Pith. sign in

REVIEW 4 major objections 5 minor 57 references

Data-driven Lake Water Quality Forecasting for Time Series with Missing Data using Machine Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A joint feasibility rule shows that lake water clarity forecasts can match full-history accuracy within 5% using about 64 recent samples and a single predictor per lake.

desk verdict Honest applied ML case study whose headline '64 samples, 1 predictor' rule is an in-sample fitted summary, not a validated out-of-sample result, because imputation and configuration selection leak test information. read the letter →

arxiv 2601.15503 v2 pith:Y6ZOWCNJ submitted 2026-01-21 cs.LG

Rishit Chatterjee , Tahiya Chowdhury This is my paper · ORCID
classification cs.LG
keywords lakewaterqualitySecchidiskdepthtime-seriesforecastingmissingdataimputationMICEridgeregressionsamplesizedeterminationfeatureselection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks how little data and how few measurements a lake water-quality forecaster needs while staying within 5% of the accuracy of a model trained on full history and all available features. Using three decades of volunteer-collected Secchi disk depth records from 30 Maine lakes, the authors impute missing covariates with MICE, pick ridge regression as the best of six forecasters, and then search over training-window length and feature count together. They report that a typical lake needs only about 64 of its most recent observations and one predictor—most often the oxic status variable—to meet the target. If this holds, monitoring programs could dramatically cut sampling effort and still produce reliable clarity forecasts.

What carries the argument

The central object is the joint feasibility function f(n,k), which formalizes 'good enough' as staying within 5% of a lake-specific reference nMAE achieved by ridge regression on all pre-test data and all features. The procedure combines MICE for missing covariates, a backward recent-history training protocol that predicts a held-out last-five-years test block, MDI-based feature ranking followed by greedy forward selection, and a lexicographic rule that prioritizes minimal n then minimal k. This yields a parameter-free, per-lake minimally sufficient configuration and an actionable aggregate target for sampling effort and measurement priorities.

What would settle it

Re-run the same backward protocol but fit MICE only on the training fold and apply it to the test block, and select (n,k) using only validation data; if the median minimal configuration moves substantially above 64 samples or 1 predictor, the paper's central rule is an artifact of leakage.

Watch

Extended reading notes

Core claim

The paper introduces a joint feasibility function f(n,k) that is 1 when a ridge forecaster trained on the most recent n pre-test observations with the top-k ranked predictors achieves test nMAE within a 5% tolerance of the full-data, full-feature reference. Taking the lexicographically smallest feasible (n,k) per lake, the authors find a median minimal configuration of 64 recent samples and 1 predictor, with OXIC selected in 90% of the 20 lakes where a single predictor sufficed. Averaged over lakes, the minimal sample count alone is about 176, and a four-feature subset (TPEC, OXIC, SCHMIDT, CONDUCT) matches the thirteen-feature baseline within tolerance.

Load-bearing premise

The evaluation protocol is leakage-free: the MICE imputer is fit on the full covariate matrix including test-block and future records, and the minimal n and k are selected using test-block errors that also define the 5% target, so if test information seeps into training the reported '64 samples, 1 predictor' rule may not hold out of sample.

