Gullies

Gully detection based on LiDAR and terrain derivatives with Deep Learning

Exploring existing gully inventories in the Gisborne area in New Zealand to develop a gully detection approach based on LiDAR and terrain derivatives.

Summary of work so far:

  1. Reference data needed some correction regarding co-registration, the approach followed is documented here. The sample preparation for deep learning is documented here.

  2. A calculation of terrain derivatives was prepared based on the LiDAR DEM data provided. The approach is documented here and to developed this, an R package terrain was created to ease the workflow.

  3. A summary of the exploration of the reference data and initial tests with deep learning approaches in ArcGIS are documented here and here.

  4. Further experimentation with ArcGIS Pro deep learning tools led to the insights documented here.

Acknowledgements

This research work is supported by the New Zealand Ministry of Business, Innovation and Employment research program “Smarter Targeting of Erosion Control (STEC)” (Contract C09X1804).