An aerial orthomosaic is a continuous, map-scaled image of the ground surface produced by joining vertical images captured with a UAV into a single seamless dataset. Each image is first tied to a real-world coordinate frame using ground control points and GNSS/IMU data, then reprojected to correct relief displacement and lens distortion before being rearranged into one grid. The result becomes the basis for area measurement, feature extraction and elevation modelling. Final mosaic quality depends above all on control accuracy, flight overlap and the ground sample distance of the imagery. Structure from Motion and Multi-View Stereo processing form the technical foundation of this workflow.
Geometric base and true orthophoto
A true orthophoto corrects relief displacement per pixel against a surface elevation model rather than a single average height. In the common case the correction uses a bare-earth model, so building and canopy heights remain as lateral lean; that effect introduces meaningful error in urban parcel mapping and area measurement.
Aerial triangulation, solved through bundle adjustment while minimising reprojection error, refines the interior and exterior orientation of every image. Using an RTK receiver or PPK processing reduces the number of required control points, yet a few independent check points should still be held back for a genuinely independent accuracy assessment.
The final geometric accuracy is reported as an RMSE in GSD units, for example a horizontal RMSE of one to two ground pixels. Inputs at this stage should be documented so the result stays reproducible.
- Flight specification: altitude, speed, flight axis, and along- and across-track overlap.
- GNSS/IMU data and raw image files together with camera metadata.
- Ground control points with known coordinates in an agreed reference frame, such as WGS84 or UTM with a stated EPSG code.
- A base elevation model, where available, to make the relief correction true rather than nominal.
Seam lines and radiometric balancing
Seam lines are the paths where two neighbouring images are stitched together. A seam should cross homogeneous areas such as roads and open ground, and stay away from sharp edges, tall trees and building facades; otherwise ghosting and repeated edges appear in the mosaic.
Colour balancing smooths brightness and colour differences between images. Vignetting, a changing sun angle across the flight and directional surface reflectance are the main sources of colour inhomogeneity, and they are reduced through radiometric normalisation and, where needed, relative normalisation against a reference image.
- Grouping images by time and lighting conditions for more homogeneous balancing.
- Avoiding manual, non-repeatable colour tweaks that destroy reproducibility.
- Field checking of colour over known surfaces such as asphalt and bright roofs.
Spatial resolution and ground sample distance
Ground sample distance follows from flight altitude, camera focal length and sensor pixel size; a lower altitude tightens the GSD but reduces per-flight coverage and increases data volume. The chosen GSD should match the smallest feature to be identified, not simply the wish for a higher-looking number.
Resampling to a pixel finer than the imagery can actually support only enlarges the file without adding information. For delivery it is better to keep the native GSD and let downsampling happen during publishing and tiling.
Effective resolution is limited by motion blur, UAV vibration and lens quality, so nominal GSD is not always equivalent to the smallest reliably detectable feature.
Tiling, overviews and Cloud Optimized GeoTIFF
Large mosaics must be split into regular tiles and stored with image overviews (pyramids) so they load and pan smoothly. These overviews give fast rendering at low zoom levels and prevent a full-file read for every view.
A Cloud Optimized GeoTIFF is a GeoTIFF whose internal layout supports HTTP range requests, so software can fetch only the needed part from the server. Combined with internal tiling and overviews, COG makes streaming of heavy data possible without downloading the whole file.
Compression is chosen among DEFLATE, LZW and JPEG based on data type; lossless compression is preferred for colour imagery where tonal fidelity matters. Internal tile size should match the access pattern of the consuming tool.
Multi-season mosaicking and overlap handling
When an area is covered across several seasons or flights, merging them introduces differences in vegetation phenology, sun angle and soil moisture. Ranking images by date and quality preserves visual consistency better than a plain stitch.
In overlap strips, feathering or gradual weighting blends the boundaries. In areas with tall features, choosing seams by distance from the image centre reduces the effects of occlusion and stretch.
Before merging, the consistency of the reference frame, a uniform GSD and matching block footprints must be verified; a mismatch in any of these produces local shifts and elevation differences in the output.
Quality control and deliverables
Quality control should run independently of production so a systematic error cannot stay hidden. Visual inspection of seams, measurement of residual deviation against independent check points, and matching the extent to the contractual boundary are the three baseline steps.
The technical report should record the processing software version, key settings, control point sources and error indicators so the result can be reproduced and compared later.
- Orthomosaic as GeoTIFF or COG with image overviews.
- Error report in GSD units with a control and check point statement.
- Metadata following ISO 19115 to describe source, date and production method.
- Where required, a surface elevation model and publish-ready tiles.
Publishing through tiled services
For organisational use the mosaic is published through standard services. WMTS and WMS for image delivery and XYZ tiles for web mapping libraries are the most common options, deployed with GeoServer or MapServer.
Setting zoom level ranges and a caching strategy controls server load; pre-rendering frequently used tiles makes responses more stable.
Access should be governed by authentication and user roles, particularly where imagery is location-sensitive or intended for a specific group.
Summary
Aerial orthomosaic production is a multi-stage process that begins with geometric orientation, continues through seam selection and colour balancing, and ends with tiling and publishing. Output quality depends not only on computational accuracy but on input accuracy and the transparency of settings.
We recommend fixing GSD, reference frame and expected accuracy at the outset, and running quality control against independent points. Following these steps turns the mosaic into data that can be relied on for measurement, monitoring and decision making.
Published through tiling services, the output reaches users across organisational GIS systems and web tools, and the data update cycle becomes considerably easier to maintain.