Skip to content

Comparing chunks for change detection: DEM, mesh, and point-cloud diff workflows

  • Status: unverified
  • Applies to: Metashape Pro 2.x — and PhotoScan 1.x; DEM differences are native (Transform DEM), while mesh/point-cloud diffs route through external tools
  • Edition: Pro
  • Diátaxis: how-to
  • Confidence: high
  • Last reviewed: 2026-06-05

Confidence: high. DEM differencing is native (Transform DEM → Calculate difference, Workflow 0). The remaining point — that there is no one-click chunk-to-chunk mesh / point-cloud diff — is forum-attested (Agisoft support, 2017) and still holds in 2.x; the external-tool triage is reconstructed from multi-author thread replies.

Problem

You have two chunks of the same site captured at different times — before / after excavation, before / after construction, seasonal change of a glacier or river. You need to quantify the change: cubic metres of material removed, areas of subsidence, average height difference. Metashape exposes per-chunk Measure Volume and Measure Area tools, and — for DEM surfaces — a native difference operator (Tools → Transform DEM with Calculate difference, described next). For mesh or point-cloud change, or for cross-package analysis, there is still no one-click chunk-to-chunk operator, so the established route is to export and diff in an external tool.

"In current version of PhotoScan you cannot subtract DEM models produced in different chunks, so you need to do that with the exported results in the external GIS applications (like GlobalMapper or Q-GIS, for example)." — Agisoft support, 2017-05-15, PhotoScan 1.3 (permalink)

That 2017 answer is now outdated for DEMs. Current Metashape Professional can subtract one DEM from another natively via Tools → Transform DEM with the Calculate difference option: import the other chunk's DEM into the project first (or keep both as DEM instances in one chunk), then difference them to produce a result DEM — DEM difference calculation (Agisoft KB). The external-tool route below is still worthwhile for finer control over resampling, for mesh / point-cloud diffs that a 2.5-D DEM cannot represent, and for cross-package analysis.

Pre-requisite: chunk co-registration

Before any diff operation produces meaningful numbers, both chunks must be in the same world coordinate frame. A 1 cm horizontal misalignment can produce 100+ m³ of spurious volume difference on a 100 m × 100 m area.

Three co-registration strategies:

  1. Shared GCPs. Most robust. Place the same surveyed marker set in both chunks; the markers' identical world coordinates force the chunks into a common frame.
  2. Align Chunks → Method: Marker. Use markers visible in both chunks (e.g., painted ground targets that survived from epoch 1 to epoch 2).
  3. Align Chunks → Method: Point cloud. ICP-style alignment on the dense clouds; works when no markers exist but stability is poor near the actively-changing area (the alignment is biased by the change region).

Verify alignment quality after Align Chunks using Tools → Reference → Inspect Markers — marker errors should be sub-cm for high-quality diff work.

For the broader chunk-alignment topic see Align Chunks: three methods.

Four diff workflows

Workflow Tool Best for Output
0. Native DEM difference Metashape Transform DEM quick in-app DEM change, same project Result DEM layer
1. DEM raster subtraction QGIS / GlobalMapper / ArcGIS rasterised surfaces (terrain) Difference GeoTIFF
2. Point-cloud cloud-to-cloud CloudCompare overhanging or non-rasterisable geometry Per-point distance values
3. Per-chunk volume measurement Metashape native quick excavation / earthwork volumes Two scalar volumes; subtract manually

Workflow 0 — native DEM difference (Transform DEM)

For DEM surfaces this is the quickest path and needs no external tool. Bring both epochs' DEMs into one project, then:

  1. File → Import → Import DEM to add the second epoch's DEM alongside the first (a chunk can hold several DEM instances).
  2. Select the DEM to subtract from, open Tools → Transform DEM…, enable Calculate difference, and pick the other DEM from the dropdown.
  3. OK produces a new difference-DEM layer (positive = accumulation, negative = loss) that you can measure or export.

It is exact and georeferenced but shares the DEM's 2.5-D limitation (no overhangs), and gives less control over resampling than a GIS. For mesh / point-cloud change use the export-and-diff workflows below. See DEM difference calculation (Agisoft KB).

Workflow 1 — DEM raster subtraction (GIS)

Standard workflow for terrain change detection (excavations, landslides, glacier retreat):

  1. Build DEM in each chunk: Workflow → Build DEM with identical projection, identical pixel size, identical bounding box.
  2. Export each as GeoTIFF: File → Export → Export DEM. Set Region identically in both exports (use the same shape boundary in both chunks).
  3. Open both rasters in QGIS: Raster → Raster Calculator 
    "after@1" - "before@1"
  4. The output is a difference raster in metres (positive = accumulation, negative = excavation).
  5. Compute volume: Raster → Zonal Statistics on the difference raster, or r.volume in GRASS, or: 
    total_volume = SUM(diff_pixel_value × pixel_area_m²)

Pros: - Preserves georeferencing; output is directly drape-able on basemaps. - Resolution-controllable (downscale before subtraction for smoothing). - Handles arbitrary horizontal extents.

