A spatially-informed interpretable deep learning framework for high-resolution nutrient monitoring in complex coastal waters
Bibliographic record
Abstract
Satellite monitoring of non-optically active nutrients remains an ongoing challenge due to their weak spectral signals. While artificial intelligence (AI) methods can improve estimates by incorporating spatiotemporal information, their dependence on simple geographic coordinates tends to oversimplify the complex dynamics of nutrient distribution in coastal waters, especially in data-sparse regions. To address this gap, we propose the Spatiotemporal Kernel Density Estimation Deep Neural Network (ST-KDE-DNN). This model leverages precomputed spatial influence fields, derived from river outlet locations using KDE, as direct inputs to establish a source-driven spatial prior that reflects complex mixing patterns. We applied the ST-KDE-DNN framework to retrieve Dissolved Inorganic Nitrogen (DIN) and Phosphorus (DIP) from Landsat-8/9 imagery in the Pearl River Estuary (PRE), a typical multi-outlet coastal system. Compared with models incorporating standard spatiotemporal predictors, incorporating the KDE achieved significantly higher precision, with a reduced Root Mean Square Error (RMSE) of 45% for DIN and 29% for DIP, respectively. The most prominent improvements occurred in complex mixing zones where river plumes overlap. Furthermore, Explainable AI (XAI) analysis confirmed that the KDE features were dominant predictors, which enhances model transparency and grounds its predictions in physical processes. Based on this framework, we generated high-resolution time series of DIN and DIP for the PRE, and further revealed their seasonal and inter-annual variabilities. The ST-KDE-DNN offers a scalable, physically-informed, and interpretable solution for advancing nutrient monitoring in complex estuarine and coastal systems worldwide.
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 | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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".