Mmm...I doubt there is a simple approach. I am working with a similar issue but on terrestrial vegetation. As far as you take nadir photographs with a high overlap among them, orthorectifying them should not be too difficult. It's much more difficult to classify the resulting image. Even more in your case, where the spectral signature of the target organisms is masked by water (lower signal to noise ration, higher variability of signal associated to light conditions and picture angle...).

In case it is useful, my workflow is more or less like this:

* Set a minimum of 4 targets in  the area to be covered, around the edges of the pictures. You must take accurate GPS position for them. 4 is the minimum, but I would not recommend less than 10. If you don't need too much precision, you can just take control points from landmarks and take the coordinates from a map or from Google Maps (2-3 m error). Then you don't need the targets.

* Take the pictures on several parallel passes, ensuring an overlap of around 70% between consecutive passes and 30% between consecutive pictures.

* Analyze the pictures with classical photogrammetric techniques or with Sfm techniques. I use AirPhotoSE, which is free. There are other newer pieces of software, but AirPhotoSe allow doing all the work. The task is easier if you have a proprietary software like EnsoMosaic (they are expensiive, though). You will get an orthomosaic.

* Georreference the orthomosaic by using the control points.

* Perform a supervised classification. You need to know the location of several patches of each organism, in order to tell the algorithm.

* Measure the precission...if not enough, improve your classification.

Best regards,

            Asier

Thank you Asier for the layout suggestion... I will consider the approach 

I have never done that, but I am aware of research done to monitor coral reef size/area in this manner. From the air you will probably not be capable of identifying the benthic organisms properly, unless diversity is low (unlikely in a tropical ecosystem) and water transparency is maximum (and reflection, minimal). What you can do (and is usually that way with landscape ecology) is calibrate your aerial photos through in situ work/checking. In this manner you may be able to identify multispecies/dominant species beds from your photos, and therefore monitor their size (again, if you can determine the exact size from some reference point in land/sea, let's say, a coastal house or a boat you can easily check the size personally. The approach mentioned by Asier is well explained and valuable! 

Hola Emmanuel, I have done extensive work in this area (we made an atlas for the entire coastline of Mauritius; mapped seagrass in Florida; intertidal vegetation in both Canadian coasts, and other projects), BUT we used our compact airborne spectrographic imager (CASI); yet, some of the methods might be useful to you.

Asier's points are good, but we found that using unsupervised classification provided better results. Normally, unless you know with some degree of certainty the composition of your study area, or you cannot account for its diversity (and have all the required end members). Having said that, if what you want is to identify certain "classes", the rule image generated from the supervised (we use mostly S.A.M.) gives you nice spatial probabilities.

I advocate for unsupervised, creating a lot of output classes, and then use clear criteria for manual groupings.

I can send you some papers if you want; let me know! and best of luck.

Would you like to email me? so we can discuss your approach, steps and goals. I think I can help you.

[email protected]

@Mar Martinez, I will get in touch thanks