Search the Community

Hello Community, this is my first post in this forum, i hope it will meet all the standarts which are necessary here I'm currently working on a project for my university course in remote sensing of vegetation. I mainly want to use R-Script, as this is what we learned during the semester and it seems to be a very powerful tool to get these things done. Well, let's get straight to the topic: I want to analyze the LandUse/LandCover change in the Euphrat basin in Syria during the civil war. So i downloaded (and preprocessed) 4 different Landsat 7 sceneries (April 2009,2011,2014, 2015). Now i want to create a time series to show if the agricultural use in my scenes changed. To do so, i thought i'll do a classification using the RandomForest package for R. My question is: how do i choose the right training pixels? Should i search in every single scene enough pixels via googleearth for each class, let RF do it's job and compare the results (like: "oh, there are much less "agricultural" pixels in the scene from 2015 then in 2009!") or should i select only pixels for the classes from my 2009 scene, let RF do it's magic and then use the same trainingpixels for the 2015 scene and map the pixels which changed? (that would probably mean the preprocessing of the 2011/2014 scenes was quite senseless?) What's the best approach to reach my goal? I'm thankful for any input! If there is any tutorial showing how to do this, i m glad for an url (although i searched quite long for something like this...) Thanks!