Help SAGA help you... Clipping your extent

Hello everyone,

Most people might be done with their analysis but for those that aren't, here is something that might help you run your tools faster and with more accuracy (although you will lose training area space..., so think about that when you decide to clip your extent, in addition to what area is relevant to the analysis). If your study area is a fraction of your extent:

-> in QGIS

• open your Landsat data
• create a shapefile that covers your desired extent
• save the shapefile as an ESRI shapefile in the same coordinate system as your Landsat scene (careful about this, if your Landsat data from the two years are in different coordinate systems (eg, if one is UTM25 and another UTM24) then export two versions of your clipping extent...)

-> in SAGA

• open the shapefile in SAGA by double-clicking, it will appear in your data tab
• go to Geoprocessing > shapes > grid > spatial extent > clip grid with polygon
• when the tool opens, do this: insert the grid that you want to clip, choose the file from that grid that you want to clip (can be more than one file), choose your polygon, and for your target extent choose the original (this way the new file will fall under the same data folder as the original image). Run the tool

The result will be good and exciting but the bands will be separated...

Export the bands as a GeoTIFF in order to automatically re-concatenate them. Follow this:

• go to your tools tab > Import/Export > GDAL/OGR > Export GEOTIFF
• when you run the tool, choose the grid in question, add all the bands IN THE RIGHT ORDER, name your file and save it somewhere smart and run the tool. The output will be a concatenated version of your clipped extent.

With a clipped extent SAGA will do a much better job at parsing out different pixel classes when you run unsupervised classification (I don't know about supervised because I have not used this method of classification). You can also increase the number of classes and the number of iterations without fearing that your tool will be running the whole night. The more classes and the more iterations (up to a point) the finer the classification and the better SAGA is at telling one thing apart from the other.