explore_australia

Access to Australia-wide public data for OZ Minerals/Unearthed Explorer challenge - the jumpstarter repo!

Installation

This is a basic Python package but it makes use of a number of system libraries to read/write geospatial data. We recommed using Anaconda to manage these libraries otherwise you're likely to get burnt with strange C++ exceptions & dependency conflicts. Download Anaconda for your system here. Either the full distribution or Miniconda is fine.

We've provided a conda environment file to install all the required dependencies - environment.yml. You can create the environment and install the dependencies with:

$ cd /path/to/explore_australia

$ conda env create --file environment.yml
Collecting package metadata/|\-
# ...snip output

$ conda activate explore_australia

(explore_australia) $ # you should see the prompt change

Then you can install the package with:

(explore_australia) $ python setup.py install
running install
# ...snip output
Successfully installed explore_australia

This should install the python package and also the get_coverages CLI tool.

$ which get_coverages
# should show where this is installed, probably in .../conda/bin

You can run the test suite with python setup.py test from the package root directory.

Head over to the jupyter folder to find notebooks to get you going with this data in Python.

Targets: cleaned deposit locations

As targets, we've provided 3034 deposit locations gleaned from Geoscience Australia's Identified Mineral Resources database. They're available as data/deposit_locations.csv and data/deposit_locations.geo.json respectively. Geoscience Australia also has a nice poster showing these locations around Australia.

To make it easier to target commodity types and remove deposits that are unlikely to be useful targets we've done the following:

• Remove very rare deposits like diamond or 'uninteresting' deposits like opal or coal (who cares about silica or carbon?)
• Concatenated the commodity types into some larger groups for prediction purposes (e.g. illmenite -> Ti, hematite -> Fe). Also combined rare-earth (Sc, Y, La, Ce, Pr, Nd, Pm, Sm, Eu, Gd, Tb, Dy, Ho, Er, Yb, Lu, REO, REE, Xen, Mnz) and platinum-group element (PGE, Pt, Pd, Os, Rh, In, Ir, Re, Ru, Nb) resources into REE and PGE resources respectively.
• Streamlined the commodity type labels into a semicolon-delimited list (;)

and provided the latitude and longitude of these deposits in WGS84 longitude/latitude (epsg:4326).

Covariates: getting geophysical coverage data

Most of the geophysical data for all of Australia is pretty big so we've created a couple of Python functions to pull the data from their web coverage service endpoints - basically a little wrapper around owslib.

The relevant functions are all in explore_australia/stamps.py. The main ones are get_coverages and get_coverages_parallel. Use them like so:

>>> get_coverages(
        name=prominent_hill,  # a name for the data/stamp area
        lat=-32.42,           # box central latitude
        lon=122.169999,       # box central longitude
        angle=239,            # optional rotation about the box centre
        distance=25           # box side length in km
    )
# will loop through and grab tifs from WCS

After running this all your coverages will be under a folder called prominent_hill, sorted by type.

For the parallel get function, just make a pandas DataFrame with the following columns:

>>> import pandas

>>> stamps = pandas.read_csv('my_stamp_locations.csv')  # or whatever you like

>>> stamps.head()
         id    rotation  centre_longitude  centre_latitude
0       foo           0               147              -36
1       bar          23               125              -18

and you can pass this off to the get_coverages which will get each row's coverage in parallel:

>>> from explore_australia.stamp import get_coverages_parallel

>>> get_coverages_parallel(stamps)
Loading futures: 100%|██████████| 2/2 [00:00<00:00, 163.23it/s]
Collecting futures: 100%|██████████| 2/2 [01:26<00:00, 29.58s/it]

Depending on your machine/network connection you might want to tweak the number of workers.

