Development of Unsupervised Classification Techniques for Water Quality Analysis

Abstract

In large thematic studies, acquiring water quality data is costly and time-consumptive. For this reason, remote sensing is used as an initial estimation technique to identify significant environmental parameters. As shown in Cox et. al. (1998), correlation/regression may be used as a statistical tool to link the radiometric measures (acquired by the remote sensing platform) to the in situ measures (acquired by field sampling). This approach is both sensible and highly effective when an adequate sampling density is used that provides significant coverage for the study region. However, most studies cannot provide this coverage since the acquisition is far too costly and/or time-consumptive to meet the stochastic measures for sampling accuracy and precision. In addition, the related problem of how best to design a spatial sample within a highly diverse system (heterogeneous environmental strata) is an ongoing field of research (Bruce and Gao 1998). This technical note examines an innovative method of performing unsupervised classification that requires no a priori field sampling. The approach is self-contained and is not statistically biased by a priori field samples. However, the technique is highly conducive to the use of an a posteriori sample of field data to refine the unsupervised classes. In this regard, the technique may be used to: (1) estimate coarse surface features that require detailed field sampling, and (2) locate unusual patterns from within large regions that require in-depth field sampling to identify cause-and-effect relationships. The unsupervised method differs significantly from traditional methods applied within commercial image processing applications (Smith and Cole 1999). The approach does not use random sampling and is not based upon equal area sampling. In addition, no a priori assumptions are made concerning the normality (symmetry, kurtosis) of the underlying image data.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jun 01, 2000
Accession Number
ADA378497

Entities

People

  • Perry Lapotin
  • Robert H. Kennedy

Tags

Communities of Interest

  • Sensors
  • Space

DTIC Thesaurus Topics

  • Accuracy
  • Algorithms
  • Atmospheric Attenuation
  • Data Science
  • Image Processing
  • Information Processing
  • Information Science
  • Network Science
  • Neural Networks
  • New York
  • Open Water
  • Pattern Recognition
  • Remote Sensing
  • Statistical Algorithms
  • Supervised Machine Learning
  • Unsupervised Machine Learning
  • Water Quality

Readers

  • Computer Vision.
  • Distributed Systems and Data Platform Development
  • Statistical inference.