Boundary‐Informed, Deep‐Learning Reconstructions of ITF Heat and Salt Transports

The Indonesian seas are among the most dynamically complex regions of the global ocean, where intricate bathymetry and intense water‐mass exchanges give rise to pronounced multiscale variability. Understanding the variability of the Indonesian Throughflow (ITF) and its climatic impacts is therefore critical. However, long‐term observations within the interior seas remain rather sparse. In this study, we develop and evaluate an attention‐enhanced, multi‐branch, residual neural network (AM‐ResNet) to achieve daily, full‐depth reconstructions of temperature, salinity, and velocity across six key ITF straits, enabling accurate diagnostics of strait‐integrated heat and salt fluxes. Post hoc attribution analysis based on SHapley Additive exPlanations (SHAP) shows that the influence of surrounding water masses on interior passage flows can be conveyed through thermohaline signals, with the North Pacific and Indian Ocean boundaries exerting the strongest control. Extending the attribution into both the frequency and lag domains further reveals a pronounced dependence of this regulation on temporal scale and propagation time. Notably, retaining only the dynamically dominant boundary branches preserves and even enhances predictive skill, indicating that predictive skill depends more on selecting physically relevant pathways than on increasing input dimensionality. Overall, this study establishes a feasible framework for inferring ITF variability from dynamically connected boundary information, while providing new insights into how remote boundary signals regulate the multiscale exchange of mass, heat, and salt between the Pacific and Indian Oceans.

Authors

Institutions

Publication Details

Journal
Journal of Geophysical Research Machine Learning and Computation
Published
2026-11-19
DOI
https://doi.org/10.1029/2026jh001405
Primary Topic
Oceanographic and Atmospheric Processes
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Boundary‐Informed, Deep‐Learning Reconstructions of ITF Heat and Salt Transports

Huijie Xue, Yuan Wang, Ruijun Zhu
Journal of Geophysical Research Machine Learning and Computation
Oceanographic and Atmospheric Processes
article

Boundary‐Informed, Deep‐Learning Reconstructions of ITF Heat and Salt Transports

Huijie Xue, Yuan Wang, Ruijun Zhu
article en

Abstract

The Indonesian seas are among the most dynamically complex regions of the global ocean, where intricate bathymetry and intense water‐mass exchanges give rise to pronounced multiscale variability. Understanding the variability of the Indonesian Throughflow (ITF) and its climatic impacts is therefore critical. However, long‐term observations within the interior seas remain rather sparse. In this study, we develop and evaluate an attention‐enhanced, multi‐branch, residual neural network (AM‐ResNet) to achieve daily, full‐depth reconstructions of temperature, salinity, and velocity across six key ITF straits, enabling accurate diagnostics of strait‐integrated heat and salt fluxes. Post hoc attribution analysis based on SHapley Additive exPlanations (SHAP) shows that the influence of surrounding water masses on interior passage flows can be conveyed through thermohaline signals, with the North Pacific and Indian Ocean boundaries exerting the strongest control. Extending the attribution into both the frequency and lag domains further reveals a pronounced dependence of this regulation on temporal scale and propagation time. Notably, retaining only the dynamically dominant boundary branches preserves and even enhances predictive skill, indicating that predictive skill depends more on selecting physically relevant pathways than on increasing input dimensionality. Overall, this study establishes a feasible framework for inferring ITF variability from dynamically connected boundary information, while providing new insights into how remote boundary signals regulate the multiscale exchange of mass, heat, and salt between the Pacific and Indian Oceans.

Journal of Geophysical Research Machine Learning and ComputationVol. 3(6)
National University of Defense Technology (CN), Xiamen University (CN), State Key Laboratory of Marine Environmental Science (CN)
Oceanographic and Atmospheric Processes
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon