How to Remove Topographic Effects Like Shadows from Landsat Image

Question

I am using QGIS 2.4.0 Chugiak on my windows 8.1, for classification of landsat 5-7 image that contains hilly mountains with dense forest and 60% area are plains. I am classifying through Semi-automatic Classification Plugin by Spectral Angle Mapping algorithm but I am not getting a properly classified image.

I have carefully collected the signatures and analysed those on a spectral signature plot. Before starting the classification algorithm, I converted the radiance to Top of Atmosphere reflectance through automated process and clipped my area of interest using "clip multiple raster" under preprocessing tab through shape file.

My question is this. Is there something wrong with whole process or first do I have to eliminate the terrain shadows after the conversion to TOA reflectance? I also have the SRTM DEM 30m resolution. If it is necessary to remove those areas that are under shadows then how; if not then how can I improve my classification through this automated plugin?

Answer 1

Yes, image classification is generally improved when you remove topographic illumination effects (same goes with atmospheric effects). However, as with anything like this, there exists a wide range of techniques for accomplishing this task and the effectiveness of it will depend both on the sophistication of the technique (its ability to model the physical processes involved) and your data. Here is an excellent quick read which overviews some of the available methods:

http://web.pdx.edu/~nauna/Topographic_Normalization.pdf

If you have access to Jensen's excellent remote sensing textbook then he also has a good section in there on topographic illumination and strategies for its removal/minimization. Basically, the various methods can be grouped into those which are based on band ratios and those that are based on some kind of physical model. Band ratios are generally not as effective but have simplicity and fewer data requirements. That is, you don't even need a DEM or derived illumination surface to use band rationing. Have you ever noticed that the amount of topographic illumination is severely reduced in an index like the NDVI? It's because it's based on a type of band ratio. Slide 3 of this presentation shows why this works. Ultimately, if you choose this approach then your classification will be based on a series of band ratios (e.g. NIR / red) rather than the raw imagery.

Since you have a DEM and can therefore derive an illumination surface (somewhat like a hillshade image but controlled for sun direction and angle), then you can use a more sophisticated approach like the Cosine correction or one of its many derivatives. Other than the illumination surface (hillshade tool), you can perform the analysis using a raster calculator based workflow. It will likely require some experimentation with various parameters to get the best results. Good luck.