How to determine appropriate threshold for flood detection (Sentinel 1)

Hi there,

I’m new to the sentinel toolbox application but not to remote sensing, so I apologize if this is basic - I searched the forum first to see if my question was answered.

I’m trying to detect flooding with Sentinel 1, and this is my first time with radar data. I’ve been following the guide here: and have gotten good results after a few software hiccups. That said, when it comes time to threshold the image to separate water from non-water, that guide is vague. It looks like they just choose the value right at the left edge of the normal distribution in the histogram, but that’s not clear.

My own histogram (above) looks quite different from the one in that guide, and choosing the value just to the left of the main part of the distribution doesn’t give results I think are valid. I’ve experimented with other values, and have found some that seem sensible, but the process of choosing a value felt too subjective. So I have a few questions, some specific to flooding and some to radar data:

  1. Is there an accepted, relatively objective method for the threshold that separates mostly water from mostly non-water using Sentinel 1 data?
  2. Can I compare the data to a prior date for the same location? I see the software has a change detection function (but I haven’t read up on it yet) - would the intensity values be comparable, where I could simply compare the resulting image to a prior date’s values and see which areas dropped significantly in intensity? I’ve been uncertain since it seems like the instrument can operate in different modes and may have different angles of capture.

Thank you for your help!


In general, you need to have a histogram of an area that is covered by land and water to around the same parts. So instead of calculating a histogram for the whole image digitize a rectangle that features the two parts equally.
In the histogram tool you can then select to calculate the histogram only for the digitized geometry.

to your first question: It is an accepted method to have the histogram that has two peaks in a numerical format, then make a mathematical graph/curve out of it and then search for the local minimum between the two peaks.


Thank you, that was very helpful - I used that method and it worked well. Thanks for taking the time to describe it!

Differentiate the absolutes of Sigma0 VH db between non flooded and flooded. (Terrain correction and conversion to UTM is a nice thing to have done) . <---- Processing!!!

The threshold is if done correctly is somewhat bigger then other values so between 5 and then depends what you are looking for.

What formula have you used to calculate the water threshold ? Using the histogram values .

A widely used method for image thresholding was provided by Otsu:

However, a global threshold is searched which often is insufficient in cases where land and water are not equally distributed in the image and the histogram doesn’t have a clear minimum. There are several approaches for local thresholding


can you send the correct link for "implementation in python (Otsu’s) ". The one you placed in this message is not opening.

1 Like

Hi Sir,
Can you please send me the detailed process of how to determine flood detection using Sentinel 1 images in Snaphu software.

Maybe these help you:


I wanted to carry out a standard thresholding procedure to segment the land the water bodies.
And I did try out the Otsu’s thresholding method, adopting the python program specified in the links provided here. But however, I am not sure how these scripts takes in the Sentinel 1 image.

If suppose i need to try out the code from the link “
Then, how and where should i include the image file.
(This is my first time working with python, kindly help me please)

Could sumone provide a possible way to find the local minima?

before you can use S1 image within python, you have to pre-process (calibrate, geocode) and export them as 8 bit integer. Then you can use it as an input as described here:
Afterwards you can convert it to a python array and process them as documented in the script you linked.


thank you lots… I have already pre-processed the S-1 image…
Will now try out as you have mentioned…

Is there any future plan to include an algorithm, such Otsu’s, to compute the threshold in SNAP ?