Editorial extensions

If this is right

  • Volunteer monitoring programs could focus effort on roughly 60 recent observations per lake and a single priority measurement (e.g., oxic status) rather than maintaining full-history, thirteen-feature records.
  • The 5% tolerance can be tuned: a looser tolerance would shrink the required sample count and feature set, while a stricter tolerance would grow them, giving program managers a direct accuracy-versus-effort dial.
  • The per-lake feasibility output (hat-n, hat-k) translates directly into operational targets, so agencies can reallocate staff time toward instrumentation or targeted lab campaigns.
  • If the result generalizes beyond the 30 data-rich lakes, the same joint rule could guide monitoring design for data-poor lakes where remote sensing is unreliable.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The 64-samples/1-predictor result plausibly reflects diminishing returns in learning curves: after a few seasonal cycles, additional history adds little signal for ridge regression, so the finding may hold qualitatively even for other lakes, though the exact numbers likely shift.
  • A natural testable extension would be to apply the same joint feasibility procedure to other targets (e.g., chlorophyll) and other imputation methods to see whether the minimal configuration is robust or an artifact of MICE's assumptions.
  • The paper's own flagged future work — a leakage-safe nested evaluation — suggests the central numbers should be treated as best-case estimates until the imputation and model-selection pipeline is re-validated without test-block information.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies Secchi Disk Depth (SDD) forecasting for 30 data-rich Maine lakes with irregular, missing multivariate time series. Missingness is handled by MICE imputation, and among six candidate forecasters ridge regression is selected as best. The authors define a normalized MAE (nMAE), then quantify a minimal sample size n* (mean approximately 176 under a recent-history backward protocol), a minimal feature set (four features), and introduce a joint feasibility function f(n,k) that selects, per lake, the lexicographically smallest pair (n,k) staying within 5% of the full-history, full-feature reference nMAE. The headline result is a median minimally sufficient configuration of n=64 recent samples and k=1 predictor, with OXIC dominant. The paper is transparent about its limitations and lists a leakage-safe nested evaluation as future work, but the current protocol fits MICE on the full covariate matrix (including the test block) and selects n,k using test-block errors, so the empirical claims are not out-of-sample validated.

Significance. If the results survive a leakage-free evaluation, the proposed joint feasibility rule would be a practically useful tool for lake-monitoring design: it unifies sample-size and feature-selection decisions under a single, interpretable accuracy target, and it addresses a real gap in the water-quality forecasting literature. The paper also contributes a scale-normalized nMAE metric that is appropriate for cross-lake comparison, and it openly acknowledges several of its own weaknesses. However, the central numeric claims (especially n=64 and k=1) currently rest on an evaluation protocol that can leak test information into both imputation and configuration selection, so the contribution is not yet demonstrated.

major comments (4)
  1. [II-B and II-F; Future Works (e)] The evaluation is not leakage-safe. MICE is fit on the full covariate matrix Z_l, which includes the held-out test block and future records, so imputed covariate values in the training window can encode test-period structure. In addition, n* and k* are defined by thresholding nMAE computed on the very same test block used to set the 5% tolerance (Sec. II-D and II-F), so the minimal configuration is selected on the test set. The paper's Future Works (e) explicitly concedes that a leakage-safe nested evaluation is still needed. Until that is run, the headline numbers (n*~176, median n=64, k=1) cannot be treated as valid out-of-sample estimates.
  2. [II-E] The feature-selection loop also selects on the test set. The MDI importance ranking is computed and then, for each k, the ridge forecaster is evaluated by nMAE on the test block T_test to pick k*_l. This is a second channel of test-set adaptation beyond the imputation leakage. Moreover, the paper does not state whether MDI comes from a RandomForest or a Ridge model, and no evidence is given that MDI importances transfer to ridge regression. The authors should either derive the ranking from ridge coefficients or perform a nested selection where the ranking and k are chosen on validation data only.
  3. [II-F] The minimally sufficient configuration depends on the lexicographic tie-break, which minimizes n first and only then k. This is an arbitrary design choice that can substantially change the reported n and k; for example, a Pareto-minimal or cost-weighted rule might choose a slightly larger n to achieve k=1. The paper provides no sensitivity analysis for the tolerance (5%) or the tie-break, and lists such an analysis only as future work (d). Without it, the headline 64 samples and 1 predictor is not robust.
  4. [III-a,d] The manuscript uses two different minimal quantities without clearly distinguishing them: n*~176 is the mean minimal sample count for the full-feature model, while n=64 is the median joint n when k is also minimized. The abstract highlights the latter but the former appears in the results. Moreover, no confidence intervals or per-lake distributions are provided for n and k, and 10 of 30 lakes revert to the full-data/full-feature configuration. Such high variability and a small convenience sample of 30 record-rich lakes do not support a strong general per-lake rule.
minor comments (5)
  1. [Abstract / III] Clarify that n*~176 is the mean minimal count for the full-feature model, while n=64 is the median joint result; otherwise readers may conflate the two.
  2. [Fig. 3] The y-axis label 'percentage of lakes' should state the denominator is the 20 lakes with k=1, not all 30 lakes.
  3. [II-E] Specify which model produces the MDI importances. The text mentions 'Reference Ridge model' and then 'MDI importances,' which are associated with tree-based models; this needs disambiguation.
  4. [II-D] The nMAE formula is not numbered and the notation 'nMAE_final' is not used consistently; use a single notation for the full-history reference.
  5. [General] There are minor typos such as 'T est' and inconsistent capitalization of 'Secchi Disk Depth' vs 'Secchi disk depth'; please polish the text.

