Quantifying the relative contributions of forcings to the variability of estuarine surface suspended sediments using a machine learning framework
Tavora, J.; El Hourany, R.; Fernandes, E.H.; Jalón-Rojas, I.; Sottolichio, A.; Salama, M.S.; van der Wal, D. (2025). Quantifying the relative contributions of forcings to the variability of estuarine surface suspended sediments using a machine learning framework. Cont. Shelf Res. 287: 105429. https://dx.doi.org/10.1016/j.csr.2025.105429
In: Continental Shelf Research. Pergamon Press: Oxford; New York. ISSN 0278-4343; e-ISSN 1873-6955, more
|   |
| Authors | | Top |
- Tavora, J.
- El Hourany, R.
- Fernandes, E.H.
- Jalón-Rojas, I.
|
- Sottolichio, A.
- Salama, M.S.
- van der Wal, D., more
|
|
| Abstract |
The influence of forcing mechanisms on the variability of suspended sediments in an estuary is, for the first time, synoptically quantified over prevailing ('normal') conditions and extreme events. This study investigates the complex and non-linear influence of tides, river discharge, and winds on the variability of suspended sediments in the macrotidal Gironde Estuary, France. Employing a machine learning-based framework, we integrated high-frequency field data, hourly numerical modeling outputs, and semi-daily satellite remote sensing to spatially quantify the relative contributions of forcing mechanisms. Our results reveal that tides are the primary driver of sediment variability (42.3–58.9%), followed by river discharge (21.2–34.7%) and wind (8.7–16.9%). Uncertainties range between 7% and 13.6%. In addition, the spatial variability of their contributions is consistent across numerical modeling and satellite remote sensing data, with differences not exceeding 10%. However, satellite data is limited by cloud cover and may miss extreme events. In contrast, hourly numerical modeling indicates tides are the dominant forcing mechanism under extreme events significantly affecting suspended sediment variability in the estuary. This study verifies the effectiveness of our machine learning approach against traditional Singular Spectral Analysis using field data. We demonstrate that machine learning techniques can effectively synthesize spatial distribution patterns of hydrodynamic and sedimentological variability, including the influence of winds. Our findings highlight not only the potential of satellite observations to analyze prevailing conditions despite data gaps but also that with hourly numerical modeling, the impact of forcings can be synoptically quantified under prevailing ('normal') conditions and extreme events. |
|
