From GIS to Remote Sensing
This post is a collection of questions and answers about the Semi-Automatic Classification Plugin (SCP) and remote sensing which were discussed in the Facebook group and the Google+ Community this month.
These questions vary from supervised classification technique to software issues, and can be useful to the readers of this blog for solving issues about the use of SCP.
These questions vary from supervised classification technique to software issues, and can be useful to the readers of this blog for solving issues about the use of SCP.
Question
Is there a way to save a custom RGB band composite to the default RGB list instead of having to re-enter it every time a new QGIS session is started? For example, I would like to input 4-1-2 into the list and have it available to me the next time I use QGIS instead of having to re-enter it each time.
Answer
RGB composites are saved inside the project. You can save an empty project with your custom RGB composites and open this project every time you work with SCP.
Question
Is there a way that you can compute the area of each class after performing the classification?
Answer
About class area you can use this tool http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/main_interface_window.html#classification-report
Question
I have a question about training, I chose Rizophora macroclase for example, then I did three classes: Rizophora 1, Rizophora 2 and Rizophora 3. I repeated for all other land use: MC Foret C Foret 1, Foret 2 and Foret 3. and so on.
For the classification use only the macroclasses, and the classification is well done, however when I want to get the value of the spectral signature of each class, I have three values, example Rizophora 1, Rizophora 2 and Rizophora 3, my question is whether it should Make a merge of the three values? To obtain a single spectral signature.
Answer
Merging the signatures depend on your objective.
When you use macroclasses, the single spectral signatures of classes are considered separately, and then the macroclass value is assigned. Basically you can use very different signatures to get one class in the classification.
If you merge the signatures you get the average spectral signature, which could be wrong if signatures are very different.
Depending on the classification algorithm, it could be useful to merge signatures to increase spectral variability (e.g. merge very small ROIs).
When you use macroclasses, the single spectral signatures of classes are considered separately, and then the macroclass value is assigned. Basically you can use very different signatures to get one class in the classification.
If you merge the signatures you get the average spectral signature, which could be wrong if signatures are very different.
Depending on the classification algorithm, it could be useful to merge signatures to increase spectral variability (e.g. merge very small ROIs).
Question
According to the article (http://www.un-spider.org/node/10955), until the step 5 you don't clip the study area. What happen if your study area is clipped in the first step (clipping the area in every tile)? It seems that the further analysis for your computer will takes less time but I'm not sure if the result of Index will be different when you use SCP with the area clipped.
Answer
If you clip the image before the DOS1 correction (which is based on the image) you get different reflectance values (please read http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/remote_sensing.html#dos1-correction). In general, it is better to perform first DOS1 correction, and then image clipping.
Question
Is there any way that I can merge several scp-files into one?
Answer
You can use this import tool http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/main_interface_window.html#import-signatures
Question
Is there a way to save a custom RGB band composite to the default RGB list instead of having to re-enter it every time a new QGIS session is started?
Answer
RGB composites are saved inside the project. You can save an empty project with your custom RGB composites and open this project every time you work with SCP.
Question
I have a Sentinel-2 .tif file which inlcudes the following bands: 1,2,3,4,5,6,7,8,9 and 10.
I wanted to use this file with SCP and start the classification (identification of ROI and classes attribution). I had a look at the tutorial http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/tutorial_2.html and noticed that you used different bands for the band set.
The question is: can i stick to the bands i have in order to classify the image for habitat mapping purposes and follow the rest of your tutorial bearing in mind the bands i have are not exactly yours?
Answer
Yes, you can use those bands for classification. However, considering the spatial resolution and spectral information of bands 1, 9, and 10 (see http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/remote_sensing.html#sentinel-2-satellite) I generally avoid using these bands. But perhaps these bands are useful for your study.
Question
I realized two map of LST, the first refer to July 2003 (during a heatwave) from Landsat 5, the second refer to July 2015 from Landsat 8.
1) in the phase of land cover classification, for the agricultural areas that appear without vegetation, is correct to classify as "bare soil",or better to check also the respective NDVI value (eg. <0.5 bare soil, > 0.5 vegetation)?
2) in order to assess the difference between July 2003 and 2015, is it better to produce two different land cover classifications (as I've done) or use the same classification for the two image (eg. derived from other source as CORINE Land Cover 4th level)?
3) for the reclassification to emissivity, could you suggest more detailed values, in addition of the value used in your tutorial?
4) finally: i notice that in july 2015 the LST values derived from Landsat 8 seem too high; I have the data of the local metereological observatory that, at the same day and hour, register less 3 °C than the value of the respective LST raster cell. Is it enough to adjust the LST raster for the total city area?
Answer
1) Yes, you can use NDVI for classification. Of course bare soil means that there is almost no vegetation cover, so NDVI could be lower than 0.3 or 0.2.
2) I think it is better to use two classification in order to assess the actual differences, but also the alternative approach is feasible.
3) Unfortunately emissivity should be collected in field. Example of emissivity can be found in internet but the difference with the ground truth can be significative.
4) I think meteorological data are referred to air temperature, while LST is temperature at the surface of land, therefore there can be differences. Possibly you can find models to relate these two temperatures. Also you could check with temperatures obtained from different satellites such as MODIS.
2) I think it is better to use two classification in order to assess the actual differences, but also the alternative approach is feasible.
3) Unfortunately emissivity should be collected in field. Example of emissivity can be found in internet but the difference with the ground truth can be significative.
4) I think meteorological data are referred to air temperature, while LST is temperature at the surface of land, therefore there can be differences. Possibly you can find models to relate these two temperatures. Also you could check with temperatures obtained from different satellites such as MODIS.
For any comment or question, join the Facebook group and the Google+ Community about the Semi-Automatic Classification Plugin.