Glossary | Terraflux

The tutorials define each term where it first matters. This page gathers them, so you can come back when a word is unfamiliar. Terms are grouped by subject; use the search (Ctrl+K) to jump to one.

Maps and data#

Raster: A map made of a grid of square cells (pixels), each holding one value: a land cover class, an elevation, a probability. Satellite images, land use maps and elevation models are rasters. Terraflux works mostly with rasters.
Cell, cell size, resolution: A cell is one square of a raster. Its size (or the raster's resolution) is the length of its side on the ground, for example 30 m. Smaller cells show more detail and make larger files. Areas computed by Terraflux come from the cell size: a 30 m cell covers 900 m², 0.09 ha.
Extent: The rectangle a map covers, given by its edges' coordinates (west, south, east, north).
Band, layer, stack: A raster file may hold several grids of the same size, called bands: the years of a land use series, the drivers of a model, the channels of an image. In Terraflux a set of maps on one grid is a stack, and each map in it a layer.
No data (nodata, missing): A cell without a value: outside the study area, under a cloud, the sea. A raster marks it with a special value; Terraflux leaves those cells out of every count and calculation.
Categorical (classes) and continuous data: A categorical raster holds codes of classes (1 forest, 2 farming, 3 water): the numbers are labels, and averaging them makes no sense. A continuous raster holds measurements (elevation, temperature, a probability). The difference decides how a map may be resampled and which colours show it.
Data type: How a raster stores its numbers: whole numbers of 8, 16 or 32 bits (uint8, int16...) or decimal numbers (float32, float64). Classes fit in uint8; probabilities need floats.
GeoTIFF, COG: The usual raster file format (.tif), with its coordinates inside. A Cloud Optimized GeoTIFF (COG) is a GeoTIFF arranged in tiles with smaller copies (overviews), so that parts of it can be read quickly, even over the web.
Vector data: Maps made of features with coordinates: points (occurrence records, plots), lines (roads, rivers) and polygons (reserves, parcels, municipalities). Each feature carries attributes in a table. Files: GeoPackage (.gpkg), Shapefile (.shp), GeoJSON.
Rasterize: To burn vector features into a raster: each cell takes the value of the feature over it.
Zonal statistics: Statistics of a raster within each polygon (or zone): the mean elevation of each reserve, the forest area of each municipality.
GDAL: The open-source library that reads and writes almost every geographic file format. Terraflux brings its own copy, so you need not install it.

Coordinates and grids#

Coordinate reference system (CRS): The rule that turns positions on the Earth into the numbers of a map. Each has a code, such as EPSG:4326 (latitude and longitude, WGS 84) or EPSG:31983 (UTM zone 23 S, SIRGAS 2000).
Geographic and projected coordinates: Geographic coordinates are angles (degrees of latitude and longitude); a degree is not a fixed distance on the ground. Projected coordinates are metres on a flat map, made by a projection. Areas, distances and most landscape metrics need projected coordinates, ideally an equal-area projection for areas.
Reprojection: Moving a map from one CRS to another. For a raster it means building a new grid and filling it by resampling.
Resampling: Filling the cells of a new grid from an old one. Nearest neighbour and mode (majority) keep the original values, so they suit classes; bilinear and cubic interpolation, and mean, suit continuous values; sum suits counts. Using interpolation on classes invents classes that do not exist.
Grid alignment, reference grid: Maps can be combined cell by cell only when they share the same CRS, cell size and cell edges. Terraflux's reference grid is the grid every map of a project is brought onto (raster.match).

Landscape ecology#

Landscape: An area that is spatially heterogeneous in at least one factor of interest, at the scale of the question. Its pattern is described by composition (what is there, and how much) and configuration (how it is arranged).
Patch: An area of one class, made of cells that touch each other. Whether cells touching only at a corner belong to the same patch depends on the neighbourhood rule: 4 neighbours (sides only) or 8 (sides and corners).
Grain and extent (scale): The grain is the finest unit of observation (the cell size); the extent is the whole area studied. Most landscape measures change with both, so report them.
Edge, edge effect: The boundary between a habitat and its surroundings, where conditions differ from the interior (light, wind, predators). The edge width is how far into a patch that influence reaches.
Core area: The part of a habitat patch farther from its edge than the edge width.
Habitat loss and fragmentation: Habitat loss is the reduction of habitat amount; fragmentation per se is the breaking apart of the habitat that remains, for the same amount. The two are often confused, and their effects on species differ.
Connectivity: The degree to which a landscape lets organisms move between habitat patches. Structural connectivity depends on the layout alone; functional connectivity depends on how a given species moves.
Resistance (cost) surface: A raster saying how hard (costly, risky) it is for an organism to cross each cell. It is the input of least-cost and circuit analyses.
Corridor: A strip of habitat, or of easy terrain, linking two patches.
Neutral landscape model: A landscape drawn by chance with a known pattern, used as a null model: what would the pattern be if only chance shaped it?

Conservation planning#

Feature: Something a plan should conserve: a species, an ecosystem, a stock of carbon. Each is mapped as a raster of its amount or probability in each cell.
Planning unit: The unit a plan selects or ranks: a cell, a hexagon, a parcel, a watershed.
Target: How much of each feature a plan must hold (30 % of each species' range, 1,000 ha of each ecosystem).
Complementarity: Choosing places for what they add to the places already chosen, not for their own richness.
Irreplaceability: How necessary a planning unit is to meet the targets: an irreplaceable unit cannot be swapped for another.

Modelling#

Model, graph, node, operator: A Terraflux model is a graph of nodes. Each node runs one operator (read a map, reclassify it, run MSPA...) on the values that reach it through its input ports and gives values from its output ports. Connections carry values from an output to an input.
Parameter (setting): A value that tunes how an operator works (an edge width, a window size), set in the node rather than connected.
Driver (explanatory variable): In land change models, a map that helps explain where change happens: distance to roads, slope, soil.
Calibration and validation: Calibration fits a model's settings on observed data from a past period; validation compares what the model predicts with observations it did not see.
Stochastic model, seed, replicate: A stochastic model uses random numbers, so two runs differ. The seed fixes the random numbers: the same seed gives the same result. Replicates are runs with different seeds, to see the spread of outcomes.
Cellular automaton: A model where each cell changes state by rules that depend on its own state and on its neighbours', step after step.
Monte Carlo: Running a calculation many times with inputs drawn from their distributions, to see the distribution of the result: its uncertainty.

Computing#

Core, thread: A core is one processing unit of a computer's processor; a laptop has 4 to 16. A thread is a sequence of work a core runs. Work split into threads runs on several cores at once.
Process, machine, node: A process is a running program. A machine is one computer; in a cluster each machine is called a node (not to be confused with a model's nodes).
Coordinator and worker: In Terraflux's distributed mode, the coordinator hands out tasks (replicates, scenarios, tiles) and the workers run them, on the same machine or on many.
Cluster, HPC, Slurm: A cluster is a group of machines run as one, often a university's high-performance computing (HPC) facility. Slurm is the scheduler most of them use: you ask for nodes, cores and time, and your job runs when they are free.
Tile, halo: A part of a large raster processed on its own. A halo is the band of cells around a tile that it needs to read from its neighbours (for a window or a neighbourhood) without writing them.
Command line, terminal: A window where programs are run by typing commands, such as terraflux run model.toml. On Windows it is PowerShell or the Command Prompt; on macOS and Linux, the Terminal.
TOML: A plain-text format for settings, easy to read and write by hand. Terraflux saves models in it.