Submerged macrophyte assessment in rivers: An automatic mapping method using Pléiades imagery

Area of Science:

• Ecology
• Remote Sensing
• Environmental Monitoring

Background:

• Submerged macrophyte monitoring is vital for hydrosystem management and understanding global change impacts.
• Current methods like field monitoring and manual image analysis are time-consuming and labor-intensive.
• Automated, efficient tools are needed for submerged macrophyte distribution mapping in rivers.

Purpose of the Study:

• To assess the suitability of very fine-scale resolution Pléiades satellite imagery for estimating submerged macrophyte cover in river sections.
• To compare the performance of Random Forest and Support Vector Regression machine learning algorithms for this task.
• To evaluate the effectiveness of different spectral datasets and field sampling configurations.

Main Methods:

• Utilized 50 cm resolution Pléiades multispectral satellite imagery for a 1 km river section.
• Applied Random Forest and Support Vector Regression (nonparametric regression methods).
• Tested four spectral bands (red, green, blue, near-infrared) and two vegetation indices (NDVI, GRVI) with varying field sampling strategies.

Main Results:

• Both machine learning algorithms achieved good predictions of macrophyte cover (R² > 0.7, RMSE ≈ 20%).
• The Random Forest algorithm using the four spectral bands was the most effective, especially for extreme cover values (0% and 100%).
• A higher number of fine-scale field samples improved prediction accuracy compared to fewer, larger samples.

Conclusions:

• Pléiades satellite imagery combined with machine learning offers a promising approach for automated submerged macrophyte monitoring in rivers.
• The Random Forest model demonstrates strong potential for accurate macrophyte cover estimation.
• Optimizing field sampling strategies is crucial for enhancing the reliability of remote sensing-based ecological assessments.