ON THIS PAGE
Satellite and aerial imagery is often presented as though acquisition is the end product: an image is captured, downloaded and used. In professional geospatial workflows, that is rarely the full story.
Raw or minimally processed imagery may still contain geometric distortion, inconsistent colour, scene edges, overlap, varying projections or file structures that are inconvenient for enterprise use. Image processing is the production stage that turns those observations into a consistent geospatial product that can be measured, compared, published and integrated with GIS.
“Map-ready” should be defined, not assumed
Different teams can mean different things by map-ready. A visualization team may simply need imagery that displays correctly in a web map. A GIS analyst may need reliable alignment with authoritative vectors. A change-detection workflow may need two dates normalized as consistently as possible. An engineering user may expect documented positional accuracy.
- intended use;
- coordinate reference system;
- required positional accuracy;
- pixel size and output resolution;
- spectral product;
- format;
- tiling structure;
- compression;
- background/no-data value;
- delivery environment.
Orthorectification corrects image geometry
Imagery is collected from a sensor looking at a three-dimensional Earth. Terrain elevation, viewing angle, sensor position and platform motion all affect where objects appear in the raw image. Orthorectification uses sensor geometry and elevation information to correct those effects and place the imagery into a map coordinate system.
Elevation data affects orthorectification quality
If terrain elevation is represented incorrectly, features can shift horizontally during correction. The effect is particularly visible in mountains and around tall structures.
Projection decisions should be explicit
Imagery can be delivered in geographic coordinates or projected coordinate systems. Government and engineering programmes may require a national or project-specific CRS. Reprojecting imagery resamples the raster grid, so unnecessary repeated reprojection should be avoided.
Mosaicking combines multiple scenes into one coverage
Large areas are commonly covered by more than one image or aerial strip. Mosaicking combines those sources into a continuous product. A technically correct mosaic needs more than edge-to-edge placement because images may overlap, have different acquisition dates or differ in brightness and geometry.
Seam lines
The seam line determines where the mosaic transitions from one source image to another. Poor seams can cut through buildings, bridges or roads. Good seam-line design places transitions where they are less noticeable while preserving valid coverage.
Radiometric balancing improves consistency
Different scenes can vary in brightness, contrast and colour because of sun angle, atmosphere, sensor conditions and date. Radiometric normalization and colour enhancement aim to create a consistent visual product while preserving useful image information.
Visual consistency and analytical integrity are not the same thing
If imagery will be used for quantitative remote sensing, aggressive balancing may be inappropriate. Analytical and visual products may need different processing strategies.
Pansharpening and multispectral outputs serve different needs
Pansharpening combines high-resolution panchromatic spatial detail with multispectral colour information. The result is often ideal for visual interpretation, while analysts requiring original spectral values may also need separate multispectral data.
No-data and black-edge handling matter operationally
Orthorectified scenes can produce empty areas around valid image data. If no-data values are handled incorrectly, those regions may appear as black blocks in mosaics or interfere with analysis.
Pyramids and overviews improve GIS performance
Very large rasters can be slow to display. Raster pyramids or overviews create reduced-resolution levels for efficient display at smaller map scales.
File format is part of product design
GeoTIFF is widely interoperable. IMG and other formats can support large mosaics. Cloud-optimized GeoTIFF can be useful in cloud/web architectures because applications can request parts of a large file rather than downloading everything.
Tiling versus a single mosaic
A single mosaic is conceptually simple, but extremely large files can be difficult to move, update and back up. Tiled delivery can improve performance and partial-update workflows.
- maximum file sizes;
- desktop GIS performance;
- server publishing;
- partial updates;
- backup strategy;
- cloud/object storage;
- client software compatibility.
Resampling method affects raster values
Whenever imagery is reprojected or placed onto a new grid, pixel values must be resampled. Common methods include nearest neighbour, bilinear interpolation and cubic convolution.
Nearest neighbour preserves original pixel values and is often preferred where categorical or analytical integrity matters. Bilinear and cubic approaches can produce smoother visual results but modify pixel values.
Compression balances size, quality and performance
Large imagery programmes consume significant storage. Lossless compression preserves values exactly but may achieve less reduction. Lossy compression can substantially reduce size but may introduce artifacts.
