Global model for riverine suspended sediment concentration from Landsat

Area of Science:

• Environmental Science
• Remote Sensing
• Hydrology

Background:

• Long-term suspended sediment concentration (SSC) data are crucial for monitoring river health and response to global changes.
• Existing methods for estimating SSC often lack global coverage or long-term consistency.
• Satellite remote sensing offers a promising avenue for large-scale, continuous SSC monitoring.

Purpose of the Study:

• To develop and evaluate a global model for estimating riverine suspended sediment concentration (SSC) using multi-decadal Landsat surface reflectance data.
• To create the largest global matchup database of in-situ SSC and remote sensing data for model training and validation.
• To harmonize surface reflectance data across different Landsat sensors (TM, ETM+, OLI) for consistent long-term analysis.

Main Methods:

• Compiled a global database of 240,224 in-situ SSC measurements and corresponding Landsat surface reflectance data.
• Developed empirical models to harmonize water surface reflectance across Landsat TM, ETM+, and OLI sensors using ~88 million observations (1984-2021).
• Employed an extreme gradient boosting (XGBoost) algorithm to develop the SSC prediction model, trained and validated with data spanning 0.1 to 5,760 mg/L.

Main Results:

• The developed SSC prediction model achieved high accuracy, comparable to or exceeding existing methods (RMSE = 5.22 mg/L, RMSLE = 0.24).
• The model demonstrated consistent spatial and temporal prediction capabilities, validated through spatial-temporal cross-validation.
• The model successfully estimated SSC across a wide range of water turbidity, from clear to extremely turbid conditions.

Conclusions:

• A robust, globally applicable model for estimating riverine SSC from Landsat data has been successfully developed.
• This model enables the creation of long-term, consistent global riverine SSC records, supporting environmental change studies.
• The harmonized surface reflectance and XGBoost-based SSC estimation provide a valuable tool for global river monitoring.