Circularity Check

2 steps flagged · score 6.0 of 10

Headline '64 samples, 1 predictor' is the self-selected argmin of a feasibility function evaluated on the same test errors that define the 5% target; leakage-safe evaluation is deferred to future work.

  1. self definitional [Section II-F (Joint selection), Eqs. for f(n,k) and lexmin; reported in Section III-d]
    "We then define the feasibility function as, f(n, k) = ( 1, if nMAE(n, k) ≤ τ; 0, otherwise, ) where τ = 1.05 nMAE full. ... We pick the lexicographically smallest pair—i.e., minimize n first and, for ties, minimize k: (ˆnℓ, ˆkℓ) = lexmin (n,k)∈Fℓ (n, k)"

    The headline 'minimally sufficient configuration has median ˆn=64 observations ... and median ˆk=1' is literally the lexicographic argmin of an indicator defined as nMAE(n,k) ≤ 1.05 nMAE_full, where nMAE_full is the model's own test-block nMAE. By construction, any reported (n,k) is a point at which the model's own test nMAE is within 5% of its own full-history nMAE; no independent reference or external error target is used. Thus the statement that '64 recent samples and one predictor suffice for 5% accuracy' restates the selection criterion plus the fitted learning curves rather than an independent prediction that could fail the defining test. The empirical content is only the location of the argmin, i.e., a fitted summary of the same test errors.

  2. fitted input called prediction [Section II-B/II-C (imputation and backward protocol), Section V Future Works (e)]
    "(e) adopt a leakage-safe, nested train–validation–test evaluation in which imputation is fit on the training fold only and applied to validation-test, with model-threshold choices made on validation and final metrics reported on the held-out test set."

    The current pipeline completes the covariate matrix by MICE before the train/test split (Sec. II-B: 'we complete the covariate series by imputation'), so imputed training features can encode test-block and future information. The joint rule then selects (n,k) by evaluating nMAE(n,k) on the same held-out test block that defines τ=1.05 nMAE_full. The paper's own future-work item acknowledges this protocol is not leakage-safe. Consequently, the 'minimal configuration' is chosen to pass the very test errors that are then reported as the 5% target; the '64 samples, 1 predictor' rule is an adaptive in-sample selection presented as a predictive sampling rule, which is the fitted-input-called-prediction pattern.

full rationale

The paper's model comparison and learning-curve traces are genuine empirical measurements: ridge is evaluated on held-out last-five-year blocks, and the MDI rankings are computed per lake. The circularity is concentrated in how the central minimal-resource numbers are produced. The minimal sample count n* and minimal feature count k* are defined as the first grid points at which the model's own test nMAE falls within 5% of its own full-history nMAE on the same test block; the joint procedure then returns the lexicographically smallest feasible (n,k). Reporting 'median n=64, median k=1' as 'meeting the 5% accuracy target required...' is therefore a restatement of the definition plus the fitted learning curves, not an independent prediction that the procedure could have failed. This is the self-definitional pattern. The concern is amplified by the paper's own future-work admission that imputation is fit on the full covariate matrix and model-threshold choices are made on the test block, so the selected configuration is chosen to pass the very errors used to judge it; a leakage-safe nested evaluation is not yet done. Because the learning curves and ridge-vs-baseline comparisons retain independent empirical content, the paper is not wholly circular, but the headline '64 samples, 1 predictor' rule is a fitted input presented as a predictive rule.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central numerical claims rest on several choices the paper makes without independent evidence: a 5% tolerance, a 5-year test block, a lexicographic tie-break, unspecified MICE and ridge settings, and a leakage-prone full-data imputation. The MAR assumption and the transferability of MDI importances to ridge are additional unverified premises.

