ABSTRACT Monitoring of inland water quality is of significant importance due to the increase in water quality related issues, especially within the Midwestern United States. Traditional monitoring techniques, although highly accurate, are vastly insufficient in terms of spatial and temporal coverage. Using a virtual constellation by harmonizing Landsat-8 and Sentinel-2 data a high temporal frequency dataset can be created at a relatively fine spatial scale. In this study, we apply a novel deep learning method for the estimation of blue-green algae (BGA), chlorophyll-α (Chl), fluorescent dissolved organic matter (fDOM), dissolved oxygen (DO), specific conductance (SC), and turbidity. The developed model is evaluated against previously studied machine learning methods and found to outperform multiple linear regression (MLR), support vector machine regression (SVR), and extreme learning machine regression (ELR) generating R2 of 0.91 for BGA, 0.88, 0.89, 0.93, 0.87, and 0.84 for Chl, DO, SC, and turbidity respectfully. This model is then applied to all available data ranging from 2013–2018 and time series for each variable were generated for four selected waterbodies. We then use the Empirical Data Analytics (EDA) anomaly detection method on the time series to identify abnormal data points. Upon further analysis, the EDA method successfully identifies abnormal events in water quality. Our results also demonstrate strong correlation between non-optically active variables such as SC with Chl and fDOM. The framework developed in this study represents an efficient and accurate empirical method for inland water quality monitoring at the regional scale.