nomlas

@ sijooss Feature space based classification are rather simplistic when compared to ensemble methods like random forest, boosting or bagging. A number of recent multispectral and hyperspectral studies have confirmed this. Due to the nature of the random forest , the pixel values are considered are number of times as a result of the bootstrapping aggregation, making the algorithm very robust and accurate. Additionally, by considering different variables (i.e bands) for creating the trees, an additionally measure of randomness is introduced. hence the more diversity thta is introduced into the algorithm the more powerful it becomes. Thanks

Hi Since you are aware of the classification accuracy measures available in Erdas (Envi also has some nice post classification accuracy measures like AUC and ROC) and Lurker has mentioned the Chi squared test another option will be to use the map comparison kit available at -->Map Comparison Kit Website, For details and examples, please see: H. Visser and T. de Nijs, 2006. The Map Comparison Kit. Environmental Modeling & Software 21, 346-358. Thanks

Hi you can run randomforest in R , using the randomForest library or alternatively if you prefer you can download the rattle GUI (rattle.togaware.com), your shapefile containing your training data should have the values from your tiff file extracted then you can use the dbf file in rattle. there is a tutorial on the togaware webpage. There is also a library that creates maps based on your RF analysis called modelmap Recently, I came across a stand alone IDL based remote sensing software called EnMAP-Box that allows you to carry out a random forest classification/regression , you check out the software at Environmental Mapping and Analysis Program Documentation also on the website The paper below explains the use of the software imageRF – A user-oriented implementation for remote sensing image analysis with Random Forests Environmental Modelling & Software, Volume 35, July 2012, Pages 192-193 Björn Waske, Sebastian van der Linden, Carsten Oldenburg, Benjamin Jakimow, Andreas Rabe, Patrick Hostert or you can try--> openmodeller.sourceforge.net Thanks

Hi You need to have a dem for your study site,and if you using erdas then try the topographic correction (using mineartequation) otherwise you might have to try and correct the image using atcor2/3 for erdas thanks

Hi You have to check the band sequence of your spot image, if you imported using dimap format then the band sequence is 3214, and it should be 1234 ,if you band sequence is correct than the vegetation classes should display a higher NDVI value