{"id":91060,"date":"2026-10-21T17:17:17","date_gmt":"2026-10-21T09:17:17","guid":{"rendered":"https:\/\/wp-productionenv-bjg9h2g2bgg5b8aa.southeastasia-01.azurewebsites.net\/?p=91060"},"modified":"2026-10-21T17:17:17","modified_gmt":"2026-10-21T09:17:17","slug":"can-stereo-satellite-imagery-produce-a-bare-earth-terrain-model","status":"publish","type":"post","link":"https:\/\/starpath.global\/blog\/can-stereo-satellite-imagery-produce-a-bare-earth-terrain-model\/","title":{"rendered":"Can Stereo Satellite Imagery Produce a Bare-Earth Terrain Model?"},"content":{"rendered":"

Stereo satellite imagery can support bare-earth terrain mapping where enough ground is visible, but it does not automatically reveal the land beneath trees or buildings. Its first elevation product is usually a digital surface model (DSM), representing the surfaces the cameras can see. Producing a reliable digital terrain model (DTM) requires identifying ground observations, handling gaps and validating the resulting elevations.<\/p>\n

For buyers, the distinction matters. A surface model may support building-height analysis or landscape visualization, while drainage assessment and slope analysis may require bare-earth terrain. Ordering \u201ca satellite DEM\u201d without defining the intended surface can result in a technically valid product that does not meet the project\u2019s needs.<\/p>\n

Start with the Surface You Need<\/h2>\n

A DSM includes visible features such as roofs, tree crowns and exposed ground. A bare-earth DTM represents terrain with vegetation and buildings removed. The term \u201cdigital elevation model,\u201d or DEM, is used differently across organizations, so the name alone is insufficient to define a deliverable.<\/p>\n

On exposed ground, a stereo-derived DSM may closely represent terrain. Under a closed forest canopy, it generally represents the canopy surface. Removing those canopy points leaves gaps; it does not create observations of the hidden ground.<\/p>\n

For hydrological applications, a bare-earth model may also need additional treatment. Bridges, culverts, embankments and drainage channels can affect modeled flow, and their treatment should follow the intended analysis. A bare-earth DTM and a terrain model prepared specifically for hydrological modeling are not necessarily interchangeable.<\/p>\n

Why Two Images Do Not Necessarily Make a Stereo Pair<\/h2>\n

Stereo reconstruction matches the same features in overlapping images taken from different viewpoints. With suitable camera geometry, those matches can be triangulated into three-dimensional points. Two images covering the same location are therefore only potential inputs: their geometry, timing and content must also support reconstruction.<\/p>\n

Viewing Geometry<\/h3>\n

The intersection angle between viewing directions affects height sensitivity. Nearly identical viewpoints provide little geometric leverage for estimating elevation. Greater angular separation can improve height sensitivity, but also increases differences in appearance and the risk that a feature visible in one image is hidden in another. The largest available angle is not automatically the best choice.<\/p>\n

Tri-stereo or additional views can improve coverage around steep slopes and buildings. However, extra optical views cannot recover ground that remains hidden in every image.<\/p>\n

Acquisition Timing and Surface Conditions<\/h3>\n

Images acquired close together reduce differences caused by vegetation movement, construction, shadows and changing surface conditions. Archive images acquired on separate dates can sometimes form a usable pair, but their suitability requires assessment.<\/p>\n

Clouds and deep shadows reduce usable coverage. Water, snow and uniform surfaces can provide too little stable texture for reliable matching. Steep terrain and dense buildings introduce occlusion, leaving areas without reconstructable points.<\/p>\n

Image Product and Sensor Metadata<\/h3>\n

Buyers should request stereo-ready imagery with the sensor geometry required by their processing workflow, commonly supplied as rational polynomial coefficients (RPCs). Ordinary orthorectified images, web-map tiles and screenshots should not be assumed suitable. Confirm the supported product level with the processor before purchasing data.<\/p>\n

