SCP Questions of This Month: March

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.

Question

I am trying to map the land cover map one woreda (AOI) and the instructor told me to classify the land cover of the area as the following types:
1.intensively cultivated land
2.moderately cultivated land
3.. dense forest
4.dense woodland
5.bush wood land
6.bush shrub land
7.bare land
8.built up
9.water body
Then, how can classify such land cover type using the qgis or SCP plugin?

Answer

Please read this tutorial http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/scp_dock.html
It can be difficult to distinguish between intensively cultivated land and moderately cultivated land.
Also, bush wood land and bush shrub land are very similar.
Possibly you can try to classify multiple images acquired in different times, in order to distinguish these vegetation classes.
NDVI calculation could help. For instance read this tutorial https://fromgistors.blogspot.com/2016/11/from-image-download-to-ndvi-calculation.html

Question

I am trying the SCP processing workflow and would like to speed up the image downloading stage. When downloading Landsat 8 images, SCP gets them band by band (and this is slow). On the other hand, when exporting the download links (pointing to .tar.bz files on http://storage.googleapis.com/earthengine-public/), the procedure in browser or program is much faster.
Is it possible to use the same DL procedure in SCP? Alternatively, is it possible to DL files and do pre-processing and band calc later in SCP? Basically, I would like to tell SCP to pre-process all the images in a folder and subfolders.

Answer

Yes, you can download the images in browser and then preprocess them. You can use the tool batch (see http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/main_interface_window.html#batch). For instance read this tutorial https://fromgistors.blogspot.com/2016/11/from-image-download-to-ndvi-calculation.html

Question

The problem is when I try to run the atmospheric correction or creation of the ROI, after five seconds (or 5% of the process) the program qgis is closed.
I have tried with Landsat sentinel and other images.
System: Kde Neon 5.9.3 (updates to day), Qgis 2.18.4, SCP  5.3.6

Answer

Probably you need to install the plugin dependecies. See http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/installation_ubuntu.html
In case you still get the same error, you can try this virtual machine http://semiautomaticclassificationmanual-v5.readthedocs.io/en/latest/semi-automatic_os.html

Question

When I download images Landsat 5 and 8 from an area of Ecuador, in the metadata note that the projection says wgs84 17 N, but in reality it is 17 S, in QGIS I work normally and I did the classification, I had no problem with the projection. My question is: I must re-do the classification and change the projection of the images?.

Answer

There is no need to reproject data for classification. In case you need a different projection, I suggest reprojecting only the classification.

Question

I tried to use scp to classify landsat 8 images of the study area. This area is mostly arid mountainous with few villages (attached). I did my best to select good trainings for LULCs classification (built-up, vegetation and bareland), But after testing all algorithms, no one could distinguish barelands and built-up areas appropriately.
I used atmospherically corrected bands 2-7 landsat 8.

Answer

Built-up and bareland can have very similar spectral signatures, therefore the classification can be very difficult.
You can try the method shown in this tutorial (https://fromgistors.blogspot.com/2016/10/land-cover-signature-classification.html)  to discriminate similar signatures . Also, you could consider other techniques using multitemporal images, performing raster calculations of spectral indices (e.g. NDVI) trying to mask bareland during the year.


If you know Python you could create the batch text itereating through the image files.
For any comment or question, join the Facebook group and the Google+ Community about the Semi-Automatic Classification Plugin.

Newer posts Older posts

Follow

RG