Enhancing Wetland Mapping: Integrating Sentinel-1/2, GEDI Data, and Google Earth Engine

Area of Science:

• Ecology
• Remote Sensing
• Geospatial Analysis

Background:

• Wetlands are critical ecosystems facing mapping challenges due to landscape complexity and spectral similarities.
• Accurate wetland mapping is essential for effective conservation and management strategies.
• Earth observation (EO) data offers potential for enhanced wetland characterization.

Purpose of the Study:

• To produce high-resolution (10 m) wetland classification maps for Newfoundland, Canada.
• To evaluate the synergistic use of optical, SAR, and LiDAR data for wetland mapping.
• To assess the added value of vegetation canopy height (VCH) derived from GEDI LiDAR in wetland classification.

Main Methods:

• Utilized multi-source EO data including Sentinel-1/2, GEDI LiDAR, MERIT Hydro, and ERA5.
• Integrated elevation and topographical derivatives (slope, aspect).
• Employed Google Earth Engine (GEE) and Random Forest (RF) model, generating a VCH map and using it as a predictor.

Main Results:

• The VCH variable derived from GEDI significantly improved wetland classification accuracy.
• VCH achieved high accuracy (R2=0.69, RMSE=1.51 m, MAE=1.26 m).
• Incorporating VCH enhanced overall accuracy to 93.45% (Kappa=0.92, F1=0.88).

Conclusions:

• Multi-source, multi-sensor EO data approaches are crucial for effective wetland mapping.
• The integration of GEDI-derived VCH is a valuable advancement for wetland classification.
• This research provides a robust methodology for future wetland mapping initiatives.