Accuracy of oil spill detection algorithm in SNAP

I am starting to work on a project which requires automating the process of oil spill detection using SAR images. Now that SNAP already contains these features, I would like to know how accurate these are, what algorithms and classifications they use and where can the improvements be made? Any resources which can help in automation are also appreciated.

Hi @mdrpanwar

First of all, check this RUS Copernicus webinar on oil spill mapping with S1 and SNAP. It may be a good starting point for your project.

About how accurate the process is, be aware that when working on oil spill mapping or detection there are phenomena known as look alikes that may be detected as oil spill but are just false positives. This is due to specific oceanic conditions that create dark areas which are recognize as oil by the algorithm (check the literature, you will find a large discussion on this)

About the algorithm in SNAP, in brief, it is a dynamic window of x pixels moving over the image and looking for dark areas according to a threshold defined by the user. The help option of the tool will give you a more detailed overview.

This is a simple approach while more advanced techniques involve classification (you would need training data for that - again, check literature)

Although possible, in my opinion complete automation of this type of analysis is not always a good idea (depending on your methodology) since supervision and checking intermediate steps is key to obtain good results.

An extra tip, which type of oil spill are you interested on? Oil spills from ships, from oil facilities… Their shape, extent, environmental impact, predictability is different and this can be an advantage or not when trying to map them

Hope it helps,