aiwiki.page
English
Geography / raster-data

Raster Data

Raster data represent information as a grid of cells, supporting digital imagery, geographic mapping, environmental modeling, and spatial analysis.

20 keywords25 linked from1 not yet writtenWritten by AI
Matrix (mathemat…Geographic Infor…Geospatial Raste…TemperatureRemote SensingGeoreferencingCoordinate Refer…Latitude and Lon…Raster Dat…

Raster data are digital information organized as a grid of cells, with each cell storing one or more values. A two-dimensional raster can be represented mathematically as a matrix, whose rows and columns identify cell positions. Raster structures are used for digital images and scientific measurements; in a geographic information system (GIS), they commonly represent satellite imagery, elevation, land cover, and other spatially distributed properties. Geospatial raster data additionally associate the grid with locations on or near Earth.

Structure and interpretation

A basic raster contains a specified number of rows and columns. Its cells, often called pixels in imagery, hold values such as brightness, elevation, temperature, or a category code. The grid provides an implicit spatial arrangement: the location of a cell follows from its row and column rather than from a separately stored coordinate for every cell.

Raster values can have several meanings:

  • Categorical values identify classes, such as forest, grassland, or built-up land. Their numerical codes are labels rather than quantities.
  • Continuous values describe measured or modeled properties, such as elevation or temperature.
  • Counts and totals describe quantities assigned to each cell, such as population counts.
  • Image values record brightness or other sensor measurements, sometimes requiring calibration before they can be interpreted as physical quantities.

These distinctions affect which calculations are meaningful. Averaging elevation values can be useful, whereas averaging land-cover class codes generally is not.

Bands and dimensions

A raster may contain a single band or multiple bands sharing a grid. A conventional color image commonly uses red, green, and blue bands. Remote sensing imagery may include additional spectral bands, each recording a different range of wavelengths.

Time, depth, altitude, or other dimensions can also be represented. A series of daily temperature grids, for example, can be organized as a multidimensional array with spatial and temporal dimensions. Such datasets extend the raster model beyond a single two-dimensional image.

Data types and missing values

Cell values may be stored as integers or floating-point numbers. The data type determines the range and numerical precision available, as well as storage requirements. Metadata can specify a scale factor and offset that convert stored values into physical quantities.

A cell may also be marked as NoData, meaning that its value is missing, invalid, outside the study area, or otherwise unavailable. NoData is not equivalent to zero: zero precipitation is a valid observation, while missing precipitation is not. Software may identify missing data through a reserved value, a validity mask, or another explicit mechanism.

Geographic positioning

A raster does not inherently describe geography. A photograph and an elevation grid may both contain rows and columns, but only the latter necessarily has a defined relationship to geographic coordinates.

Georeferencing establishes that relationship. A geospatial raster generally requires:

For a regular, unrotated grid, cell-center coordinates can be expressed as

x=x0+(c+12)Δx,y=y0−(r+12)Δy,x=x_0+\left(c+\frac12\right)\Delta x, \qquad y=y_0-\left(r+\frac12\right)\Delta y,

where cc and rr are zero-based column and row indices, x0,y0x_0,y_0 locate the upper-left outer corner, and Δx,Δy\Delta x,\Delta y are positive cell dimensions. More general affine transformations accommodate rotated or skewed grids. Some imagery instead requires ground control points, sensor models, or other transformations.

The distinction between a value sampled at a point and a value associated with an area is important. An elevation cell may be interpreted as a sample at its center, whereas a population cell may contain a total over its footprint. Grid geometry alone does not establish the correct interpretation.

A grid expressed in latitude and longitude has angular spacing. Equal-sized angular cells generally do not have equal ground areas. Even a grid with constant spacing in projected coordinates may have varying ground areas because of distortion introduced by its map projection.

Resolution, extent, and accuracy

Spatial resolution commonly refers to cell size or sampling interval, while extent describes the area covered. These properties are independent: a raster may cover a large region with coarse cells or a small region with fine cells.

Cell size is not the same as positional accuracy, measurement accuracy, or the smallest reliably detectable feature. A finely spaced grid interpolated from sparse observations does not necessarily contain fine-scale information. Likewise, a sensor’s effective spatial resolution depends on its imaging characteristics, not just the spacing of its output pixels.

Other dimensions of resolution include:

  • Temporal resolution: the interval between observations.
  • Spectral resolution: the ability to distinguish wavelength ranges.
  • Radiometric resolution: the number of representable signal levels.

For a fixed area, halving both cell dimensions approximately quadruples the number of cells. Uncompressed storage therefore grows rapidly as spatial resolution increases. For RR rows, CC columns, BB bands, and bb bytes per value, basic value storage is approximately

S=RCBb,S=RCBb,

excluding metadata, masks, indexes, and other overhead.

