the output from the gdal-config command.

You can then install hrds from source using the standard:

python setup.py install

or from PyPi:

pip install hrds

Functionality

• Create buffer zones as a preprocessing step if needed

• Obtain value at a point based on user-defined priority of rasters

The software assumes all rasters are already in the same projection space and using the same datum.

This example loads in an XYZ file and obtains data at each point, replacing the Z value with that from hrds.

from hrds import HRDS

points = []
with open("test_mesh.csv",'r') as f:
    for line in f:
        row = line.split(",")
        # grab X and Y
        points.append([float(row[0]), float(row[1])])

bathy = HRDS("gebco_uk.tif",
             rasters=("emod_utm.tif",
                      "inspire_data.tif"),
             distances=(700, 200))
bathy.set_bands()

print(len(points))

with open("output.xyz","w") as f:
    for p in points:
        f.write(str(p[0])+"\t"+str(p[1])+"\t"+str(bathy.get_val(p))+"\n")

This will turn this:

$ head test_mesh.csv
805390.592314,5864132.9269,0
805658.162910036,5862180.30440542,0
805925.733505999,5860227.68191137,0
806193.304101986,5858275.05941714,0
806460.874698054,5856322.43692232,0
806728.445294035,5854369.81442814,0
806996.015889997,5852417.19193409,0
807263.586486046,5850464.56943942,0
807531.157082069,5848511.94694493,0
807798.727678031,5846559.32445088,0

into this:

$ head output.xyz
805390.592314	5864132.9269	-10.821567728305235
805658.16291	5862180.30441	2.721575532084955
805925.733506	5860227.68191	2.528217188012767
806193.304102	5858275.05942	3.1063558741547865
806460.874698	5856322.43692	5.470234157891056
806728.445294	5854369.81443	1.382685066254607
806996.01589	5852417.19193	1.8997482922322515
807263.586486	5850464.56944	4.0836843606647335
807531.157082	5848511.94694	-2.39508079759155
807798.727678	5846559.32445	-2.401006071401176

An example of use via thetis:

import firedrake
import thetis
from hrds import HRDS

mesh2d = firedrake.Mesh('test_mesh.msh') # mesh file

P1_2d = firedrake.FunctionSpace(mesh2d, 'CG', 1)
bathymetry2d = firedrake.Function(P1_2d, name="bathymetry")
bvector = bathymetry2d.dat.data
bathy = HRDS("gebco_uk.tif", 
             rasters=("emod_utm.tif", 
                      "inspire_data.tif"), 
             distances=(700, 200))
bathy.set_bands()
for i, (xy) in enumerate(mesh2d.coordinates.dat.data):
    bvector[i] = bathy.get_val(xy)
thetis.File('bathy.pvd').write(bathymetry2d)

# rest of thetis code

These images show the original data in QGIS in the top right, with each data set using a different colour scheme (GEBCO - green-blue; EMOD - grey; UK Gov - plasma - highlighted by the black rectangle).The red line is the boundary of the mesh used (see figure below). Both the EMOD and UK Gov data has NODATA areas, which are shown as transparent here, hence the curved left edge of the EMOD data. The figure also shows the buffer regions created around the two higher resolution datasets (top left), with black showing that data isn't used to white where it is 100% used. The effect of NODATA is clear here. The bottom panel shows a close-up of the UK Gov data with the buffer overlayed as a transparancy from white (not used) to black (100% UK Gov). The coloured polygon is the area of the high resolution mesh (see below).

After running the code above, we produce this blended dataset. Note the coarse mesh used here - it's not realistic for a model simulation!

If we then zoom-in to the high resolution area we can see the high resolution UK Gov data being used and with no obvious lines between datasets.

Community

We welcome suggestions for future improvements, bug reports and other issues via the issue tracker. Anyone wishing to contribute code should contact Jon Hill ([e-mail]) to discuss.