I have tried Region based Segmentation - through Seed Growing, Watershed algorithm, and Graphcut algorithm, but none have been successful so far. I am using MATLAB.
You could try image saliency or entropy filtering. Entropy filtering calculates the entropy of gray values in the image Image within a rectangular mask. The result is written to the output image. The process is repeated in a sliding window fashion.
In the resulting image, the forest areas should appear more whitish while the barren areas appear black, because they don't contain enough information.
Of course I assume the roads would make a problem. However you could get rid of them using morphological edge finding operations.