An Advanced Data Fusion Method to Improve Wetland Classification Using Multi-Source Remotely Sensed Data.

Area of Science:

• Environmental Science
• Remote Sensing
• Geospatial Analysis

Background:

• Accurate wetland classification is crucial for environmental monitoring and management.
• Traditional methods often struggle to fully utilize diverse remote sensing data, leading to classification inaccuracies.

Purpose of the Study:

• To improve wetland classification accuracy by integrating multi-source remote sensing data.
• To develop a novel classification methodology that maximizes information extraction and classification performance.

Main Methods:

• Utilized a combination of Landsat-8, Sentinel-2 (multi-spectral), Sentinel-1 (SAR), and Digital Elevation Model (DEM) data.
• Employed three distinct random forest (RF) classifiers and integrated their results using Dempster-Shafer theory (D-S theory).
• Tested the methodology in a study area in Northern Alberta, Canada, classifying fen, bog, marsh, swamps, and upland.

Main Results:

• Achieved an overall classification accuracy of 0.93, a 5% improvement over traditional methods.
• Significantly reduced high-confidence misclassifications (by ~10%) compared to the traditional approach.
• Identified and incorporated previously overlooked features that enhanced the separation of compound wetland classes.

Conclusions:

• The proposed D-S theory-integrated RF classification method effectively leverages multi-source remote sensing data for improved wetland mapping.
• This approach offers a significant advancement in wetland classification accuracy and reliability.
• The methodology demonstrates the value of fully exploiting diverse geospatial data for environmental applications.