Transitioning from visual image interpretation to semi-automated segmentation methods for seagrass mapping

The Massachusetts Dept. of Environmental Protection is exploring the pros and cons of changing their eelgrass mapping methods from screen digitizing (visual interpretation) of aerial imagery to semi-automated methods (e.g., eCognition). Has anyone done comparative studies between the two methods or explored what this kind of impact this kind of methods change has on trend data? Any references or points of contact are much appreciated.

Hi Mark,

Kris’ paper touches on similar concepts:

  • Dave
1 Like

Thanks Dave (and Kris!)

Hi @mfinkbeiner, I compared mapping eelgrass using Heads-Up against image segmentation and supervised classification in ArcPro for my masters thesis last year, but using drone imagery. At that ultra high resolution, the two methods were apples and oranges in some regards. The machine tools produced extremely detailed maps, delineating individual patches that a photo-interpreter would have lumped together. The classifier did a pretty good job separating rocks from eelgrass from algae but it needed an excessive amount of training. All that said, the acreage calculations were not comparable between machine and Heads Up methods. However the extent was fairly consistent between the two. I could see a scenario where machine-derived maps help a photointerpreter along during Heads Up.

It’s possible some of the MMU issues might disappear with lower resolution imagery or by using different software, but I don’t think automation alone in is our future for eelgrass in temperate waters.

Figure shared as an example.

Hi Mark,
I don’t have any studies to share unfortunately, but I’m excited to see the interest in this topic. We started down this path with a segmentation of our 2004 heads-up digitized habitat map using 2010 imagery (was a smokey summer so not ideal imagery), see example below of those segments overlaid on our 2018 imagery. I’d really like to have a segmentation method for quickly identifying change each year using our 6" ortho photo mosaics. We are still acquiring the imagery, but are in need of assistance with the tools to segment (ENVI or eCognition software or even better - could ArcPro Image Analysis tools produce the product?)