Question about area of analysis

[Environment Arabia]

Hello, 

I am wondering about how big the area of analysis is on any single data point within an image. 

For example, lets say that a point lands close to a coral but quite on top of the coral. What is the area of that point that is used for analysis? I am worried about potentially training the software wrong because I'm not sure how much leeway there is in terms of point accuracy.

Another example could be a large Porites with a very small dead patch which a point lands on. For training purposes, should I classify that point as dead coral with turf algae, or as the Porites?

Essentially, how much of a "bigger picture" does the software account for when using training data?

I hope I have been able to explain this.

Thanks, 

Gabriele

[Stephen Chan]

Hi Gabriele,

The automated classifier's assumption is that you are annotating what is at the center of a point. So, even if the point lands on a small dead patch or algae patch which is surrounded by healthy coral, you would still classify the point as dead/algae.

Of course, the same principle can apply in reverse: if you have a healthy Porites which is just barely visible in the photo because there is something else in the way, and your point lands on that tiny unobstructed part of the Porites, then you classify the point as Porites.

For more insight on this topic, see David Kriegman's message in this thread from a little while back:

https://groups.google.com/g/coralnet-users/c/7yOQM3QhFzU/m/rpXnDH83BAAJ