Georeferencing is the process of assigning real-world geographic coordinates to digital data that doesn’t already have location information. Most commonly, this means taking a scanned paper map, aerial photograph, or satellite image and aligning it so that every pixel corresponds to an actual place on Earth. Once georeferenced, that image can be layered with other spatial data in geographic information systems (GIS), making it possible to measure distances, compare datasets, and perform location-based analysis.
Why Georeferencing Matters
A scanned historical map is just a picture. It might show streets, rivers, and property boundaries in fine detail, but without coordinate information attached to it, GIS software has no way to know where on Earth that map belongs. Georeferencing solves this by creating a mathematical link between points in the image and their corresponding locations in a known coordinate system. After that transformation, the image snaps into its correct geographic position and can be overlaid with modern satellite imagery, census data, environmental layers, or anything else that carries location information.
This makes georeferencing essential for a wide range of work: urban planners digitizing old city surveys, archaeologists mapping historic sites, environmental scientists tracking land-use change over decades, and construction teams aligning drone imagery with site plans. Any time you need to bring spatial data without coordinates into a system that depends on them, georeferencing is the bridge.
How the Process Works
The core idea is straightforward. You identify features that appear both in the image you’re georeferencing and in a reference source that already has accurate coordinates, like modern satellite imagery or a georeferenced base map. These shared features are called control points (sometimes called tie points or ground control points). A road intersection visible on both a 1950s aerial photo and current Google satellite imagery, for example, makes a good control point.
You typically need between four and ten control points, spread across different parts of the image. GIS software uses these matched pairs to calculate a transformation that stretches, rotates, and shifts the image until it fits its real-world position. The more control points you place, and the more evenly you distribute them, the more accurate the result. Clustering all your points in one corner leaves the opposite side of the image poorly anchored.
Good control points share a few qualities. They should be features that are stable over time: building corners, road intersections, or bridge abutments rather than tree lines or shorelines that shift. They should be clearly identifiable in both the source image and the reference data. And they should be spread across the full extent of the image, including corners, edges, and the center, so the transformation is consistent everywhere.
Coordinate Systems and Datums
Before you start placing control points, you need to choose a coordinate system for your project. A coordinate system is a reference framework that defines how locations on Earth’s curved surface get expressed as numbers on a flat map. There are two main types. Geographic coordinate systems use latitude and longitude to describe positions on a three-dimensional model of the Earth. Projected coordinate systems take that sphere and flatten it mathematically onto a two-dimensional plane, which is what you see on a printed map or screen.
Every coordinate system relies on a datum, which is essentially the specific model of Earth’s shape that the coordinates are built on. The most widely used datum in civilian GPS and web mapping is WGS 84, defined by the U.S. Department of Defense. For higher-precision scientific work, the International Terrestrial Reference Frame (ITRF) serves as the current global standard. Choosing the wrong datum or coordinate system can shift your georeferenced image by meters or even kilometers, so matching your project’s settings to your reference data is a critical first step.
What Happens to the Pixels
When GIS software transforms an image to fit new coordinates, the original grid of pixels rarely lines up perfectly with the output grid. The software has to decide what color or value each new pixel should get. This step is called resampling, and there are three common methods.
- Nearest neighbor assigns each new pixel the value of whatever original pixel is closest. It’s fast and preserves the original data values exactly, which matters when your image represents categories like land-use types. The tradeoff is a slightly blocky appearance and the possibility of small positional shifts (up to half a pixel).
- Bilinear interpolation averages the four nearest original pixels, weighted by distance. This produces smoother results and works well for continuous data like elevation models or temperature maps, but it alters the original values slightly.
- Cubic convolution uses 16 surrounding pixels and a more complex weighting formula. It produces the sharpest visual results with enhanced edges, making it the best choice for aerial photos and remote sensing imagery. It’s slower to compute and can generate values slightly outside the original data range.
For most people georeferencing scanned maps, bilinear interpolation is a safe default. If you’re working with classified data where exact category values matter, use nearest neighbor.
Measuring Accuracy
After placing control points and running the transformation, GIS software reports how well the image fits its new coordinates. The standard metric is root mean square error (RMSE), which measures the average distance between where your control points ended up and where the reference data says they should be. A lower RMSE means a tighter fit.
There’s no single universal threshold for acceptable accuracy. The American Society for Photogrammetry and Remote Sensing (ASPRS) publishes positional accuracy standards that let project managers define the required precision based on the work at hand. A construction survey might demand accuracy within a few centimeters. Georeferencing a 19th-century hand-drawn map to study general urban growth patterns might tolerate errors of several meters. What matters is that the accuracy matches the purpose. If one control point shows dramatically more error than the others, it’s often worth removing or repositioning it and re-running the transformation.
Georeferencing vs. Geocoding
These two terms sound similar but describe different operations. Georeferencing takes image data (pixels) and assigns it geographic coordinates so it can be placed on a map. Geocoding takes a text address, like “123 Main Street, Austin, TX,” and converts it into a latitude and longitude point. Geocoding relies on large databases of street addresses and place names to make that translation. Both produce geographic coordinates as output, but they start from fundamentally different inputs: images on one side, text addresses on the other.
A Typical Workflow in GIS Software
Most georeferencing today happens in free or commercial GIS platforms like QGIS, ArcGIS Pro, or ArcMap. The general workflow is similar across tools. In QGIS, for example, you start by setting your project’s coordinate system, then load a reference layer such as Google satellite imagery through a plugin. You open the Georeferencer tool, load your scanned map image, and begin placing control points by clicking a recognizable feature on the scanned map and then clicking the same feature on the satellite imagery.
After placing at least four well-distributed points, you choose a transformation type (a first-order polynomial works for most scanned maps with minimal distortion) and a resampling method, then run the process. The software outputs a new version of your image with embedded coordinate information. You can then overlay it with any other georeferenced data in your project, trace features from the old map into new vector layers, or run spatial analyses that combine historical and modern information.
The entire process for a single map typically takes 15 to 30 minutes once you’re familiar with the software. The time-consuming part is usually finding reliable control points, especially on older maps where landmarks may have been demolished or streets rerouted. Loading multiple reference sources, like both satellite imagery and a modern street map, can help you identify matching features more confidently.