site-suitability

Program to automate site suitability / site selection analysis

Features

• Transparent: open source means the code can be inspected and audited, especially the use of inputs and function parameters
• Reproducible: the configuration file that describes the calculations are independent of the code that does the actual work, so it can be released without releasing the backend code.
• Modular: if one step fails and aborts the program, once you fix it you can simply re-run the program. It will skip calculating existing files and continue as if it was never aborted
• Not a black box: every pre-processing, standardization, and weighting step is given back.

Type safety

Helps distinguishes between vector and raster files at compile time. While it does not do any runtime verification, it is easier to see if you made a mistake. For example, doRasterOperation (Raster "roads.shp") is obviously wrong and doVectorOperation (Vector "solar_radiation.tif") is obviously wrong too.

Now, doRasterOperation "roads.shp" can also be spotted relatively easily, but what about:

layer = "roads.shp"
doRasterOperation layer
-- many lines later
doVectorOperation layer

Sure, you can name things like:

rasterLayer = "roads.shp"
doRasterOperation rasterLayer
-- many lines later
doVectorOperation rasterLayer

But why not have the compiler catch it for you at compile time, before your program even runs?

layer = Raster "roads.shp"
-- doRasterOperation :: Raster -> IO ()
doRasterOperation layer
-- many lines later
-- doVectorOperation :: Vector -> IO ()
doVectorOperation layer  -- compile-time error, layer :: Raster but expected Vector

Installation

Prerequisites

• Install Haskell Stack
• Install an updated version of GDAL

Your GDAL version must be at least (aka, cannot be older than this):

$ gdalinfo --version
GDAL 3.3.1, released 2021/06/28

For example commands, see example_ci.yml

Compile

1. git clone and cd into dir
2. stack build

Usage

Note: these are instructions to replicate the Arizona case study. To use other data sets, just modify the configuration

Prerequisites

Download the datasets and do some slight pre-processing. See README_datasets.md and data/README.md.

TLDR: run this

cd data
python download_simple.py
python download_roads.py
./crop_border.sh
./land_use.sh

Running

Edit configs/run.txt. Put the names of the configs you want to run on a new line. Do not type extensions (.json)

If your shell's python is from (ana)conda but QGIS is not installed with (ana)conda, please disable it temporarily Use conda deactivate (To enable it back, type conda activate base, where base is your environment name)

Therefore the command to compile and run would be like:

conda deactivate && stack build && stack exec site-suitability

Customizing

See ./configs/README.md for documentation on how to edit or write a configuration file

Integration tests

Install pytest and imagehash (pip install imagehash). Run the program at least once to generate outputs, then run:

cd tests
pytest -vvvv --log-cli-level=INFO test.py

This will calculate and store similarity-hashes of all outputs and save them as the baseline. Future runs will test existing outputs against the stored baseline

To customize which configs are saved, see tests/README.md

Q&A

• Why not use Python (rasterio) or R?

Because I started work on this before those were taught to me. They're already used to generate figures.

• Why not Rust?

Because this is the signature of the first function I ported:

pub fn batch<A, B>(
    f: fn((<A as Iterator>::Item, <A as Iterator>::Item)) -> B,
    input_files: A,
    output_files: A,
) // -> B // doesn't even work!
where
    A: Iterator

Compared to Haskell:

-- Haskell language server automatically generates the signature
batch = zipWithM

• But why Haskell?

This program is just a fancy wrapper around a series of command line instructions anyway, any programming language can do this, so I picked my favourite programming language.

License

(MIT) or (GPLv3 or later), at your choice