Land cover classification using sentinel 1 GRD data

Continuing the discussion from GRD product-land cover classification:

I am trying to create a land cover classification of a hilly terrain using sentinel 1 GRD data. The preprocessing steps that I have used are

  1. Apply orbit file
  2. Calibration
  3. Radiometric calibration
  4. Speckle filtering
  5. Terrain correction

I used Random forest classification by creating vector files but I didn’t get the accurate map. I also tried by adding VV- VH, VV/ VH but didn’t get the correct classification.

Can anyone please help me?

(I’m trying to create a map having classes forest, agriculture, urban, barren land and water)

one input image is not sufficient for a random forest classifier, because it needs a large variety of different input features (bands). We had some discussions on this here::

Calculating image textures is a good way to increase the input features. Also using two images of different seasons often brings more variation.

Have you seen this approach: Landcover classification with Sentinel-1 GRD

I used glcm textures of the preprocessed image. Then also I didn’t get the correct result.
The main problem occuring in the map is that the forest area is classified as agriculture.

Can you tell me whether the steps I used for the classification is correct or not?
I used only one GRD image. Is it necessary to take two images?

the preprocessing is alright, but about the selection of two images: Crop pixel are likely to be different at two times of the year while forest pixels are comparably constant. Using images from two or three dates (various seasons at best) will give you better results in terms of class separabilit than just increasing the feature space by textures.

Ok. I got the concept.
There are some more problems. Since I am classifying a hilly terrain, there are mountain shadows in the image. These shadows are classifying as water in both sentinel 1 and sentinel 2 images.
How can I rectify this?

if the topography is not too steep, you might consider radiometric terrain flattening Radiometric & Geometric Correction Workflow

Yes, it’s very steep. I’ll try terrain flattening method.