NISAR Level-2 GCOV SAR Analysis (VVVV + HVHV)

This project processes NISAR Level-2 Geocoded Polarimetric Covariance (GCOV) data to detect water and vegetation using SAR backscatter, filtering, thresholding, and K-Means clustering.

1. Nisar_Level2_GCOV.ipynb for VVVV polarization analysis(Done in QGIS):

VVV.tiff

2. Nisar_Level2_GCOV_img2.ipynb for HVHV polarization analysis(Done in QGIS):

Overview

The repository contains two notebooks:

• Nisar_Level2_GCOV.ipynb for VVVV polarization analysis.
• Nisar_Level2_GCOV_img2.ipynb for HVHV polarization analysis.

VVVV is used for water-sensitive analysis, while HVHV is used for vegetation-sensitive analysis.

Features

• Loads multi-band SAR GeoTIFF data.
• Normalizes and smooths backscatter using Gaussian filtering.
• Detects water regions using Otsu thresholding on VVVV.
• Detects vegetation regions using Otsu thresholding on HVHV.
• Applies morphological cleanup to binary masks.
• Uses K-Means clustering to classify surface types from backscatter intensity.
• Visualizes SAR intensity, masks, and cluster-based land-cover categories.

Workflow

VVVV Processing

The first notebook processes the VVVV band from the GCOV product.

Main steps:

• Load the VVVV band from the GeoTIFF.
• Replace invalid values and normalize intensity to 0–1.
• Apply Gaussian smoothing to reduce speckle noise.
• Use log transformation and Otsu thresholding to extract water-like regions.
• Clean the mask using opening and closing operations.
• Apply K-Means clustering with 3 classes to separate dark, medium, and bright surfaces.

HVHV Processing

The second notebook processes the HVHV band for vegetation analysis.

Main steps:

• Load the HVHV band from the GeoTIFF.
• Normalize and smooth the data.
• Use log transformation and Otsu thresholding to identify vegetation.
• Refine the mask with morphological operations.
• Overlay vegetation on the SAR image.
• Apply K-Means clustering with 3 classes to represent vegetation density.

Results

VVVV Water Detection Overlay

This image shows the processed VVVV SAR scene over a basemap. The bright white regions indicate strong low-backscatter or water-dominant areas in the SAR footprint.

Requirements

Install the required Python packages:

pip install rasterio numpy matplotlib scipy scikit-image scikit-learn

How to Run

1. Open Nisar_Level2_GCOV.ipynb.
2. Set the correct path to the VVVV GeoTIFF file.
3. Run all cells to generate the water mask and clustering output.
4. Open Nisar_Level2_GCOV_img2.ipynb.
5. Set the correct path to the HVHV GeoTIFF file.
6. Run all cells to generate the vegetation mask and clustering output.

Notes

• The helper function process_sar_tiff() loads a selected band, handles NaN values, normalizes the data, and applies Gaussian smoothing.
• Otsu thresholding is used for automatic separation of classes based on intensity distribution.
• K-Means clustering is used as an unsupervised method to group pixels into meaningful surface categories.

Output Interpretation

• VVVV low-backscatter regions: likely water or smooth surfaces.
• VVVV bright regions: stronger scattering, often associated with rough or built-up areas.
• HVHV high-response regions: likely dense vegetation.
• HVHV low-response regions: sparse vegetation, bare land, or water.