Robust prediction of chlorophyll-A from nitrogen and phosphorus content in Philippine and global lakes using fine-tuned, explainable machine learning
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
• Chl-a predicted from lake nutrients in Philippines, Japan, USA, Canada, Uganda. • Best models are KRR (92.5%) for Laguna Lake and GPR (82.05%) for global lakes. • KRR, SVR, GPR, and MLP are robust to sensor noise but kNN, RF, and GBR were not. • In both case studies, phosphorus content has the most impact on Chl-a predictions. • Fine-tuned, explainable, and robust models can be trusted more to drive policy. Chlorophyll-a (Chl-a) content in waterbodies is a primary indicator of algal biomass and is used to detect impending harmful algal blooms. This paper presents a methodology using 8 popular machine learning (ML) models for estimating Chl-a concentration from nutrient content in lakes. Different from previous works, we introduce 3 novel steps: (i) the use of Bayesian optimization for fine-tuning ML hyper-parameters to improve performance; (ii) the use of explainability methods to understand the most influential inputs to Chl-a prediction; and (iii) the use of robustness analysis to assess how models are affected by measurement noise. Two case studies were used to test our approach: Laguna Lake, Philippines, and various lakes from Japan, the United States of America, Canada, and Uganda. We found that fine-tuned Kernel Ridge Regression and Gaussian Process Regression are consistently the most accurate (>80%) and robust models in both case studies. In Laguna Lake, Shapley explanations revealed that phosphate and nitrate ions are the most important predictors of Chl-a, while total phosphorus is that for global lakes. Hence, these parameters are suggested to be monitored more closely for detecting algal blooms. By making our codes accessible, we hope that our methods can serve as a benchmark for the data-driven modeling of Chl-a content in lakes, and aid in their management through model deployment.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 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 it