Performance of Machine Learning Algorithms and Multi-sensor Data in the Retrieval of Optically-active Water Quality Parameters in Inland Water Bodies. A Systematic Review

Remote sensing has emerged as a powerful and cost-effective alternative, providing synoptic and repeatable observations of Optically Active water quality Parameters (OAPs), such as Chlorophyll- a (Chl- a ), Total Suspended Solids (TSS), Secchi Disk Depth (SDD), Coloured Dissolved Organic Matter (CDOM), and turbidity. This paper synthesised, for the first time, recent advances in the application of remote sensing and machine learning for OAP assessment and monitoring, with a specific focus on comparative analysis of their retrieval performance in inland water bodies. We applied the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol to select 127 peer-reviewed studies published between 2010 and June 2025. The findings highlight the predominant use of Sentinel-2 and Landsat-8 multispectral data in 43% of the publications, while machine learning algorithms, such as Artificial Neural Networks, Random Forest, and Support Vector Regression, were preferred but showed variable retrieval performance. Recent methods, such as Convolutional Neural Networks and Extreme Gradient Boosting, achieved the highest mean prediction accuracies of R 2 = 0.85 and R 2 = 0.78, respectively. However, there is a skewed geographical distribution, with the majority of the publications emanating from China and the United States of America; hence, the retrieval performance in other regions is largely unknown. Among OAPs, Chl- a was by far the most studied OAP; turbidity and TSS also featured frequently, whereas CDOM and SDD received relatively little attention. In terms of performance, TSS and turbidity are retrieved with high accuracy (i.e., a mean R 2 of 0.72), attributed to the presence of crucial spectral regions such as 520–720 nm and 800–900 nm in all optical sensors. Finally, this study found several gaps which should pave a trajectory for future studies: (1) CDOM and SDD remain underrepresented and difficult to predict, (2) there is a need to adopt multi-output models (simultaneously estimating multiple OAPs) for improved accuracy as OAPs interact in the same spectral regions and water bodies, and (3) there is a persistent challenge of a lack of robust atmospheric correction algorithms for water bodies. Overall, this study provides critical information on the performance of machine learning algorithms and multi-sensor data in retrieving OAPs, allowing for further development and informing operational water quality monitoring.