Unsupervised Classification

by Megan Kanaga Creutzburg

Other Names:

None known

Description

One common application of remotely-sensed images to rangeland management is the creation of maps of land cover, vegetation type, or other discrete classes by remote sensing software. In unsupervised classification, image processing software classifies an image based on natural groupings of the spectral properties of the pixels, without the user specifying how to classify any portion of the image. Conceptually, unsupervised classification is similar to cluster analysis where observations (in this case, pixels) are assigned to the same class because they have similar values. The user must specify basic information such as which spectral bands to use and how many categories to use in the classification, or the software may generate any number of classes based solely on natural groupings. Common clustering algorithms include K-means clustering, ISODATA clustering, and Narenda-Goldberg clustering.

Unsupervised classification yields an output image in which a number of classes are identified and each pixel is assigned to a class. These classes may or may not correspond well to land cover types of interest, and the user will need to assign meaningful labels to each class. Unsupervised classification often results in too many land cover classes, particularly for heterogeneous land cover types, and classes often need to be combined to create a meaningful map. In other cases, the classification may result in a map that combines multiple land cover classes of interest, and the class must be split into multiple classes in the final map. Unsupervised classification is useful when there is no preexisting field data or detailed aerial photographs for the image area, and the user cannot accurately specify training areas of known cover type. Additionally, this method is often used as an initial step prior to supervised classification (called hybrid classification). Hybrid classification may be used to determine the spectral class composition of the image before conducting more detailed analyses and to determine how well the intended land cover classes can be defined from the image.

Similar Methods

Output

Unsupervised classification methods generate a map with each pixel assigned to a particular class based on its multispectral composition. The number of classes can be specified by the user or may be determined by the number of natural groupings in the data. The user must then assign meaning to the classes, and combine or split classes where necessary to generate a meaningful map.

Successful Rangeland Uses

Unsupervised classification has been used extensively in rangelands for a wide range of applications, including:

  • Land cover classes
  • Major vegetation types
  • Distinguishing native vs invasive species cover
  • Vegetation condition
  • Disturbed areas (eg. fire)
  • Land use change

Application References

The following references are only a few examples of this widely used technique

  • Everitt, J. H., C. Yang, D. E. Escobar, R. I. Lonard, M. R. Davis. 2002. Reflectance Characteristics and Remote Sensing of a Riparian Zone in South Texas. The Southwestern Naturalist 47: 433-439 – used unsupervised classification to map dominant vegetation types
  • Everitt, J. H., C. Yang, R. S. Fletcher, and D. L. Drawe. 2006. Evaluation of High-Resolution Satellite Imagery for Assessing Rangeland Resources in South Texas. Rangeland Ecology and Management 59:30-37 – used unsupervised classification to identify rangeland cover types
  • Kreuter, U.P., H.G. Harris, M.D. Matlock, and R.E. Lacey. Change in ecosystem service values in the San Antonio area, Texas. Ecological Economics 39: 333-346 – used unsupervised classification to detect land use change due to urbanization
  • Levien, L.M., P. Roffers, B. Maurizi, J. Suero, C. Fischer, and X. Huang. 1999. A machine-learning approach to change detection using multi-scale imagery. Presented at the American Society of Photogrammetry and Remote Sensing 1999 Annual Conference. Portland, Oregon, May 20, 1999 – used unsupervised classification to create a map of changes in land use that was used to select field sampling sites
  • Stitt, S., R. Root, K. Brown, S. Hager, C. Mladinich, G.L. Anderson, K. Dudek, M.R. Bustos, and R. Kokaly. 2006. Classification of Leafy Spurge With Earth Observing-1 Advanced Land Imager. Rangeland Ecol Management 59:507–511 – use of unsupervised classification to track invasive species

Technical References

  • Congalton, R.G. 1991. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment 37:35-46.
  • Ghorbani, A., D. Bruce, and F. Tiver. 2006. Specification: A problem in rangeland monitoring. In: Proceedings of the 1st International Conference on Object-based Image Analysis (OBIA), 4th-5th July 2006, Salzburg, Austria.
  • Jensen, J. R. 1996. Introductory digital image processing. Prentice-Hall, Inc., Upper Saddle River, NJ.
  • Karl, J. W., and B. A. Maurer. 2009. Multivariate correlations between imagery and field measurements across scales: comparing pixel aggregation and image segmentation. Landscape Ecology. DOI: 10.1007/s10980-009-9439-4
  • Lillesand, T. M., and R. W. Kiefer 1994. Remote sensing and image interpretation. John Wiley & Sons, Inc., New York.

Limitations

Arbitrarily changing classification parameters can result in very different land use classifications and maps. Without field data, it can be difficult to interpret the maps and determine how land use classes correspond to the software-derived classes. A combination of supervised and unsupervised classification (hybrid classification) is often employed; this allows the remote sensing program to classify the image based on the user-specified land cover classes, but will also classify other less common or lesser known cover types into separate groups. Supervised and unsupervised classification are both pixel-based classification methods, and may be less accurate than object-based classification (Ghorbani et al. 2006, Karl and Maurer 2009).

Data Inputs

Unsupervised classification can be performed with any number of different remote-sensing or GIS-derived inputs. Commonly, spectral bands from satellite or airborne sensors, band ratios or vegetation indices (e.g., NDVI), and topographic data (e.g., elevation, slope, aspect) are used as inputs for unsupervised classification.

Software/Hardware Requirements

Unsupervised classification is relatively easy to perform in any remote sensing software (e.g., Erdas Imaging, ENVI, Idrisi), and even in many GIS programs (e.g., ArcGIS with Spatial Analyst or Image Analysis extensions, GRASS).

Sample Graphic

A false color satellite image of the Welder Wildlife Refuge clearly differentiates at least three of the major vegetation types shown as numbered arrows: 1- riparian woodland, 2- green herbaceous vegetation, 3- spiny aster (A). Unsupervised vegetation classification resulted in map B with 6 vegetation classes identified: red- riparian woodland, yellow- green herbaceous vegetation, purple- spiny aster, green- stressed herbaceous vegetation, white- sparsely vegetated/bare soil, and blue- water. Accuracy assessments based on field data showed that the classification was 79-89% accurate (image from Everitt et al 2006).

Additional Information

Comments are closed.