STARPATH GLOBAL\u2019s coverage of Jilin-1 stereo mapping<\/a> describes two-view, three-view and multi-view acquisition capabilities. These expand reconstruction options, while bare-earth suitability still depends on ground visibility and validation.<\/p>\n

What Can Be Delivered Under Different Surface Conditions?<\/h2>\n
\n
Stereo Imagery: Terrain Suitability by Surface Condition<\/div>\n\n\n\n\n\n\n\n\n
Surface condition<\/th>\nPotential deliverable<\/th>\nVerification and supplementary data<\/th>\n<\/tr>\n<\/thead>\n
Exposed ground<\/td>\nA terrain surface where ground texture, geometry and coverage support matching.<\/td>\nIndependent ground checkpoints; checks on slopes, shadowed areas and matching gaps.<\/td>\n<\/tr>\n
Sparse vegetation<\/td>\nA filtered terrain estimate supported by visible ground patches, with interpolation between them.<\/td>\nCheck vegetation residuals, ground-point distribution and interpolated areas separately.<\/td>\n<\/tr>\n
Dense forest<\/td>\nPrimarily a canopy DSM. Bare-earth recovery is conditional on sufficient ground visibility.<\/td>\nConsider airborne LiDAR or surveyed terrain. Evaluate leaf-off imagery for deciduous forest and verify under-canopy elevations.<\/td>\n<\/tr>\n
Dense buildings<\/td>\nA roof and visible-ground DSM; a DTM requires building removal and reconstruction across footprints.<\/td>\nInspect narrow streets, building edges and inferred ground beneath buildings; supplement with surveyed elevations or suitable terrain data.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

Can Filtering Recover Ground Beneath Forest?<\/h2>\n

Ground filtering classifies or removes points associated with vegetation and structures. Its success depends on having enough genuine ground observations to preserve. An algorithm can produce a smooth surface across missing areas, but smoothness is not evidence that the hidden terrain has been measured.<\/p>\n

A 2017 study in GIScience & Remote Sensing<\/a> examined stereo satellite imagery in a densely vegetated deciduous forest in Virginia. Filtering improved terrain estimates, but the leaf-on results retained canopy artifacts. Leaf-off imagery produced substantially better results, with filtered elevations comparable to the reference LiDAR terrain model within that study.<\/p>\n

That finding supports testing seasonal acquisition options, rather than assuming that filtering will work equally well in every forest. Evergreen canopy, dense understory and limited ground visibility can leave substantial uncertainty.<\/p>\n

Airborne LiDAR can provide ground returns where some laser pulses pass through canopy gaps. It also needs sufficient ground sampling and careful classification. Ground control points can improve the positioning of a stereo reconstruction, but a few control points cannot replace missing terrain observations across an entire forest.<\/p>\n

Can Sentinel-1 or Sentinel-2 Provide the Missing Terrain?<\/h2>\n

Sentinel-2 repeat coverage should not be treated as a ready-made stereo acquisition service. Two dates of optical imagery do not automatically provide the geometry and detail needed for an engineering terrain model.<\/p>\n

Sentinel-1 uses radar, and interferometric DEM generation follows a different process from optical stereo matching. It requires suitable spatial separation between acquisitions, sufficient coherence and careful handling of atmospheric and phase-unwrapping errors. Vegetation and changing surface conditions can degrade the result. Radar deformation measurements should also not be confused with a new absolute bare-earth terrain model.<\/p>\n

For a project requiring terrain before and after an accident, first establish whether suitable observations exist for each period. Frequent image availability does not guarantee frequent terrain measurements, and new tasking cannot recreate an unobserved historical surface.<\/p>\n

How to Specify and Accept a Bare-Earth DTM<\/h2>\n

Define the Reference Systems and Dates<\/h3>\n

Specify the horizontal coordinate system, vertical datum, elevation units and acquisition dates. Establish whether heights are ellipsoidal or referenced to a geoid-based or local vertical datum, and document any conversion. A datum mismatch can appear as an elevation bias even when the reconstruction is internally consistent.<\/p>\n

