Photogrammetric DTM for tropical forests

Hi there! I hope everyone is keeping well:)

I am Ivan, a university student working on a project in which we aim to use a UAV to estimate Above ground mass (AGB) in tropical forests.

Our research has informed us that the following data pipeline is needed:

  1. Obtain DSM and DTM, either from orthophotos, or from a lidar point cloud
  2. Obtain a CHM (canopy height model) from the DSM and DTM
  3. Perform segmentation on the CHM to delineate individual tree crowns
  4. Obtain individual tree metrics (such as crown diameter, and height) from delineated tree crowns:
  5. Apply allometric equations to estimate AGB

We have found that DSM can easily be generated from orthophotos, using ODM. A photogrammetric-DTM (photo-DTM) can also be generated, but non-ground objects, such as trees, need to be filtered out first.

In order for the photo-DTM to be accurate, both “ground” and “non-ground” points must be readily available. Hence accurate photo-DTMs can be generated from areas with plenty of visibility of the ground.

However, in tropical forests, due to dense canopy cover and little visibility of the ground, the resultant photo-DTM generated will likely be inaccurate (with a large RMSE). The use of GCPs may help to generate better photo-DTMs in this situation, as per this source.

A sample image of the tropical forest we aim to estimate AGB for:

I would like to ask if anyone has found a workaround to generate fairly accurate photo-DTMs for tropical areas, using ODM? Did the use of GCPs help?

If so, and if I have correctly identified ODM’s DTM/DSM feature as the most appropriate output setting, which custom settings (e.g. smrf related) would you recommend?

1 Like

Hey @Ivantan :hand: use --pc-rectify, it sounds like it might work well for your case. See the description of the method here:

1 Like

Thanks @pierotofy! This is promising :slight_smile: Will check it out and update as I go along!