Mixture density networks for re-constructing historical ocean-color products over inland and coastal waters: demonstration and validation
Bibliographic record
Abstract
Ocean color remote sensing tracks water quality globally, but multispectral ocean color sensors often struggle with complex coastal and inland waters. Traditional models have difficulty capturing detailed relationships between remote sensing reflectance ( R rs ), biogeochemical properties (BPs), and inherent optical properties (IOPs) in these complex water bodies. We developed a robust Mixture Density Network (MDN) model to retrieve 10 relevant biogeochemical and optical variables from heritage multispectral ocean color missions. These variables include chlorophyll-a ( Chla ) and total suspended solids ( TSS ), as well as the absorbing components of IOPs at their reference wavelengths. The heritage missions include the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard Aqua and Terra, the Environmental Satellite (Envisat) Medium Resolution Imaging Spectrometer (MERIS), and the Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership (Suomi NPP). Our model is trained and tested on all available in situ spectra from an augmented version of the GLObal Reflectance community dataset for Imaging and optical sensing of Aquatic environments (GLORIA) (N = 9,956) after having added globally distributed in situ IOP measurements. Our model is validated on satellite match-ups corresponding to the SeaWiFS Bio-optical Archive and Storage System (SeaBASS) database. For both training and validation, the hyperspectral in situ radiometric and absorption datasets were resampled via the relative spectral response functions of MODIS, MERIS, and VIIRS to simulate the response of each multispectral ocean color mission. Using hold-out (80–20 split) and leave-one-out testing methods, the retrieved parameters exhibited variable uncertainty represented by the Median Symmetric Residual ( MdSR ) for each parameter and sensor combination. The median MdSR over all 10 variables for the hold-out testing method was 25.9%, 24.5%, and 28.9% for MODIS, MERIS, and VIIRS, respectively. TSS was the parameter with the highest MdSR for all three sensors (MODIS, VIIRS, and MERIS). The developed MDN was applied to satellite-derived R rs products to practically validate their quality via the SeaBASS dataset. The median MdSR from all estimated variables for each sensor from the matchup analysis is 63.21% for MODIS/A, 63.15% for MODIS/T, 60.45% for MERIS, and 75.19% for VIIRS. We found that the MDN model is sensitive to the instrument noise and uncertainties from atmospheric correction present in multispectral satellite-derived R rs . The overall performance of the MDN model presented here was also analyzed qualitatively for near-simultaneous images of MODIS/A and VIIRS as well as MODIS/T and MERIS to understand and demonstrate the product resemblance and discrepancies in retrieved variables. The developed MDN is shown to be capable of robustly retrieving 10 water quality variables for monitoring coastal and inland waters from multiple multispectral satellite sensors (MODIS, MERIS, and VIIRS).
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.007 | 0.012 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".