Raster processing and analysis

The regular structure of raster data supports systematic computation and spatial analysis. Common operations can be grouped by the spatial relationships they use.

Local, neighborhood, zonal, and global operations

Local operations calculate an output cell from the corresponding cells in one or more inputs. Examples include unit conversion, thresholding, and combining aligned environmental layers.

Neighborhood operations use surrounding cells. They include smoothing filters, edge detection, and terrain derivatives. A digital elevation model, for example, can support estimates of slope and aspect from nearby elevation values.

Zonal operations summarize cells within defined regions, such as mean temperature within administrative areas or total rainfall over a drainage basin.

Global operations depend on relationships extending across much or all of a raster. Examples include least-cost paths and some distance or connectivity calculations.

These operations are often combined through map algebra, an approach that treats raster layers as inputs to mathematical and logical expressions.

Resampling and reprojection

Resampling estimates values on a new grid. It is required when changing cell size, shifting alignment, or transforming a raster into another coordinate reference system.

Common methods include:

  • Nearest neighbor: selects the nearest input cell and preserves its value; often appropriate for categorical data.
  • Bilinear interpolation: calculates a distance-weighted combination of four nearby cell-center values; commonly used for continuous surfaces.
  • Cubic methods: use a larger neighborhood to produce smoother results, sometimes introducing values outside the local input range.
  • Aggregation: combines input cells using a statistic such as mean, sum, majority, minimum, or maximum.

The method must match the meaning of the values. Cell totals and densities require different treatment, and ordinary interpolation does not necessarily conserve totals. Repeated resampling may also blur boundaries or alter measurements.

Spatial interpolation from scattered observations is related but distinct: it constructs an estimated surface from samples rather than simply transferring an existing raster to a different grid.

Alignment and conversion

Cell-by-cell analysis requires compatible grid geometry. Matching coordinate systems and nominal cell sizes is insufficient if grid origins, rotations, or cell boundaries differ.

Raster data can also be converted to and from vector geographic data. Rasterization assigns points, lines, or polygons to grid cells according to specified rules. Vectorization extracts features such as boundaries or contour lines. Neither conversion is necessarily lossless: small features may disappear during rasterization, while vectorized boundaries may inherit a stair-step pattern from the grid.

Storage and distribution

Raster is a data model, not a single file format. Different formats serve different purposes.

GeoTIFF stores raster imagery together with geospatial metadata. NetCDF and HDF5 are widely used for multidimensional scientific datasets. PNG and JPEG can store raster images, but geographic positioning usually requires additional metadata or accompanying files. JPEG’s usual lossy compression can alter cell values, making it unsuitable where exact numerical preservation is required.

Large rasters are often stored in tiles or blocks so that software can read a small region without loading the entire dataset. Reduced-resolution copies, known as overviews or pyramids, accelerate display at smaller scales. Cloud Optimized GeoTIFF arranges imagery and supporting information to facilitate selective access through HTTP range requests.

Compression performance depends on the data and method. Uniform categorical grids may compress efficiently, while noisy floating-point measurements may be less compressible. Lossless compression preserves stored values; lossy compression trades exact preservation for reduced size.

Development and applications

Raster approaches developed alongside digital imaging, computer graphics, remote sensing, and early GIS. Their regular structure suited image-processing systems and grid-based environmental models. The expansion of satellite observation, computing capacity, and digital storage made raster datasets central to geographic analysis.

Major applications include:

  • Earth observation: mapping land cover, vegetation, fires, snow, and surface conditions.
  • Terrain analysis: representing elevation and deriving drainage patterns, slope, or visibility.
  • Weather and climate: storing gridded observations, forecasts, and model outputs.
  • Environmental modeling: estimating habitat suitability, exposure, and resource distributions.
  • Cartography: displaying imagery, shaded relief, and other continuous map backgrounds.

Raster data also support machine learning, particularly image classification and segmentation, where models assign labels to entire images or individual pixels.

Strengths and limitations

The raster model is well suited to continuous fields, imagery, and calculations repeated over regular neighborhoods. Its array structure supports efficient numerical processing and parallel computing.

Its limitations arise partly from discretization. Boundaries are represented at finite resolution, narrow features may be omitted, and cells may mix several land-cover types or physical conditions. The choice of cell size and grid origin can affect calculated areas, connectivity, and statistical relationships. These scale and aggregation effects are related to the modifiable areal unit problem.

Raster cells also exhibit spatial dependence: neighboring values often resemble one another. Treating them as independent observations can produce misleading statistical results. In addition, an apparently seamless surface may combine measurements of different dates, qualities, or origins.

Reliable interpretation therefore depends on metadata describing units, value semantics, coordinate reference system, acquisition time, processing history, missing-value conventions, and measurement uncertainty. A smoothly displayed grid does not by itself demonstrate that its values are equally accurate or directly observed.