Configuration

Landschaft has no configuration file: a handful of settings, passed to the functions that use them or set once for the Python session.

Threads

Every function that runs in parallel uses all processor cores but one by default, so that the computer stays responsive while it works. You can change this per call or for the whole session:

import landschaft as ls

ls.max_threads()          # processors available, e.g. 20
ls.default_threads()      # threads used when you do not say: 19

forest = ls.load_raster(ls.example_data("input"))
ls.mspa(forest, n_threads=4)       # this call only: 4 threads
ls.set_default_threads(8)          # from now on, every call: 8 threads
ls.set_default_threads(None)       # back to all cores but one

n_threads=None or n_threads=0 always means the default. The results never depend on the number of threads: the same input gives the same bits with 1 or 64 threads.

Functions that process many rasters at once (ls.batch, ls.mspa on a collection, ls.mspa_tiled) also take n_workers, the number of rasters or tiles in flight (ls.batch also threads_per_worker); by default the workers share the same cores (all but one).

Jobs that run side by side

If you start several Python processes at once (e.g. on a cluster node), give each a share of the cores with ls.set_default_threads(n) at the top of each script, so that together they do not use more cores than the machine has.

Other computers

Functions that run many independent tasks (batch, mspa_batch, simulate.scenarios, simulate.calibrate) take on=ls.helpers() and share the tasks with the computers of a group running landschaft worker, with the same results; see several computers. The group's secret is kept in %APPDATA%\landschaft (Windows) or ~/.config/landschaft (LANDSCHAFT_CONFIG_DIR overrides it).

Memory

Setting	Where	What it does
max_memory="16GB"	ls.batch, ls.mspa on a collection	rasters are started only while their estimated memory fits the budget
block_rows=...	ls.Landscape, ls.landscape_metrics	landscape metrics of a file read in blocks of rows, for rasters larger than memory
tile_size=...	ls.mspa_tiled	MSPA of rasters larger than memory, tile by tile
memory_limit=...	ls.omniscape	caps the memory of the moving windows (fewer threads for very large windows)

Output files

ls.write_raster(data, "out.tif") writes a compressed, tiled GeoTIFF with the georeferencing, NoData and band names of data and, for class maps, their colour table (colormap="auto"); pass colormap=None to leave it out or a {value: (r, g, b, a)} dictionary for your own colours. It chooses what the data need: NaN as the NoData of floats (or nodata=-9999), uint8 for True/False maps, the smallest integer type for labels, the format from the extension (.tif, .asc, .img...; driver="COG" for a Cloud-Optimised GeoTIFF). Dask arrays are written block by block, and missing folders are created. Tables are pandas DataFrames; large metric tables can be streamed to Parquet or CSV (see the landscape metrics tutorial).

Environment variables

Variable	Effect
LANDSCHAFT_CIRCUIT_AVX2=0	use the generic build of the circuit solver instead of the AVX2 one (for comparisons; the default picks AVX2 when the processor has it)
GDAL_*	GDAL's own settings (cache size, number of threads for decoding...), honoured by all file reading

Reproducibility

Landschaft has no random numbers: the same input and options give the same output on the same platform, whatever the number of threads. Record the version (ls.__version__) with your results.