I am trying to detect some sea level rise/fall for a specific region, from Landsat imagery.

I went through the recommended methods of identifying water/land (MNDWI, AWEI, Bandratio 5/2, etc.).

Now I am at the point where I need an accurate vectorized coastline. I tried with Geomatica Lineament extractor, but somehow I wasn't happy with the results.I have Geomatica 2013, ENVI 5.3 and ArcGIS 10.2.


I was going to suggest you convert the classification map to a vector map which then can be used to identify the coastline. But then I realized that what you are trying to do eventually is change detection. If you are indeed interest in identifying the area of change between multiple images, you can do that without extracting the coastline. Just overlay the classification maps and using Band Math in ENVI, you can assign different values to different change types (ie, unchanged, increased, decreased, etc).

  • Wouldn't you have to take tide levels into consideration for that kind of analysis? – David Mar 2 '16 at 22:40
  • @David I agree. What i said was a general thought. It needs to be adjusted based on the actual conditions of the research (eg, data availability, time of the images, etc) – TonyC Mar 2 '16 at 22:48
  • Well, in my thesis, I am actually trying to correlate the tidal activity/sea level movement with groundwater levels through remote sensing techniques. The thing is, the satellite passes through my area at the same time of day every time, mostly at high tide. So the coastline itself would be useful for measuring the distance from a reference line and then correlate that with groundwater levels. That's just my idea, I'm open to suggestions, though – Alin Radu Mar 3 '16 at 9:59
  • @TonyC Actually, the first suggestion is exactly what I want to do. My question was exactly: how do I convert the classification map to a vector map? So far, whatever I tried (Geomatica Lineament extraction), gave me differences of 40m or more, which is impossible to happen in 1 year. – Alin Radu Mar 10 '16 at 13:20
  • @AlinRadu Try convert the raster to polylines in ArcGIS (you can either do it directly, but it may give you a lot of small line segments, or you can conver the raster to simplified polygons first, then convert the resultant polygons to polylines), and then delete all unwanted line segments. – TonyC Mar 10 '16 at 16:28

my guess is the (in)accuracy you are seeing is due to the spatial resolution of the landsat imagery ~40m.. The vectors will always be created within about one pixel of error. In your case, just one pixel of error/resampling for Landsat is several tens-of meters.

The coastlines can be extracted from Landsat imagery nicely but i would not expect higher spatial accuracy then the sensor used. thanks

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.