For analytical source imagery, preserving values may be important. For visual basemaps, controlled lossy compression may be acceptable.
Processing for change detection requires extra care
When two dates will be compared automatically, geometric and radiometric consistency become critical. Small alignment errors can appear as false change along every building edge.
- fine co-registration;
- resampling to a common grid;
- radiometric normalization;
- cloud and shadow masks;
- stable-area selection;
- residual-alignment metrics.
Quality assurance should be built into every stage
Geometric QA
- projection correct?
- pixel size correct?
- expected extent covered?
- alignment acceptable?
- no shifts between adjacent tiles?
Radiometric QA
- brightness consistent?
- seams acceptable?
- no clipped highlights?
- no band mismatch?
File QA
- files open correctly?
- overviews built?
- no-data correct?
- metadata complete?
- naming follows specification?
Production lineage should be documented
When several scenes are processed into a final mosaic, users should be able to trace the output back to source identifiers, acquisition dates, processing steps, elevation source, resampling method and production date.
Enterprise publishing changes the definition of delivery
For some clients, handing over files is only half the job. The imagery must also be exposed through ArcGIS Enterprise or another platform with appropriate pyramids, cache strategy, permissions and service metadata.
SIME image-production workflows
SIME’s production capabilities include strip mosaicking, orthorectification, mosaicking, seam-line correction, colour enhancement, background-value handling, black-edge removal, GIS pyramid/definition work and final mosaic preparation.
Explore Image Processing or describe the final imagery product your project requires.
Resampling method can affect analytical values
Whenever imagery is reprojected, rotated or aligned to a new raster grid, new pixel values must be calculated. Nearest-neighbour, bilinear and cubic resampling behave differently.
Nearest-neighbour resampling preserves original pixel values and is often preferred when categorical or spectral integrity matters. Bilinear and cubic methods interpolate values and can create smoother visual results. The processing team should select the method based on the intended use rather than relying on a software default.
Compression is a technical decision, not only a storage decision
Lossless compression reduces file size without changing pixel values. Lossy compression can reduce size much more aggressively but may introduce artifacts. For a visual web basemap, controlled lossy compression may be acceptable. For analytical source imagery, it may not be.
Many organizations benefit from retaining a high-quality analytical master and creating optimized delivery derivatives for web or desktop display.
Production lineage supports auditability
When several scenes are transformed into a final mosaic, future users should be able to understand where the data came from. Useful lineage can include source scene identifiers, acquisition dates, processing software, elevation source, orthorectification parameters, resampling method, seam edits and production date.
This is especially valuable when imagery is later used for legal evidence, long-term monitoring, AI training or government archival purposes.
Cloud-oriented imagery delivery is changing raster architecture
Traditional workflows often assume a user downloads entire files. Cloud-oriented formats and image services allow applications to request only the pixels and resolution required for the current view.
For large programmes, this can reduce transfer overhead and simplify enterprise access. However, cloud optimization should be designed alongside security, storage and GIS publishing requirements.
Automated production still requires visual QA
Many processing steps can be automated, but automation can propagate systematic errors quickly. A bad seam rule, incorrect no-data value or CRS mistake can affect thousands of tiles.
Automated checks should therefore be combined with visual sampling and exception review. QA should be risk-based: focus human attention where automated metrics indicate unusual geometry, colour or file behavior.
Tile naming and folder structure affect long-term usability
Large deliveries often contain hundreds or thousands of raster files. A consistent naming convention should encode enough information to identify the tile or product without creating unreadable filenames. Folder structure should separate source, intermediate, final and metadata products where appropriate.
For enterprise clients, naming should align with ingestion scripts and future updates so a new delivery can replace or append data predictably.
Automated ingestion should be tested against final outputs
If imagery will be loaded into an enterprise catalog, image service or cloud workflow automatically, test the ingestion process before full production. Small differences in metadata, extension, band order or no-data handling can break automated pipelines.
Preserve an analytical master before creating display derivatives
A good production architecture often keeps a high-quality master product and creates separate derivatives optimized for web display, mobile use or reporting. This avoids repeated processing from already compressed or resampled outputs.
Document exceptions rather than hiding them
Some areas may contain unavoidable cloud, source-date differences, seam compromises or incomplete coverage. These should be recorded in QA notes and metadata so future users understand the limitation.
