Performance across WorldView-2 and RapidEye for reproducible seagrass mapping

Area of Science:

• Environmental Science
• Remote Sensing
• Ecology

Background:

• Ground-based and aerial mapping methods face challenges in assessing global seagrass extent.
• Commercial satellite platforms offer high spatial resolution suitable for patchy seagrass ecosystems.
• A lack of standardized protocols for commercial satellite data processing hinders reproducibility and cross-temporal/spatial comparisons.

Purpose of the Study:

• To develop and validate a reproducible image processing workflow for commercial satellite data (WorldView-2 and RapidEye).
• To radiometrically assess the performance of WorldView-2 and RapidEye sensors over coastal waters.
• To evaluate the accuracy of seagrass classification using these commercial satellite data products.

Main Methods:

• Acquired WorldView-2 and RapidEye imagery from St. Joseph Bay, Florida, during a November 2010 field campaign.
• Developed a reproducible processing regime to generate analysis-ready data from basic commercial satellite products.
• Compared satellite-derived surface reflectances with field measurements and classified imagery using a deep convolutional neural network.

Main Results:

• WorldView-2 showed high disagreement in blue spectral bands, while RapidEye exhibited slight overprediction across all bands.
• Both platforms achieved up to 97% agreement with aerial seagrass estimates, despite resolution differences.
• Radiometric calibration issues in WorldView-2's shorter wavelengths were identified but did not impede classification accuracy.

Conclusions:

• A reproducible image processing workflow was successfully developed and tested for WorldView-2 and RapidEye imagery.
• Commercial satellite data, with appropriate processing, can be effectively used for seagrass mapping.
• While calibration issues exist, the developed workflow provides a robust method for seagrass extent assessment and may be adaptable to future sensors.