Sentinel-1 SNAP toolbox image view does not export as Geotiff with spatial reference

yes, this is normal, because they originate from different sources. We discussed this here: GLCM Range Value for the SAR texture analysis

Glad to know it’s normal.
So, technically PCA should work. How much should I set maximum count to? I didnot change the default -1 for the first time and changed it to 4 for second time, the error was same in both cases.

Also, can I interpret anything from the glcm histogram apart from distribution? I suppose it’s got no unit since it’s a statistical measure between grey level pixels.

How in snap can I derive more information on it ? (any supporting information apart from the professor of university of Calgary would be highly appreciated).

-1 means that you compute as much as components as possible, but often the first 5 components are sufficient because anything after that is noise. What error message did you get for the PCA?

yes, bright pixels mean high contrast, energy etc. and dark pixels represent low textures. They are unitless or as defined by their inventors.

Textures represent image patterns at different levels. I could not give any better explanation besides the 74 page summary of Mryka Hall-Beyer (you named it).

Please correct me if my impression is false, but it seems to me that you are not quite sure yourself what exactly you want from the data and how you can achieve it. Maybe you can first share your overall aim and then we can discuss how to approach it in the most suitable way.

1 Like

Thanks, Andreas for answering each point.
I want to do classification - on a dynamic region. Basically looking for transition between seasons of the surface.
For which, textures + Masking (based on DB ranges) - as an i/p to Random Forest seems convincing in my area (this is again mentioned in ur time series tutorial), so am trying to attempt a similar one.

Also, out of topic ques, is it that all the c-band sensors behave similarly? (in terms of intensity ranges - radarsat, sentinel, risat?) Because, I don’t have in-situ data, so if I take a variable range in radarsat - would that apply to S1 as well ? like if water ranges below -17.5/20 DB in radarsat, will that be the same in S1 too?
I want a multi-temporal classified image (which can capture the dynamics of the surface) - I hope am clear.

Thanks for your time

Sorry, I’m still not quite clear. Do you have images of multiple dates and you want to classify these images separately? If so, working with textures can be challenging, because these textures are quite sensitive to look direction, degree of speckle and incidence angle.
Or do you want to combine images of different dates in one stack and then analyze this stack for dynamics? In that case, PCA is more effective than textures.

C-band data is comparable once it is calibrated to Sigma0, because after that, the impact of spatial resolution on backscatter intensity is reduced. But it is always a good idea to compare how well this worked.

Images of multiple dates & want to classify them. I need seasonal behaviour- & I thought stacking them is a better idea because I can use a time-series graph. But again, it just takes point-wise data.

So, combine images of different dates in one stack and then analyze this stack for dynamics - is what I want. Am looking for methods.
So, Texture works better for single images & PCA for stack? (did I get that right?)

so I would first coregister all images into a stack, then terrain correct the stack and then create RGBs, apply PCA or use the mask manager to extract the dynamics you are interested in.

1 Like

Thank you so much.

Hello Andreas,
In continuation to the above info, I just need to ask a few things. I have created a stack and am trying to generate a classified image based on intensity ranges (like the one u did in the Time-series tutorial).
I just wanna ask if there is a way, I can extract all the pixels within a range in the entire stack and assign a class to it? Is it possible using band math?

Also, What exactly do I get out of PCA? I just know that it would give around 3 to 4 bands - but what does this mean? Can you please share any material?

Also, in one of your tutorials on Land cover classification, you mentioned rule-based classification - is it possible in ESA’s SNAP (if yes, how?) Page 15 - it is mentioned that “systematic thresholding is applied to the four bands in order to separate the four target classes in a hierarchical way” - what does this mean?

You also mentioned applying Textures which adds as an additional i/p to RF. Can you please explain how it works- as I assume textures don’t work on a stacked product.

Sorry for the long message, but I’m a little confused with these techniques.

Sorry, It’s not clear to me. You mean you want to apply the same threshold to all bands in the stack?

PCA stands for principal component analysis. It shows you the variation throughout a multitude of bands but reduced to less bands. This introduction is brief and clear: Principal Components Analysis []

Currently, the rule-based classification is not directly implemented in SNAP, but you can combine rules with the mask manager to systematically identify pixels which fulfill several criteria.

Applying textures to a full stack makes little sense because many of the bands are rather redundant. You can select one band with high meaning (e.g. average VV) in the module so the textures will only be computed on this band. The result is a new product containing the textures. Sometimes these are rather similar, so you can inspect them andcopy the bands into the multi-temporal stack with the Band Maths (because they have the same dimensions).

1 Like

Yes, the same threshold to all bands in the stack - as in everything above -20db to be OW in all the bands of the stack. But I think you answered it here

I wanted to give the rules for the entire stack at a time.
I initially thought I can copy the masks from one band to another (which might copy only range information), but it highlights the mask of band 1 on band 2 when imported and not the range.
So, am having to manually define ranges every time. Hence, I thought there would be the quickest way to do this.

Regarding this, my data has got quite a dynamic seasonality, Andreas - And I assume that averaging the bands would lose the information I need. But, will sure give it a try.

Gracias for the great assistance.

you can create one mask which adresses all like this.
band1 > -20 AND band2 > -20 AND band3 > -20...

If your data has seasonal dynamics, you can create textures from the images which differ the most, e.g. one from the dry season and one from the rainy season, so you cover both extremes with all textures. Anything in between is rather too much.

1 Like

Great will test and get back.

Thanks a ton