CE712 Course Project
ID: 12 Module 2
Sahil Vijay Bhoge(210040132)
Rowthu Vivek(210040128)
Topic: Differentiating between cloud,
ice and sand from multispectral imagery
Overview
In this project, we aimed to differentiate between clouds, ice, and sand using
multispectral imagery. We began by collecting relevant datasets from Earth
Explorer and importing necessary Python libraries. After downloading the
required spectral band images, we converted the Digital Number (DN) values into
NumPy arrays for efficient processing. We then calculated indices such as the
Normalized Difference Snow Index (NDSI), a Cloud Index, and a Sand Index to
highlight the unique spectral signatures of each surface type. By comparing the
results of these indices, we effectively distinguished between clouds, ice, and
sand in the imagery.
The process involved several key steps:
Data Acquisition:
For data acquisition, we used the USGS Earth Explorer platform to gather
satellite imagery, selecting images that covered diverse surface types like
deserts, polar areas, and cloud-prone regions. We specifically chose images
with spectral bands crucial for our analysis, including the Blue, Green, Near-
Infrared (NIR), and Short-Wave Infrared (SWIR) bands, as these are essential
for calculating indices to differentiate between clouds, ice, and sand.
Importing python Libraries and required Images:
For data preparation, we imported essential Python libraries, including
NumPy for numerical operations, rasterio and Matplotlib for visualizing
results. We then converted the Digital Number (DN) values from the satellite
images to reflectance values and required indices.
Earth Explorer was used to download satellite image needed for the
Project
Snow Identification
We used the Normalized Difference Snow Index (NDSI) to identify snow.
Snow typically has high NDSI values because it reflects strongly in the Green
band and less in the SWIR band.
We applied a threshold to the NDSI values to isolate pixels representing snow
and ice.
This method effectively highlighted snow-covered areas, clearly distinguishing
them from sand, and other surfaces as shown in next slide
Raw image: Masked Image:
Cloud identification
•We used the Cloud Index to identify clouds.
•Clouds generally have high Cloud Index values due to their strong reflectance in the Blue band and lower
reflectance in the SWIR and NIR band.
•We also used cloud reflectance properties in NIR band and compared the results with the previous method
•By setting a threshold on the Cloud Index values, we were able to distinguish cloud-covered pixels from
other surfaces.
•This approach effectively highlighted cloud areas, making it easy to separate them from snow, sand, and other
land features.
Original image used for cloud
identification:
Using NIR band:
Using SWIR band:
Sand identification
We used the Sand Index to identify sandy areas.
The Normalized Sand Index formula is:
Sand typically has high Sand Index values because it reflects more strongly in
the Red band compared to the SWIR band.
By applying a threshold to the Sand Index values, we isolated pixels that
represent sandy regions.
This method effectively highlighted sandy areas, allowing for clear
differentiation of clouds from snow, and other surface types.
Original Image: Processed Image:
Reference:
CE712 slides
Tutorial 4
https://www.usgs.gov/landsat-missions/normalized-difference-snow-index
https://www.researchgate.net/figure/Representative-spectral-reflectance-
of-snow-clouds-soil-and-vegetation-in-the-VIS-and_fig7_324383856
https://www.researchgate.net/figure/Spectral-reflectance-curves-of-snow-
soil-vegetation-deciduous-conifer-and-grass-and_fig3_276291826
https://www.mdpi.com/2072-4292/14/15/3802