Separate Grid Spacing from Vertical Accuracy<\/h3>\n

A \u201c1 m DTM\u201d may describe a one-meter output grid. It does not establish one-meter vertical accuracy, nor prove that every cell contains an independent terrain observation. Resampling an elevation model to a finer grid does not add measured terrain detail.<\/p>\n

Agree on vertical error limits, required feature detail and acceptable completeness separately. Avoid applying a single accuracy claim to exposed ground, forest and dense urban areas without evidence from each relevant condition.<\/p>\n

Keep Checkpoints Independent<\/h3>\n

Use checkpoints that were not used to adjust the imagery or tune ground filtering. Their survey accuracy should be sufficient to assess the agreed tolerance, and their locations should represent the project\u2019s terrain, slopes and land cover.<\/p>\n

Report checkpoint numbers and distribution, signed elevation bias, vertical RMSE and any additional statistics required by the agreed specification. Explain exclusions and reference-data uncertainty. Good results on open roads do not establish accuracy beneath adjacent forest.<\/p>\n

Inspect Gaps, Interpolation and Terrain Features<\/h3>\n

Request masks distinguishing reconstructed areas, interpolated areas, externally supplied terrain and remaining voids. Review the ground-classified point cloud where available. A complete raster may conceal large regions with little direct observational support.<\/p>\n

For drainage or slope applications, inspect channels, embankments, steep breaks and other features that control the analysis. These can be poorly represented even when overall checkpoint statistics appear acceptable.<\/p>\n

Compare the DSM and DTM Along the Same Profile<\/h2>\n

A useful acceptance review places an orthorectified image, the stereo DSM and an independently validated reference DTM side by side, with an elevation profile crossing the same ground features. Use compatible datums and check acquisition dates before interpreting differences.<\/p>\n

Across trees or buildings, the DSM should show the visible elevated surface. The terrain profile should follow ground observations or clearly identified estimates. Highlight locations where the proposed DTM bridges gaps, removes real terrain features or retains vegetation artifacts.<\/p>\n

Differences between the DSM and reference DTM are not automatically reconstruction errors: part of the difference may represent actual object height. Where no reliable reference or independent checkpoints exist, the comparison can explain processing behavior, but cannot establish verified bare-earth accuracy.<\/p>\n

Choose the Data Around the Terrain Requirement<\/h2>\n

Before ordering, provide the area of interest, required observation period, land cover, smallest terrain features of interest and acceptance criteria. Ask for a representative sample that includes difficult surfaces, together with its validation evidence and interpolation masks.<\/p>\n

STARPATH GLOBAL helps match satellite data and analysis to project requirements. Explore stereo imagery and elevation examples in our imagery catalog<\/a>, or discuss your project<\/a> with our team to assess acquisition options and where supplementary terrain measurements may be needed.<\/p>\n","protected":false},"excerpt":{"rendered":"

Stereo satellite imagery can support bare-earth terrain mapping where enough ground is visible, but it does not automatically reveal the land beneath trees or buildings. Its first elevation product is usually a digital surface model (DSM), representing the surfaces the cameras can see. Producing a reliable digital terrain model (DTM) requires identifying ground observations, handling […]<\/p>\n","protected":false},"author":1,"featured_media":91061,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"footnotes":"","_links_to":"","_links_to_target":""},"categories":[3],"tags":[10693,10694,10696,6958,10692,165,10691,10695],"class_list":["post-91060","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog","tag-dsm","tag-dtm","tag-hydrology","tag-lidar","tag-photogrammetry","tag-satellite-imagery","tag-stereo-imagery","tag-terrain-mapping"],"acf":[],"_links":{"self":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts\/91060"}],"collection":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/comments?post=91060"}],"version-history":[{"count":2,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts\/91060\/revisions"}],"predecessor-version":[{"id":91063,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts\/91060\/revisions\/91063"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/media\/91061"}],"wp:attachment":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/media?parent=91060"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/categories?post=91060"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/tags?post=91060"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}