In a previous experience, working in a project about land/use dynamics, we use Landsat 8 images and using the 6 bands of visible and infrared (from 2 to 7 bands), we made a Principal Components Analysis to performance a supervised classification by the method of maximum likelihood, using the first three principal components, the NDVI, and the digital terrain model… of course, also using an adequate sample of training polygons over real settlement areas… we obtained very good results in the identification of urban areas.

You could try it; I hope it is useful for you

Good luck… best regards

René

Hi M. Alvioli

thanks a lot for your attention.

Hi rene

Thank you for your suggestion. I already tried this method but the result was not very accurate.

you first try unsupervised classification by choosing a big number of classes (around 100), then you merge to the most significant. second you investigate the urban area class if the accuracy is not acceptable then you have to turn to GIS and use semiautomatic methods by manual digitizing the missing / deleting the non urban areas. then you convert and integrate the whole into the already created urban class.

Hello Salem mohammad Ghaleb Issa sir,

Thanks for your help.

You can also try a soft classification approach but you must note that what you call a settlement area could be a mixture of different land use/cover endmembers. Good luck @Tek

Exact settlement area? Do you mean that you'd like 100% classification accuracy or do you want something highly accurate?

If you are talking about impervious surface, then you should be able to capture with better training samples. But when you are saying settlement area, it is a whole different story. Settlement is not exactly land cover rather it is land use so it will have mix of vegetation and buildings and so many other things. Specially in Nepalese context, when houses are thatched, corrugated and concrete side by side with invisible roads (because of dirt roads and/or covered by trees). Remote sensing of just onetime scene could hardly capture what you want. I would suggest to using time series (especially winter and summer, if you can find cloud free add all 4 seasons) and add more index in your input lists, like MSAVI, IDWI and principle component etc. But no matter what you may end up doing some manual editing. Good luck.

This is so true @Devendra. Settlement boundary is a politico-administrative approach which will deviate from any form of objective assessment, except the actual ground area as well as (prominent) property boundaries are surveyed. ....#manual.

follow this band combinations