Using remote sensing in support of environmental management: A framework for selecting products, algorithms and...

Area of Science:

• Environmental Science
• Remote Sensing
• Geospatial Analysis

Background:

• Traditional remote sensing mapping relies on selecting a single best-performing model.
• This single-model approach can be suboptimal as classifier performance varies with data properties.

Purpose of the Study:

• To develop and demonstrate an improved remote sensing mapping methodology.
• To create a probability map integrating multiple consistently strong models for environmental feature mapping.

Main Methods:

• Evaluated multiple classification algorithms (maximum likelihood, support vector machine) and satellite data (SPOT 5, Landsat 8).
• Utilized omission/commission plots and standard accuracy measures to assess model performance.
• Compiled a probability map based on consistently high-performing models across various data and algorithms.

Main Results:

• No single 'best fit' model was identified across all data and algorithm combinations.
• Classifier performance is data-dependent, confirming existing literature.
• A probability map derived from multiple robust models offers a more rigorous mapping solution.

Conclusions:

• The proposed probability mapping approach is easy to use and applicable to various remote sensing studies.
• This method provides a more reliable and rigorous approach to mapping natural features for management.
• Integrating results from multiple well-performing models enhances the reliability of remote sensing-based environmental maps.