2

I am working on my master thesis on multi temporal classification of LULC using S2 imagery in Google Earth Engine.

The goal: to distinguish urban area from the landscape in the Inner Niger Delta. This cannot be done using supervised classification as the houses are build of the same material as the surroundings, so the profile spectra are very similar. So my plan is to do this based on time series, where I assume that urban areas are very stable in terms of the NDVI profile and the surroundings fluctuate more. I would have to be able to have NDVI time serie info of every pixel and them cluster the similar ones together.

The problem: I know how to extract the time series with NDVI information (script:https://code.earthengine.google.com/9069f8232d61a5a51546d28a92b27110), and I know how to apply clustering for one (LS8) image (script:https://code.earthengine.google.com/915d919c2adef5a15a140c1a0d5d328d). But I have not managed to combine these two. So to do a clustering based on the NDVI time series PER PIXEL.

Is it possible to extract the (mean) time-serie chart, of e.g. NDVI, per cluster so I can determine the class?

Improve this question
5
  • 1
    I don’t have an answerer for you, more of an alternative that could be worth trying, if you haven’t already. The mean/stdDev/max NDVI for a whole year should be able to capture that difference in urban vs. other areas. Try to include these band in your supervised classification. code.earthengine.google.com/0d3b59965150645021207af1dc792f35
    – Daniel Wiell
    Commented Feb 8, 2020 at 7:25
  • 1
    The Earth Engine developers group might be a good fit for this type of question - opinion/suggestion for methodology to achieve a goal.
    – Justin Braaten
    Commented Feb 10, 2020 at 23:15
  • You say that you have not managed to combine two lots of code so can you include your attempt to do that as formatted text within your question body, please? That would see me voting to re-open your question.
    – PolyGeo ♦
    Commented Feb 11, 2020 at 20:42
  • @PolyGeo, I would love to but I really dont know where to start.. thats the problem mainly
    – Sanneeeliza
    Commented Feb 12, 2020 at 15:59
  • @Daniel and JustinBraaten have made suggestions as to where you might look for some ideas or starting code. Once you’ve made a code attempt your question is likely to attract re-open votes. We help people with where they are stuck on their code but we try not to write their code for them so as not to be seen as a code writing service.
    – PolyGeo ♦
    Commented Feb 12, 2020 at 19:11

1 Answer 1

Reset to default
0

it might be a bit late, but still I got an answer to you question.

What you need to do is convert the image collection with the NDVI time series to an unique image with as many bands as dates you have in the image collection by using the "toBands()" function.

Once there you can perform either a supervised or unsupervised classification

Improve this answer

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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