Fast, landscape analysis for Python.
Habitat patterns, fragmentation, landscape metrics and ecological connectivity,
from a single raster to maps larger than memory.
Landschaft brings the established methods of landscape ecology together in one Python package, with a C++17/OpenMP core behind a numpy / xarray / dask / rasterio API. It reproduces the reference tools of each method, checked against their outputs in the test suite, and runs from a few to several hundred times faster.
| Area | What it covers | Reproduces |
|---|---|---|
| Habitat patterns | MSPA, SPA, object accounting, contortion | GuidosToolbox |
| Fragmentation | density at many scales (FAD, FOS), landscape mosaics, contagion, entropy | GuidosToolbox |
| Landscape metrics | 133 patch, class and landscape metrics, sample plots, moving windows, time series | landscapemetrics, FRAGSTATS |
| Habitat networks | PC, IIC, ECA, node and link importance, ProtConn | Conefor, Makurhini |
| Circuit connectivity | omnidirectional and pairwise current flow, rasters and networks | Circuitscape, Omniscape |
| Distance & restoration | cost surfaces, least-cost paths and corridors, isochrone maps | GuidosToolbox |
| Change detection | simple and morphological change, fragmentation transitions | GuidosToolbox |
| Routes on grids | transitions from any rule, travel-time surfaces, least-cost distances and paths, random walks, randomised shortest paths, route overlap | gdistance |
| Surface metrics | texture of continuous rasters (ISO 25178): roughness, height distribution, bearing area curve, slopes, peaks and valleys, autocorrelation, orientation, fractal dimension, trend surfaces, texture images; terrain indices (TPI, TRI, VRM, curvature, HLI...) | geodiv, spatialEco |
| Terrain hydrology | depression filling, D8, D-infinity and multiple flow directions, flow accumulation, streams, height above the nearest drainage (HAND), wetness, stream power and sediment transport indices | WhiteboxTools |
| Neutral landscape models | simulated landscapes as null models: random, percolation, gradients, fractal surfaces, Gaussian fields, mosaics, clusters, classification | NLMR |
| Biodiversity analysis | species records in grid cells: richness, endemism, phylogenetic diversity and endemism, randomisations, CANAPE, clustering | Biodiverse |
| Spatial prioritization | ranking of whole landscapes for many features (CAZ and ABF rules), weights, costs, hierarchic masks, connectivity transforms, performance curves, post-processing, impact assessment; maps larger than memory | Zonation 5 |
| Species distributions | cleaning and thinning occurrence records, accessible area, collinearity filters, background points, spatial cross-validation folds, evaluation (AUC, TSS, Boyce index), Maxent (MaxNet), ensembles of scikit-learn models, projection and extrapolation (MESS, MOP, ExDet) | CoordinateCleaner, spThin, usdm, dismo, blockCV, ENMeval, ecospat, predicts, maxnet, mop |
| Land change simulation | change analysis (intensity, components of difference, Markov chains), transition potential (weights of evidence, logistic regression, any scikit-learn model), allocation (CLUE-S, ordered, land systems, Expander/Patcher, random patch seeds, patch growing), dynamic variables, regional demands, validation at several resolutions, fuzzy similarity, calibration, scenarios with replicates | lulcc, OpenLand, diffeR, CLUMondo, FUTURES, MOLUSCE, Dinamica EGO, PLUS |
| Ecosystem services | carbon storage and sequestration with valuation, habitat degradation, quality and rarity | InVEST |
| Large data | out-of-core tiles and row blocks, parallel batches of rasters, tasks shared with the other computers of a lab (landschaft worker) |
pip install landschaftBinary wheels are published for CPython 3.10 to 3.14 on Windows (x86-64), macOS (Apple silicon and Intel, 11+) and Linux (x86-64, ARM64), with the OpenMP runtime included: no compiler is needed, and everything every tool uses is installed with the package. In a conda environment, install the geospatial libraries from conda-forge first and Landschaft with pip:
conda create -n landschaft -c conda-forge python=3.12 rasterio rioxarray geopandas
conda activate landschaft
pip install landschaftBuilding from source (other platforms, development) needs a C++17 compiler; see the installation guide.
The image needs no compiler or Python on the host and includes every optional feature:
docker build -t landschaft .
docker run --rm -it -v "$PWD:/data" landschaft # Python shell
docker run --rm -v "$PWD:/data" landschaft python analysis.py # run a script from the current folderThe current folder is mounted at /data, the working directory of the container.
--build-arg PYTHON_VERSION=3.11 picks another Python, and
docker build --target test -t landschaft:test . && docker run --rm landschaft:test
runs the test suite in the container.
import landschaft as ls
forest = ls.load_raster(ls.example_data("input")) # bundled sample map
morph = ls.mspa(forest, edge_width=2) # cores, edges, bridges, loops...
ls.write_raster(morph, "forest_mspa.tif") # GeoTIFF with its colour table
print(ls.mspa_stats(morph))
landcover = ls.load_raster(ls.example_data("clc"))
ls.landscape_metrics(landcover, level="landscape", metric=["shdi", "contag", "ed"])Every function takes numpy arrays, georeferenced xarray.DataArrays or dask arrays,
and returns the same type; parallel functions use all cores but one unless given
n_threads.
The documentation website, https://sacridini.github.io/landschaft/, has a
getting-started guide, tutorials by topic, the full API reference with runnable
examples, and the benchmarks and accuracy report. It is built from docs/ with
MkDocs and published with each release; to build it locally:
pip install -e ".[docs]"
mkdocs serve # http://127.0.0.1:8000, or `mkdocs build` for a static site/Each tool is compared in the test suite with the program it reproduces: pixel for pixel where the output is a map, value for value (to the last bit where the original arithmetic can be reproduced) for tables, and to the original's own precision for floating-point solvers. Typical speed-ups on the benchmark cases:
Methods, machines, full tables and the known differences are on the benchmarks and accuracy page. To run the tests:
pip install -e ".[test]"
pytestLandschaft follows Semantic Versioning: from 1.0 on, the public API changes incompatibly only in a new major version. Changes are listed in the changelog.
GPL-3.0-or-later. Landschaft ports code from GuidosToolbox and miallib,
© European Union (Joint Research Centre), authors Pierre Soille, Peter Vogt et al.,
both GPL-3.0-or-later; the example maps in src/landschaft/data come from the
GTB repository. GTB's mspa notes that its
interface is made available for non-commercial purposes: contact the JRC for
commercial use of MSPA.
Please cite the methods you use. For MSPA:
- Soille, P. & Vogt, P. (2009). Morphological segmentation of binary patterns. Pattern Recognition Letters 30(4), 456–459. doi:10.1016/j.patrec.2008.10.015
- Vogt, P. & Riitters, K. (2017). GuidosToolbox: universal digital image object analysis. European Journal of Remote Sensing 50(1), 352–361. doi:10.1080/22797254.2017.1330650