Skip to content

Method · Geodata & remote sensing · AI consulting

A woodland beneath the leaves

Method case: official laser scan data become a terrain model, six relief views become a short, reasoned candidate list – reviewed blind instead of believed blindly.

The same woodland: aerial image on the left, LiDAR terrain model with highlighted hollows on the right
Abb. 01The same woodland: aerial image on the left, LiDAR terrain model with highlighted hollows on the right
verified relief views of the same terrain, each revealing different forms

6

verified relief views of the same terrain, each revealing different forms

LGL BW, dl-de/by-2.0, modified

every candidate is reviewed without knowing the location; every rejection is reasoned and cross-checked

blind

every candidate is reviewed without knowing the location; every rejection is reasoned and cross-checked

method presented publicly: talk “Context Before Pixels”

CAA 2026

method presented publicly: talk “Context Before Pixels”

CAA Joint Chapter Meeting, Münster, 17 Sep 2026

01Starting point

Under forest, landforms are invisible in aerial imagery. Laser scan data from the survey administration show the ground – but they are hard to read, and an AI that delivers unchecked hits produces long lists full of forest tracks and timber yards.

02Approach

  1. 01From the point cloud to the terrain model: vegetation drops away, the ground remains.
  2. 02Six relief views – hillshade, sky-view factor, local relief model, openness, local dominance, multi-scale relief – because each view highlights different forms.
  3. 03AI screening tile by tile; the AI suggests, it does not decide.
  4. 04Blind review: the reviewer knows neither location nor expectation. Cross-check against historical maps and aerial images.
  5. 05The result is a short, reasoned list with a recommendation for field inspection – never labelled a “find”.

03Outcome

The method was presented at the CAA Joint Chapter Meeting 2026 and is available as a service for heritage authorities and research. We deliberately do not show locations.

Data: LGL Baden-Württemberg, dl-de/by-2.0, modified. Without location details.

  • LiDAR / DGM
  • PDAL
  • Python
  • KI-Sichtung
  • Blindprüfung
  • GeoTIFF
View

A similar task on your desk?

In the intro call we tell you how we would approach it – and how quickly you would see results.