SatBuoy : Mississippi Satellite-based Virtual Buoy Observation Network
How does a virtual buoy work?
Each SatBuoy station is a fixed geographic location where satellite data are repeatedly sampled over time. For each satellite overpass, water-quality values are extracted from clear-sky pixels surrounding the location and appended to a growing time series.
This approach allows SatBuoy to:
-
Monitor many locations simultaneously;
-
Track short-term variability (days to weeks) and long-term trends (years);
-
Provide consistent observations since 2016.
Unlike physical buoys, virtual stations can be easily added, relocated, or expanded without field deployment.
Why does SatBuoy matter?
SatBuoy supports coastal management by delivering:
-
Continuous water-quality information
-
Long-term historical records
-
Spatial coverage across sensitive ecosystems
In the Mississippi Sound, SatBuoy helps track water-quality changes driven by river discharge, sediment transport, and biological activity, offering valuable insight for ecosystem conservation and restoration planning.
Overview
SatBuoy integrates satellite observations, field measurements, and data-driven models to generate reliable water quality time series at virtual buoy stations. The workflow consists of four main steps:
1) Satellite data processing (S3CARD framework)
SatBuoy uses Sentinel-3 Ocean and Land Colour Instrument (OLCI) imagery, and all satellite data are processed using the Sentinel-3 Coastal Analysis Ready Data (S3CARD) framework. S3CARD converts raw satellite observations into analysis-ready surface reflectance by applying:
-
Atmospheric correction
-
Adjacency correction to reduce land contamination
-
Sun-glint correction
-
Robust water and cloud masking
This processing step is critical for coastal environments, where atmospheric effects and proximity to land can significantly affect satellite signals.
2) Water Quality Modeling Using Machine Learning
A series of in situ data collection campaigns were performed in 2024 and 2025 at 25 locations across Mississippi Sound, and we prepared the radiometric and limnological datasets for water quality modeling. The following parameters are measured:
-
Turbidity, describing the degree to which light is scattered or absorbed by suspended and colloidal particles present in the water
-
Total Suspended Solids (TSS), quantifying the mass of suspended material in the water column
-
Secchi depth, indicating water clarity and light penetration
-
Chlorophyll-a, used as a proxy for phytoplankton biomass
-
Absorption by CDOM (Colored Dissolved Organic Matter), the attenuation of light in water caused by dissolved organic compounds
Field sampling is conducted as close in time as possible to the satellite overpass to ensure consistency between satellite observations and in situ conditions. This timing is critical in dynamic coastal waters, where water quality can change rapidly. The in-situ measurements are used in two key ways. First, they are used to train the water quality models, allowing the satellite reflectance to be linked to actual water conditions. Second, they are used for independent validation, where satellite-derived estimates are compared against field observations to assess model accuracy and reliability.
Water quality modeling: SatBuoy product uses Mixture Density Network (MDN) models to retrieve water quality parameters from satellite reflectance. MDNs are machine learning models that represent complex, nonlinear relationships between satellite observations and water properties. Unlike traditional regression models that produce a single value, MDNs estimate a probability distribution for each prediction. SatBuoy uses a one-to-one modeling strategy, meaning a separate MDN model is trained for each parameter (turbidity, TSS, Secchi depth, chlorophyll-a, and aCDOM). Therefore, each model is optimized specifically for the optical characteristics of that parameter. Model validation: All MDN models used in SatBuoy are validated using independent satellite–in situ matchups, where satellite-derived estimates are directly compared with field measurements collected under comparable environmental conditions. Model performance was evaluated using standard statistical metrics that quantify accuracy, relative error, and systematic bias. The model validation results are shown below.
3) Product generation
Once trained and validated, the MDN models are applied to the full archive of S3CARD-processed Sentinel-3 images. This step generates spatially continuous maps of each water-quality parameter. Quality flags are used to exclude unreliable pixels, such as those affected by clouds or extreme glint conditions
4) Virtual Buoy Data Extraction
At each virtual buoy station:
-
Water-quality values are extracted from a small group of surrounding pixels
-
Only high-quality observations are retained
-
Values are aggregated into a continuous time series
This method reduces pixel-level noise and ensures representative measurements for each location.
References
Paulino, R. S., Martins, V. S., Caballero, C. B., Lima, T. M., Maciel, D. A., Santos, J. C., & Liu, B. (2025). Performance of glint correction algorithms for Sentinel-3 OLCI data. Frontiers in Remote Sensing, 6, 1690337.
Caballero, C. B., Martins, V. S., Paulino, R. S., Lima, T. M., Butler, E., & Sparks, E. (2025). Sentinel-3 Coastal Analysis Ready Data (S3CARD): An operational framework for coastal water applications. Water Research, 124432.
Key description from SatBuoy Explorer
SatBuoy Explorer provides time series estimates of water quality parameters observed over multiple virtual stations (spaced 5 km from each other) across Mississippi Sound. These parameters include:
-
Chlorophyll-a (mg m⁻³): a proxy for phytoplankton biomass
-
TSS (Total Suspended Solids, mg L⁻¹): concentration of suspended organic and inorganic particles
-
Turbidity (FNU): water cloudiness related to suspended particles
-
Secchi depth (cm): an indicator of water clarity
-
aCDOM (absorption coefficient of Colored Dissolved Organic Matter at 440 nm, m⁻¹): a proxy for dissolved organic particles in the water column
A quick overview of the SatBuoy Explorer is provided below:
On the SatBuoy Explorer, water quality parameter values are classified into three anomaly conditions: Normal, Moderate Anomaly, and Severe Anomaly. These levels help users quickly identify reliable values, potentially suspicious observations, and uncommon events in the Mississippi Sound. Anomaly conditions are defined by comparing current parameter values with the corresponding historical monthly average (2016 - now) for each station.
🟢 Normal: Current values fall within ±1 standard deviation of the historical average for the parameter. Values classified as Normal are considered reliable and representative of typical conditions in the region, showing minimal deviation from the historical records.
🟡 Moderate Anomaly: Current values fall between ±1 and ±2 standard deviations from the historical average for the parameter. Values classified as Moderate Anomaly indicate moderate changes in water quality dynamics relative to the historical records. These deviations may reflect real environmental variability or potential uncertainties related to image processing and should therefore be interpreted with caution.
🔴 Severe Anomaly: Current values exceed ±2 standard deviations from the historical average for the parameter. Values classified as Severe Anomaly indicate strong deviations from typical water quality conditions. These anomalies may be associated with significant environmental stressors or processing uncertainties and warrant closer investigation and continued monitoring.
The uncertainties displayed in the SatBuoy Explorer are based on validation against in situ measurements and are quantified using the Root Mean Square Error (RMSE) metric. See validation for details.