free parameters (5)
  • accuracy tolerance (5%) = 1.05 × nMAE_full
    The feasibility target tau is set to 5% above the full-history, full-feature test error; all reported minimal sample/feature counts are relative to this arbitrary threshold and would change with tolerance.
  • recent-history test block length = 5 years
    The held-out test block spans the last 5 years; nMAE_final and every learning curve depend on this choice.
  • lexicographic tie-break = minimize n first, then k
    When multiple (n,k) meet the 5% target, the rule reports the smallest n; this drives the median n=64, k=1 outcome.
  • ridge regularization alpha = not reported (library default assumed)
    Same hyperparameters used throughout but never stated; the 'ridge is best' comparison and learning curves depend on it.
  • MICE imputation settings = not reported (m, iterations, predictor matrix)
    Imputation quality determines covariate values and thus forecasts; settings are unspecified.
assumptions (5)
  • domain assumption Missingness is Missing At Random (MAR)
    MICE is valid under MAR; the paper explicitly states this assumption. Missingness from ice cover, weather access, and human error may be MNAR, which would bias imputations. (Section II-B)
  • ad hoc to paper Full-data MICE imputation before forecasting does not leak test information into training
    They impute the entire covariate matrix, including test-block and future timestamps, before splitting; the paper's Future Work item (e) acknowledges this is leakage-unsafe. (Sections II-B, V)
  • ad hoc to paper The 30 most data-rich lakes represent the general lake-monitoring setting
    The top-30 selection by completeness is used for all conclusions; the authors note results may be skewed toward long-record lakes. (Section II-A, Limitations)
  • domain assumption nMAE = MAE / test-set mean SDD is a valid cross-lake comparability metric
    Normalizing by the test set mean assumes the error scales with typical clarity; this is a modeling choice that affects all reported numbers. (Section II-D)
  • ad hoc to paper MDI feature importances are transferable to ridge regression
    MDI is defined for tree ensembles, but the paper computes MDI importances and then applies them to rank features for the ridge forecaster, without justifying transferability. (Section II-E)

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-driven Lake Water Quality Forecasting for Time Series with Missing Data using Machine Learning." pith.science (2026). https://pith.science/paper/Y6ZOWCNJ

@misc{pith2026260115503,
  author       = {Pith},
  title        = {Pith review of: Data-driven Lake Water Quality Forecasting for Time Series with Missing Data using Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y6ZOWCNJ}},
  note         = {Machine review of arXiv:2601.15503}
}
read the original abstract

Volunteer-led lake monitoring yields irregular, seasonal time series with many gaps arising from ice cover, weather-related access constraints, and occasional human errors, complicating forecasting and early warning of harmful algal blooms. We study Secchi Disk Depth (SDD) forecasting on a 30-lake, data-rich subset drawn from three decades of in-situ records collected across Maine lakes. Missingness is handled via Multiple Imputation by Chained Equations (MICE), and we evaluate performance with a normalized Mean Absolute Error (nMAE) metric for cross-lake comparability. Among six candidates, ridge regression provides the best mean test performance. Using ridge regression, we then quantify the minimal sample size, showing that under a backward, recent-history protocol, the model reaches within 5% of full-history accuracy with approximately 176 training samples per lake on average. We also identify a minimal feature set, where a compact four-feature subset matches the thirteen-feature baseline within the same 5% tolerance. Bringing these results together, we introduce a joint feasibility function that identifies the minimal training history and fewest predictors sufficient to achieve the target of staying within 5% of the complete-history, full-feature baseline. In our study, meeting the 5% accuracy target required about 64 recent samples and just one predictor per lake, highlighting the practicality of targeted monitoring. Hence, our joint feasibility strategy unifies recent-history length and feature choice under a fixed accuracy target, yielding a simple, efficient rule for setting sampling effort and measurement priorities for lake researchers.