All of the endpoints are stored in explorer_australia/endpoints.py (note you can also load these in any decent GIS package as well as see them in nationalmap.gov.au). We've provided endpoints for continent-wide magnetics (TMI and VRTP), gravity (isostatic residual and bouger anomaly), a number of ASTER products (which map surface mineralogy at a 30 m scale), and radiometric data (K, Th, U and total dose).

get_coverages CLI

You can use the CLI to pull out aligned coverages for any piece of Australia that you'd like (for example over deposit locations). This should be useful for generating exploratory, test and train datasets for building your models.

If you've got a particular area that you'd like to look at (e.g. over a known deposit), then you can pull out a box of (roughly) size distance using:

$ get_coverages --help
Usage: get_coverages [OPTIONS] NAME

  Get coverages for a given centre and angle

  Note that because we are getting a square on the surface of the earth, it
  might not be an exact square! Distances are approximate, depending on
  latitude.

Options:
  --lat FLOAT         Central latitude of the coverage, in degrees
  --lon FLOAT         Central longitude of the coverage, in degrees
  --distance INTEGER  The approximate length of the sides of the coverage (in
                      km)
  --angle FLOAT       An angle to rotate the box, in degrees
  --help              Show this message and exit.

$ get_coverages --lon=122.169999 --lat=-32.42 --angle=239 test_output
# will loop through and grab tifs from WCS

# Show all the downloaded geotiffs
$ ls test_output/**/*
test_output/geophysics/gravity:
bouger_gravity_anomaly.tif  isostatic_residual_gravity_anomaly.tif

test_output/geophysics/magnetics:
total_magnetic_intensity.tif  variable_reduction_to_pole.tif

test_output/geophysics/radiometrics:
filtered_potassium_pct.tif  filtered_terrestrial_dose.tif  filtered_thorium_ppm.tif  filtered_uranium_ppm.tif

test_output/remote_sensing/aster:
aloh_groun_content.tif      ferric_oxide_content.tif  ferrous_iron_index.tif  mgoh_group_content.tif  thermal_infrared_gypsum_index.tif  tir_quartz_index.tif
aloh_group_composition.tif  ferrous_iron_content.tif  kaolin_group_index.tif  opaque_index.tif        thermal_infrared_silica_index.tif

The CLI code is in explore_australia/cli.py if you want to see the regridding and alignment process (using rasterio).

We've also provided a Jupyter notebook showing you how to use rasterio to read the data once you've downloaded it.

Covariates: Geological mapping data

Geoscience Australia provides national coverages of surface geology. The data is available on data.gov.au, with a direct download here (about a Gb of shape files so we haven't just included here). You can see a rendered version here. There are seperate layers for geological unit polygons, linear features (e.g. faults, shear zones and dykes) and contacts. There are a lot of attributes to dig into, including age, lithology, history and some chemistry. The geological data comes with extensive documentation on the meanings of terms, and semantic identifiers (via GeoSciML and other standards) for ages, lithologies and other geological aspects.

Some of the state surveys (GSWA in particular) provide more detailed geology (down to 1:100k scale) with the covering rocks removed, seperate regolith (cover) maps, as well as extra information about tectonic history etc. You might be interested in using these datasets for the deposits in a single state to see whether they improve your predictions or change your certainty of existing predictions. If this sounds like you, head over to the WA state data portal: https://catalogue.data.wa.gov.au/dataset?q=geology.

We've provided a jupyter notebook which you can use to pull out the geology for a particular bounding box. GeoPandas and shapely make this very easy in Python.

Covariates: Other data?

There's nothing that's stopping you from using other data to train or validate your models if you think it will make for a better outcome or submission. Make sure you also take a look at the data portals of the other state and federal geological surveys for tons of useful data. For starters, try:

• Geological Survey of South Australia
• Geoscience Australia Data and Publications
• Geological Survey of Western Australia
• National map - you can use this to find other Web Coverage Services (WCS) to plug into the data getter.

Any issues?

Any problems, ask a question on the forum or in the Unearthed community slack. Feel free to submit pull requests if you find a bug in this repo.