Hi…how can i detect wet and dry snow using sentinel 1 data using snap?
Hi,
I assume that your research area is over land and not sea ice. If that’s the case then you can use two different methods to detect snow:
- Based on backscatter intensity of snow
- InSAR and coherence analysis
In snow free areas we would expect a higher backscatter intensity compared to areas covered in snow (interaction of microwaves with snow is complicated and can produce higher backscatter signal in some cases)
You can also derive a coherence map where snow free areas will have higher coherence due to more stable scatterers.
For more details, look at this paper
1 Like
Hi,johngan.Could you please give me some examples about this?This happened in my current study, but I don’t know how to explain it .Thank you!
Hi, sorry for the late reply, i was on holidays
The interaction of microwaves is very very complex and there is a lot of depth in it. So, i will just give you short summary of how microwaves behaves in dry and wet snow.
When it comes to snow or ice, the interaction of microwaves is heavily depend on the effective dielectric constant of the medium. The dielectric constant determines the propagation and absorption of microwaves through the snow. High dielectric constant mean little or no propagation at all (specular reflection), low dielectric constant means high penetration (diffuse reflection).
The interaction of the electromagnetic wave with snow is a result of
- geometrical structure of the snow (e.g crystal structure within the snow)
- the electromagnetic properties of its components, air, ice, water vapor, and-when the snow is wet-liquid water.
The snow is not just vacuum, but instead consists of crystals, ice, dust particles (and water liquid when snow gets wet). So, The microwaves interact with the grains and ice within the snow. When the wavelength is sufficiently larger than the grain size of the snow, the backscatter increases, on the other hand, if the wavelength is significanlty larger, then the backscatter decreases. As you can see, this is wavelenght dependent. If using L-band to do ice/snow observation, due to its long wavelength, the backscatter intensity will not be as high as the one of X-band. Using high frequency microwaves, we expect a much higher backscatter on a dry snow
On the other hand, wet snow, due to liquid water, microwaves do not penetrate, hence, the backscatter intensity decreases. The higher the microwave frequency, the less the penetration to a wet snow.
2 Likes
Thanks for your reply,it is helpful!And I have two more questions.
- If I have a snow-free image and a wet snow image,is it possible that the backscattering coefficient(DB) of wet snow is higher than that of snow-free?
- When I try to extract wet snow using Nagler’s method(σwet snow/σsnow-free，threshold：-3db),I get the strange result as follow.Is anything wrong?
Excuse my pool English,thank you again!
In terms of your first question, the backscatter signal between snow free and snowy areas depends on the terrain of your AOI. If there is terrain in snow free areas then the backscatter will be much higher compared to areas with wet snow. If your AOI is relatively flat, then the intensity of the image acquired over wet snow between the the image acquired overt flat area can be very similar and hard to discriminate.
I do not really know what methodology you have followed to extract wet snow. I can give you my advice on that (which might not be the most appropriate in your case, you need to do bit more research on your own)
- first of all you need to determine the backscatter of the snow free area collecting a time series of data.
- Collect data over the snow area.
- You might need to coregister the SAR data collected
- Use the ratio between the image acquired over snow and the image acquired over snow-free area
5 determine an appropriate threshold
This is roughly the methodology i would follow.
3 Likes
Thank you very much!
Exactly,I did the 1-4 steps that your mentioned above and I got the data (in my last reply) that the range of values is large(-487~102).About the step 5,i really don’t know how to determine an appropriate threshold.Could you please give me some advice about this?
In terms of selecting an appropriate threshold, you have two options
-
Choose a threshold based on image inspection. In your case, i suspect that the values for wet snow (this is just my guess) will be range between (-22 and -15db, roughly), while values in snow free areas might range between (-7 and 2db, depending on the terrain). Hence, when we get the ratio between a big negative number and a small negative number, this gives us a large positive number. So, you can have a look at the histogram of your image and select large positive numbers assuming that this is the wet snow. Of course this is not very accurate
-
A more sophisticated thresholding algorithm can be applied. You can apply a Thresholding Method Based on Standard Deviation. Have a look at the following papers:
4 Likes
Thank you!!I just want to know the approximate snow cover for the moment,but i will look at the papers.
- To extract snow,is using vv band more efficient than vh band?
- According to you,in flat area ,it is difficult to distinguish snow areas by backscatter signal, while in topographic areas,it can be identified by reducing backscattering of wet snow.If my image includes both the flat region and the high mountain region,I need to think about the two cases separately, right?
I’m afraid I’m not making it clear, just to make it easier to understand.My image just as follow:
DEM
VH band
Thank you again!
-
In order to extract snow from SAR image, i would say VV polarization might be slightly better than using cross-polarization. This is because, snow (especially when it’s wet) tends NOT TO depolarize the signal (when a surface depolarize the signal, it means it change the polarization, lets say from V to H), hence we have higher return in co-polarization (VV). On the other hand, using VH, cross-polarized signal might reach the noise floor of Sentinel-1. I would say, a combination of both polarizations might help you discrimination the snow
-
In flat areas where there might be areas covered in snow (especially if it is dry snow), then, using VH polarization you might see some difference between dry snow and flat area, This is because due to the internal structure of the dry snow (e.g crystlas, dust, particals) this might depolarize the microwave signal (from VV to VH) hence we have higher return in VH over snow compared to flat areas (but this is not the case for wet snow though).
So, you might need to apply two different threshold values. One threshold value to discriminate flat areas from snow (looking at VH polarization) and a second threshold value to discriminate terrain from snow (using VV polarization or a combination of both)
4 Likes
Thank you very much!It is very helpful to me 
1 Like
hi,johngan!
Are there any good algorithms in snap that can be used directly to extract dry snow/wet snow (I’m not sure if they need to be extracted separately using different algorithms or can be identified simultaneously using one algorithm)?I have S1 SLC data now.
Thanks.
Hi,
SNAP does not have such an algorithm where you can automatically extract wet and dry snow. It is not an easy task where SNAP developers can develop an automated algorithm for that. It requires a lot of things such as: 1) knowing the area 2) a combination of inputs and 3)visual inspection.
As I see it, you can try three different methodologies to extract the snow of your AOI.
- Using the ratio of the two SAR polarizations as we discussed above
- Use the coherence layer derived from SLC data. Coherece over snow-free areas are going be high, but coherence over snow drops. Hence, you can set a threshold to extract areas of snow
- You can use a classification algorithm. This requires you to have some ground truth data over snow. Hence, you can use these ground truth as training sets and classify your area of interest
You can compare the results of those three methodologies and see how much agreement there is amongst them.
3 Likes
There are a lot of classification algorithms, are there any better methods of classification?Does the unsupervised classification approach work poorly (like h/alpha wishart)?
Supervised classification works better only if you have ground truth data. To collect training dataset for each class, you must know your area of interest well.
Unsupervised classification can be used in case you do not have training data. The algorithm creates cluster of pixels based on their intensity value.
You can try unsupervised classification and see what you get. If you are not happy with the result, then you should follow a supervised classification with known ground truth. There is not right and wrong way. If you have no ground truth and you do not know where wet-dry snow is located, then you might find unsupervised classification more useful.
2 Likes
Thank you!
And if I use a landsat8 image(1% cloud) to verify the accuracy of SAR snow extraction, what are the general pre-processing steps and snow cover extraction algorithms for landsat8?
there are Fractional Snow Covered Area products for North America.
Furthermore, snow products are provided by THEIA.
As an alternative, you can extract the snow areas with a supervised classification as described here: Rndom forest classification steps
1 Like