Figures

Figures reproduced from arXiv: 2601.15503 by the authors.

Figure 1
Figure 1. Cochnewagon Pond: observed and ridge-predicted SDD over the held-out last five years. The x-axis represents chronological sampling dates and [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. (a) Minimal samples and (b) features for forecasting water quality [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Bar height (y-axis) gives the percentage of lakes selecting each [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 1 canonical work pages

  1. [1]

    Securing water as a resource for society: An ecosystem services perspective,

    C. Liquete, J. Maes, A. La Notte, and G. Bidoglio, “Securing water as a resource for society: An ecosystem services perspective,”Ecohydrology & Hydrobiology, vol. 11, no. 3, pp. 247–259, 2011

    2011

  2. [2]

    Progress in understanding harmful algal blooms: Paradigm shifts and new tech- nologies for research, monitoring, and management,

    D. M. Anderson, A. D. Cembella, and G. M. Hallegraeff, “Progress in understanding harmful algal blooms: Paradigm shifts and new tech- nologies for research, monitoring, and management,”Annual Review of Marine Science, vol. 4, pp. 143–176, 2012

    2012

  3. [3]

    Cyanobacterial harmful algal blooms in aquatic ecosystems: A comprehensive outlook on current and emerging mitigation and control approaches,

    A. Sukenik and A. Kaplan, “Cyanobacterial harmful algal blooms in aquatic ecosystems: A comprehensive outlook on current and emerging mitigation and control approaches,”Microorganisms, vol. 9, no. 7, 2021, art. no. 1472

    2021

  4. [4]

    Increasingly severe cyanobacterial blooms and deep water hypoxia coincide with warming water temperatures in reservoirs,

    N.J.Smucker,J.J.Beaulieu,C.T.Nietch,andJ.L.Young,“Increasingly severe cyanobacterial blooms and deep water hypoxia coincide with warming water temperatures in reservoirs,”Global Change Biology, vol. 27, no. 11, pp. 2507–2519, 2021

    2021

  5. [5]

    Trophic state assessment of global inland waters using a MODIS-derived Forel-Ule index,

    S. Wang, J. Li, B. Zhang, E. Spyrakos, A. N. Tyler, Q. Shen, F. Zhang, T. Kuster, M. K. Lehmann, Y. Wu, and D. Peng, “Trophic state assessment of global inland waters using a MODIS-derived Forel-Ule index,”Remote Sensing of Environment, vol. 217, pp. 444–460, 2018

    2018

  6. [6]

    The human factor: Weather bias in manual lake water quality monitoring,

    J. M. Rand, M. O. Nanko, M. B. Lykkegaard, D. Wain, W. King, L. D. Bryant, and A. J. Hunter, “The human factor: Weather bias in manual lake water quality monitoring,”Limnology and Oceanography: Methods, vol. 20, no. 5, pp. 288–303, 2022

    2022

  7. [7]

    Secchi disk depth estimation from water quality parame- ters: Artificial neural network versus multiple linear regression models?

    S. Heddam, “Secchi disk depth estimation from water quality parame- ters: Artificial neural network versus multiple linear regression models?” Environmental Processes, vol. 3, pp. 525–536, 2016

    2016

  8. [8]

    Improving remote sensing estimation of Secchi disk depth for global lakes and reservoirs using machine learning methods,

    Y. Zhang, K. Shi, X. Sun, Y. Zhang, N. Li, W. Wang, Y. Zhou, W. Zhi, M. Liu, Y. Li, G. Zhu, B. Qin, E. Jeppesen, J. Zhou, and H. Li, “Improving remote sensing estimation of Secchi disk depth for global lakes and reservoirs using machine learning methods,”GIScience & Remote Sensing, vol. 59, no. 1, pp. 1367–1383, 2022

    2022

Show all 57 references
  1. [9]

    A remote sensing and machine learning-based approach to forecast the onset of harmful algal bloom,

    M. Izadi, M. Sultan, R. E. Kadiri, A. Ghannadi, and K. Abdelmohsen, “A remote sensing and machine learning-based approach to forecast the onset of harmful algal bloom,”Remote Sensing, vol. 13, no. 19, 2021, art. no. 3863

    2021

  2. [10]

    Remote sensing for regional lake water quality assessment: Capabilities and limitations of current and upcoming satellite systems,

    L. G. Olmanson, P. L. Brezonik, and M. E. Bauer, “Remote sensing for regional lake water quality assessment: Capabilities and limitations of current and upcoming satellite systems,” inAdvances in Watershed Science and Assessment, ser. Handbook of Environmental Chemistry, T. Yo...

