IMAGINE Discussions

Discuss and share topics of interest using ERDAS IMAGINE the world’s leading geospatial data authoring system.
Showing results for 
Search instead for 
Do you mean 
Reply
Highlighted
Contributor
Posts: 73
Registered: ‎01-04-2017
Accepted Solution

oil Spill Mapping with Sentinel-1 in ERDAS Imagine - Procedure

Hello,

 

I am processing the Sentinel-1 data for oil spill mapping in SNAP, the area is about 100x100km in sea. 

 

The procedure involves speckle filtering, converting bands to decebal scale, running Oil spill algroithm which involves Land-sea masking, converting pixel values to caliberated values and clustring, Projecting the image to a coordinate system. The oil spill detection background window size, I kept, is kept about 1400 and the Threshold shift (dB) is about 4.

 

All the above procedure takes a minimum of 14 hours.

 

I would like to perform Oil Spill Mapping in ERDAS.
I would like to know if there is any algroithm or Indecies to map the oil spill mapping in ocean in ERDAS ? and
What are the step by step procedure to Map the Oill spill in ERDAS with Sentinel-1 data ?

 

Best Regards,

RSGIS

Highlighted
Technical Evangelist
Posts: 879
Registered: ‎10-01-2015

Re: oil Spill Mapping with Sentinel-1 in ERDAS Imagine - Procedure

Hi RSGIS,

 

Here's some pointers from our SAR experts:

 

First, to be clear, this approach is not actually “mapping oil spills”. The phenomenology is that a film of oil will dampen the microrelief of the water surface making it smoother. This results in more specular reflection, radar signal reflected away from the sensor, and thus a darker return. To actually map oil would require multispectral or hyperspectral data.

 

Despeckling as a first step makes sense for many operations as the erratic distribution of bright and dark pixels complicates a processes such as this (or flood mapping) that are mapping light vs dark pixels. There are a variety of speckle reduction filters in ERDAS IMAGINE (Radar-Utilities) and most any would work fine for this purpose; even a simple Median would be good. I might use a 5x5 window and follow with a 7x7 or 9x9 Local Regions (clustering) filter. Despeckling a typical Sentinel-1 swath took five minutes on my laptop.

 

It's not clear why you would need to convert to Decibels. This is normally done when one is working toward absolute radiometric analysis, such as soil moisture quantification or merging datasets from different sensors or dates. That is not the case here; we are going to separate light vs dark pixels based on the relative values in the dataset being analyzed. I would tend to leave the data in linear mode so that the histogram is better distributed. However, if desired, is it easy to convert the despeckled Sentinel-1 image to Decibels using the Radar Conversions option. Takes about 1 minute.

 

Automatic Land-Sea masking is difficult as scenes vary so much; lakes, roads and runways are as dark as seawater, maybe darker depending on wind conditions. We are currently working on Deep Learning approaches to this, but it is not perfected or released. In reality, manually delineating the coastline is pretty quick and operationally the coastline should already be available to mask the data.

 

At this point I would display my Despeckled image in the 2D View and select the Radar Analyst (Raster-> Radar Toolbox-> Radar Analyst). The Display mode can then be changed to Level Slice. The histogram of the image can be optimally displayed using the Rescale Limits and then the carets on the X-axis used to separate light from dark (oil damped) areas. This same Tool is used in a similar fashion for Flood Mapping.

 

If you wanted to develop a fully automated algorithm, the IMAGINE SAR Feature extraction module has two flexible Constant False Alarm Rate (CFAR) algorithms that could be used to more exactly replicate the SNAP approach. Exact processing parameters would depend on the dataset and the CFAR used. Also note that IMAGINE SAR Feature allows a very flexible and powerful Despeckling operation to be inserted into the CFAR processing chain. So this would be the “automated” approach.

 

There is no need to project to a coordinate system. In ERDAS IMAGINE, radar scenes are automatically and precisely georeferenced. But ERDAS IMAGINE does allow conversion to any desired projection.

 

This procedure was developed for Flood Mapping, where the results are needed in minutes, not hours. 

 

Hope that helps.

 

Cheers

 

 

 

 

Ian Anderson
Chief Product Owner, Desktop Remote Sensing
Hexagon Geospatial
Highlighted
Contributor
Posts: 73
Registered: ‎01-04-2017

Re: oil Spill Mapping with Sentinel-1 in ERDAS Imagine - Procedure

Thanks Ian,

 

Your reply, rather a good article, gave me a great inshight of "Oil Spill" science besides ERDAS procedures.

I will follow the procedures and compare the results and get back to you if I have more questions.

 

Thanks and Best Regards,

 

RSGIS