A prediction model for 137Cs in surface seawater of China’s coastal waters based on AI oceanography method
- 1Nanjing University of Information Science and Technology, Nanjing, China
- 2Third Institute of Oceanography, Ministry of Natural Resources, Xiamen, China
Abstract
A hybrid framework named “TRR–LSTM–TRR” is developed to reconstruct and predict 137Cs activity concentrations in surface seawater of China’s coastal seas. The model integrates a bound-constrained Trust-Region Reflective (TRR) exponential baseline, an LSTM-based residual-learning network, and a physically constrained exponential refit. Historical observations from the Bohai Sea, Yellow Sea, East China Sea, and South China Sea during 1960–2017 were compiled (833 station records after data-quality corrections). The reconstructed environmentally effective half-lives were 10.52, 14.84, 15.05, and 16.40 a for the four seas, respectively, indicating the fastest attenuation in the Bohai Sea and the slowest in the South China Sea. The residual correction reduced absolute prediction errors in all basins; under the temporal holdout (training on 1960–2010, testing on 2011–2017) mean absolute errors were 0.30–0.61 Bq m−3. Because the recent records are few and low-valued, long-term projections are reported with 90% ensemble prediction intervals rather than as point estimates. Taking 0.2 Bq m−3 as the detection limit, the Bohai, East China, Yellow, and South China Seas are expected to fall below this threshold around 2036, 2049, 2053, and 2055, respectively.
Related
- Prioritization of contaminants of emerging concern in European surface waters using occurrence, exposure, prediction, and regulatory status perspectives
- Statistical Oceanography of Profiling Floats and Surface Drifters
- Optical mediation of chlorophyll-a trends and radiative heating in coastal waters
- Impacts of simulated coastal ocean alkalinity enhancement on the seasonal carbon cycle in European waters under a low- and a high-emission scenario
- Steady-state semi-analytical solutions for assessing the shape of the freshwater–seawater interface induced by a finite-sized underground physical barrier and a pumping well in a sloping coastal aquifer
BibTeX
@article{yu2026prediction,
title = {{A prediction model for 137Cs in surface seawater of China’s coastal waters based on AI oceanography method}},
author = {Zixin Yu and Wu Men and Fenfen Wang},
journal = {Frontiers in Marine Science},
year = {2026},
month = oct,
doi = {10.3389/fmars.2026.1924778},
url = {https://doi.org/10.3389/fmars.2026.1924778},
note = {CC BY},
}
RIS
TY - JOUR
TI - A prediction model for 137Cs in surface seawater of China’s coastal waters based on AI oceanography method
AU - Zixin Yu
AU - Wu Men
AU - Fenfen Wang
PY - 2026
DA - 2026/10/26
JO - Frontiers in Marine Science
DO - 10.3389/fmars.2026.1924778
UR - https://doi.org/10.3389/fmars.2026.1924778
AB - A hybrid framework named “TRR–LSTM–TRR” is developed to reconstruct and predict 137Cs activity concentrations in surface seawater of China’s coastal seas. The model integrates a bound-constrained Trust-Region Reflective (TRR) exponential baseline, an LSTM-based residual-learning network, and a physically constrained exponential refit. Historical observations from the Bohai Sea, Yellow Sea, East China Sea, and South China Sea during 1960–2017 were compiled (833 station records after data-quality corrections). The reconstructed environmentally effective half-lives were 10.52, 14.84, 15.05, and 16.40 a for the four seas, respectively, indicating the fastest attenuation in the Bohai Sea and the slowest in the South China Sea. The residual correction reduced absolute prediction errors in all basins; under the temporal holdout (training on 1960–2010, testing on 2011–2017) mean absolute errors were 0.30–0.61 Bq m−3. Because the recent records are few and low-valued, long-term projections are reported with 90% ensemble prediction intervals rather than as point estimates. Taking 0.2 Bq m−3 as the detection limit, the Bohai, East China, Yellow, and South China Seas are expected to fall below this threshold around 2036, 2049, 2053, and 2055, respectively.
SN - 2296-7745
ER -