Supervised Classifications of Landsat Images: an Overview of Applications Using the Semi-Automatic Classification Plugin for QGIS


Remote sensing is “the science and technology by which the characteristics of objects of interest can be identified, measured or analyzed the characteristics without direct contact” (Japan Association on Remote Sensing, 1993. Remote Sensing Note).
In the past few decades the use of remote sensing has rapidly grown because of the technology advances (e.g. spatial and spectral resolutions of data have largely improved) which fostered the development of new applications related with Geographic Information Systems (GIS) and environmental monitoring.

The land cover classification of remote sensing images is one of these applications.
In the past few months I have posted several tutorials about the land cover classifications of remote sensing images using the Semi-Automatic Classification Plugin for QGIS.
Please, read this article dated 1999 by NASA, which illustrates the main features of land cover classifications using remote sensing images, as well as the importance of spectral signatures in the identification of the land cover types.


Highlighting the importance of the interpretation of remote sensing images, I have described how to perform classifications of multispectral images and hyperspectral images, and specific remote sensing applications for environmental monitoring using a semi-automatic approach.
Following, a brief overview of these applications using Landsat images (available from the U.S. Geological Survey):


Classification of Cropland

In this tutorial, the aim is to classify an area mainly covered by cropland using a Landsat 8 image.
The following land cover classes were identified in the image:
  1. crop (e.g. fields with green vegetation);
  2. low vegetation (e.g. fields without green vegetation, or shrubland);
  3. built-up (e.g. artificial areas, buildings and asphalt);
  4. farms (e.g. farm areas);
  5. bare soil (e.g. soil without vegetation);
  6. water (e.g. surface water).

Landsat 8 color composite RGB = 543
Classification workflow

The main phases of the classification are:
  1. Conversion of raster bands from DN to reflectance;
  2. Definition of the classification inputs;
  3. Creation of the ROIs;
  4. Classification of the study area;
  5. Calculation of classification accuracy;
  6. Calculation of classification statistics.

Land cover classification (cropland represented in green)

Forest Monitoring

In this post we aim at assessing the land cover change of an area mainly covered by forest, using a Landsat 5 image of the 1985 and a Landsat 8 image of 2013.
The following land cover classes are identified:
  • Forest (e.g. trees);
  • Non-Forest (e.g. bare soil, grassland, artificial areas);
  • Water (e.g. surface water).


Landsat 5 color composite RGB=432
Landsat 8 color composite RGB=543

The following are the main classification phases:
  1. Conversion of raster bands from DN to Reflectance for the Landsat 5 image;
  2. Creation of the ROIs and spectral signatures for the Landsat 5 image;
  3. Classification of the Landsat 5 image;
  4. Conversion of raster bands from DN to Reflectance for the Landsat 8 image;
  5. Creation of the ROIs and spectral signatures for the Landsat 8 image;
  6. Classification of the Landsat 8 image;
  7. Calculation of the land cover change from 1985 to 2013.

Land cover change from 1985 to 2013 (deforestation represented in blue)

Flood Monitoring

In this tutorial I illustrate how to monitor flood areas performing the supervised classification of two Landsat 8 images, before and after the flood.
The only class identified is water, using a threshold for the classification algorithm.

Landsat 8 before flood color composite RGB=543

Landsat 8 after flood color composite RGB=543

The following are the main classification steps:
  1. Conversion of raster bands from DN to Reflectance for the image before flood; 
  2. Creation of the ROIs and spectral signatures for the image before flood; 
  3. Classification of the image before flood; 
  4. Conversion of raster bands from DN to Reflectance for the image after flood; 
  5. Classification of the image after flood; 
  6. Calculation of the land cover change.

Land cover change (flood are represented in red)

Wildfire Monitoring

The aim of this tutorial is to identify the burnt area of a wildfire using a Landsat 8 image. Therefore, the only class identified is the burnt area, using a threshold for the classification algorithm.

Landsat 8 color composite RGB=543 


The main classification phases are:
  • Conversion of raster bands from DN to Reflectance;
  • Creation of the ROIs and spectral signatures;
  • Classification of the image using a threshold;
  • Estimation of the burnt area.

Land cover classification of burnt areas

Surface Temperature Estimation

In this tutorial a Landsat 8 thermal band is used for the estimation of surface temperature.
In particular, a land cover classification is used for the definition of surface emissivity, which is required for the calculation of the land surface temperature.

Landsat 8 color composite RGB=543

The following phases are required:
  1. Conversion of raster bands from DN to reflectance and At Surface Temperature;
  2. Land cover classification of study area;
  3. Reclassification of the land cover classification to emissivity values;
  4. Conversion from At Surface Temperature to Land Surface Temperature.

Land Surface Temperature map (in kelvin)


These are only a few applications of remote sensing for environmental monitoring. The main aim of these tutorials is to demonstrate how it is possible to classify the land cover in a rapid and affordable way.

The cost of data (satellite images) for performing land cover monitoring is one of the main constraints for administrations or organizations. In fact,the use of free images and open source software means no cost for administrations or environmental organizations; in the light of the free Landsat availability, a big thank the NASA and USGS for providing their invaluable image database free to all.

Also, it is worth mentioning the upcoming European satellite Sentinels-2 that will be a valuable alternative and integration to Landsat data for classifications.

For the next few years a large amount of remote sensing data will be available for free. This means a great opportunity for monitoring land cover and integrating remote sensing data into spatial analyses. There is really no excuse for ignoring environmental change.

Please, remember that a Facebook group and a Google+ Community are available for sharing information and asking for help about the Semi-Automatic Classification Plugin.
Newer posts Older posts