Cons: - 2.5D — fails on overhanging geometry (cliffs, caves, buildings). - Requires both DEMs to be in the same horizontal CRS. - Pixel-size mismatch between exports forces a resampling step (use bilinear interpolation; nearest-neighbour exaggerates noise).

Quick QGIS Python recipe

Demo verified: ✗ — This is QGIS Python, not Metashape Python; verify the QGIS API on your install.

# QGIS Python console
from qgis.analysis import QgsRasterCalculator, QgsRasterCalculatorEntry

before = iface.mapCanvas().layers()[0]  # your "before" DEM layer
after  = iface.mapCanvas().layers()[1]  # your "after" DEM layer

entries = [
    QgsRasterCalculatorEntry().__init__(name="b@1", layer=before, ref=1),
    QgsRasterCalculatorEntry().__init__(name="a@1", layer=after,  ref=1),
]
calc = QgsRasterCalculator(
    "a@1 - b@1",
    "/path/diff.tif",
    "GTiff",
    after.extent(),
    after.width(),
    after.height(),
    entries,
)
calc.processCalculation()

Workflow 2 — CloudCompare cloud-to-cloud

For high-precision diffs or non-rasterisable geometry (cliffs, indoor scans, vegetation canopy):

  1. Build dense point cloud in each chunk: Workflow → Build Point Cloud. Use Mild filtering for change-detection accuracy (Aggressive removes legitimate small-scale features alongside noise).
  2. Export both: File → Export → Export Points with crs = chunk.crs (see exportPointCloud defaults: GUI vs Python CRS difference).
  3. Open both in CloudCompare.
  4. Verify alignment in 3D view — should overlap visually except in the change region.
  5. Tools → Distances → Cloud-to-Cloud Distance:
  6. Reference cloud: "before"
  7. Compared cloud: "after"
  8. Output: per-point signed distance.
  9. Apply a colour ramp on the distance scalar field for visualisation; export the coloured cloud as PLY.

Pros: - Higher spatial resolution than rasterised DEMs. - Handles overhanging / non-2.5D geometry. - Free, well-documented (CloudCompare wiki).

Cons: - The output is a per-point distance; converting to volume requires extra steps (e.g., 2.5D rasterisation followed by workflow 1). - Asymmetric: cloud-to-cloud distance is not symmetric (B→A ≠ A→B in general); use M3C2 (a CloudCompare plugin) for signed-distance with normal-aware comparisons.

Workflow 3 — Per-chunk volume measurement

For quick excavation / earthwork volume estimates without external tools:

  1. In each chunk, draw the same polygon shape covering the area of interest. Critical: the polygons must have identical vertices in world coordinates. The simplest way: draw the shape in chunk 1, copy it to chunk 2 via: 
    chunk_1.shapes → right-click → Copy →
chunk_2.shapes → right-click → Paste
  2. In each chunk: right-click the shape → Measure Volume. The dialog asks for a reference plane; use Best Fit Plane in both chunks for consistency.
  3. Subtract: 
    excavated_volume = volume_before - volume_after

Pros: - No external tools required. - Polygons-from-Metashape preserve exact vertex coordinates, guaranteeing identical reference geometries.

Cons: - The "reference plane" choice (best-fit vs three-point) must match between epochs. Using Best Fit in one chunk and Three Points in the other yields incomparable volumes. - One scalar per epoch: no spatial distribution of change. - Multi-region projects need one polygon per region — cumbersome for many regions.

You can also draw a polygon shape and use the Measure Volume tool, which reports the volume between the polygon and the DEM or mesh surface (Tools → Measure → Volume, on a closed shape).

Caveats

  • Chunk alignment quality dominates. If your alignment marker error is 5cm RMS, your volume diff is unreliable below that scale. Use post-processed kinematic (PPK) GNSS or surveyed monuments for high-precision diff work.
  • Vegetation, wind, lighting changes between epochs add noise that the diff cannot distinguish from real change. Where possible: capture both epochs at similar time of day, similar leaf-on/leaf-off state, similar wind. For glacier / earthwork monitoring, late-summer captures with low vegetation give the cleanest diffs.
  • Edge effects. Both DEMs and point clouds have lower density and higher noise at the project's edges. Crop the diff result to the well-overlapped interior using a smaller inner shape boundary.
  • Differential scale errors. If chunk 1 has scale-bar control but chunk 2 does not, a small scale difference can manifest as systematic Z-error proportional to height above ground. Use shared scale bars or shared GCPs.
  • Filter consistency. Both chunks must use the same point-cloud filter setting (all Mild, all Moderate, all Aggressive) — otherwise you're diffing different representations of the surface, not different surfaces.

See also

References