Use a vector of zeros to estimate shadow fractions using a mixture tuned match filter, and threshold them using a suitable threshold. However, I would recommend rather than classifying-out shadows, you could use a topographic normalization method such as the SCS-C by Soenen et al (2005) "SCS-C: A Modified sun-canopy-sensor topographic correction in forested terrain" IEEE Trans. GRS, 43(9) 2148-2158.

Use a vector of zeros to estimate shadow fractions using a mixture tuned match filter, and threshold them using a suitable threshold. However, I would recommend rather than classifying-out shadows, you could use a topographic normalization method such as the SCS-C by Soenen et al (2005) "SCS-C: A Modified sun-canopy-sensor topographic correction in forested terrain" IEEE Trans. GRS, 43(9) 2148-2158.

I would also go for topographic normalization, but beforehand some radiometric enhancement might improve the result, for example contrast stretching with histogram equalization. And generally, some mask bands are available with the images where such radiometric errors are expected, sometimes they help.

very good answer by aditya singh...... the shadow regions can be also be classified based on topographic normalization techniques. It depends on which region you want to apply this...... if you are doing this on a hilly terrain then go for cosine correction method......

For ETM+ images I would use Fmask to automatically detect the shaddows and clouds. The software is free from Boston University and can be downloaded here

http://code.google.com/p/fmask/

Dear Atri,

I developed a vegetation index (for Landsat ETM+) that I believe is relatively insensitive to terrain shadow at least in Australia where mountains are not of high altitude. If it is insensitive to terrain shadow in the Himalayas then it could possibly be used as the basis for change detection.

The index has not been used for images outside of Australia.

Paper:

Motivation, development and validation of a new spectral greenness index: A spectral dimension related to foliage projective cover ISPRS Journal of Photogrammetry and Remote Sensing, Volume 65, Issue 1, January 2010, Pages 26–41

You can find the paper here: http://dx.doi.org/10.1016/j.isprsjprs.2009.08.002

The paper provides the means to construct the index. It does not discuss terrain insensitivity and does not provide the details on how to convert from index to estimate of foliage cover.

Regards,

Trevor

Please see this link

www.geocarto.com.hk/cgi-bin/pages1/june05/33_saha.pdf

It may be good idea to extract reflectance value of the shadow area and normalise with respect to open area. Use the normalised image for classification

Well .. I will agree Trevor Moffiet, that you can use indices or may be band ratios that can can cut down the effects of shadow. I have used a set of indices and band ratios in conjunction with terrain related variables derived from DEM and then fitted a regression model (Generalized additive model) to map the forest vegetation in Palas valley (W Himalays, Pakistan). I got very encouraging results for my study.

I read all of the above answers, but if you didn't get the required accuracy or result, you can try to classify ratio images instead of a single band or colour composite of three bands. also you can try to classify the principal component image if possible. I think it will give reliable results. good luck

To over come the shadow area, you can use Digital elevation Model as a additional input, to classify LULC of area more accurately and its a suggestion. You can refer the attachment.

Hi, I have been working with Landsat ETM+ images apply to Andean regions of south America (with high mountains), and in my own experiences, I can suggest a couple of things. The first thing is to do what Khalid Elsayed Zeinelabdein says, at least i works really well for me, with the selection of an umbral over the first pribcipal component in combination 453. Also, I work with the NDSI index, that helps me a lot, and in other hand you can use another combination for space color, like HSV or maybe YCbCr that can help you not only with the shadows. Hope it helps to you, and any question just ask. Also recomend you to read my paper:

https://www.researchgate.net/publication/232421515_Land_cover_differentiation_using_digital_image_processing_of_Landsat-7_ETM_samples_of_Cauca-Colombia

Conference Paper Land cover differentiation using digital image processing of...

The most simple approach is to classify the shadowed areas as separate classes, equivalent to the non-shadowed classes, and group them only in the final output product.

Thank you all for all of your suggestions.....I have successfully overcome the shadow menace and have done the classification perfectly to some extent ....Now the ground verification will say how much accurate I were to delineate the various classes.....Thank you all and researchgate...

Atri to do it really correctly....conceptually, the best way to classify shadows would be to use a DEM of the same area (ideally with the same or finer spatial resolution as the sensor for your scene of interest), and apply a hill shade model to this smoothed/interpolated DEM surface - based on the same solar and viewing angle as the RS scene of interest. Then use simple thresholding to define the shadows as objects of interest - and voila you have shadow objects of specific sizes and shapes. This can be done for each RS image by changing the solar/viewing angle to match the acquisition characteristics of the sensor. In a GIS these shadow objects can then be overlaid and or queried to evaluate their differences based on size, shape, area, etc...