    2015

  3. [11]

    Monitoring inland water quality using remote sensing: potential and limitations of spectral indices, bio-optical simulations, machine learning, and cloud computing,

    V. Sagan, K. T. Peterson, M. Maimaitijiang, P. Sidike, J. Sloan, B. A. Greeling, S. Maalouf, and C. Adams, “Monitoring inland water quality using remote sensing: potential and limitations of spectral indices, bio-optical simulations, machine learning, and cloud computing,” Ear...

    2020

  4. [12]

    Remote sensing of inland waters: Challenges, progress and future directions,

    S. C. J. Palmer, T. Kutser, and P. D. Hunter, “Remote sensing of inland waters: Challenges, progress and future directions,”Remote Sensing of Environment, vol. 157, pp. 1–8, 2015

    2015

  5. [13]

    Lagos-us landsat: Remotely sensed water quality estimates for u.s. lakes over 4 ha from 1984 to 2020,

    P. J. Hanly, K. E. Webster, and P. A. Soranno, “Lagos-us landsat: Remotely sensed water quality estimates for u.s. lakes over 4 ha from 1984 to 2020,”Scientific Data, vol. 12, p. 1315, 2025. [Online]. Available: https://www.nature.com/articles/s41597-025-05600-w

    1984

  6. [14]

    Machine learning approach for water quality predictions based on multispectral satellite imageries,

    V. Anand, B. Oinam, and S. Wieprecht, “Machine learning approach for water quality predictions based on multispectral satellite imageries,” Ecological Informatics, vol. 84, p. 102868, 2024

    2024

  7. [15]

    Prediction of algal chlorophyll-a and water clarity in monsoon-region reservoir using machine learning approaches,

    M. Mamun, J.-J. Kim, M. A. Alam, and K.-G. An, “Prediction of algal chlorophyll-a and water clarity in monsoon-region reservoir using machine learning approaches,”Water, vol. 12, no. 1, 2020, art. no. 30

    2020

  8. [16]

    Forecasting water quality index in groundwater using artificial neural network,

    M. Kulisz, J. Kujawska, B. Przysucha, and W. Cel, “Forecasting water quality index in groundwater using artificial neural network,”Energies, vol. 14, no. 18, 2021, art. no. 5875

    2021

  9. [17]

    Water-quality data imputation with a high percentage of missing values: A machine learning approach,

    R. Rodríguez, M. Pastorini, L. Etcheverry, C. Chreties, M. Fossati, A. Castro, and A. Gorgoglione, “Water-quality data imputation with a high percentage of missing values: A machine learning approach,” Sustainability, vol. 13, no. 11, 2021, art. no. 6318

    2021

  10. [18]

    Review of automated time series forecasting pipelines,

    S. Meisenbacher, M. Turowski, K. Phipps, M. Rätz, D. Müller, V. Ha- genmeyer, and R. Mikut, “Review of automated time series forecasting pipelines,”WileyInterdisciplinaryReviews:DataMiningandKnowledge Discovery, vol. 12, no. 6, p. e1475, 2022

    2022

  11. [19]

    Effect of missing data on performance of learning algorithms for hydrologic predictions: Implications to an imputation technique,

    M. K. Gill, T. Asefa, Y. Kaheil, and M. McKee, “Effect of missing data on performance of learning algorithms for hydrologic predictions: Implications to an imputation technique,”Water Resources Research, vol. 43, p. W07416, 2007

    2007

  12. [20]

    Missing data and multiple imputation in clinical epidemiological research,

    A. B. Pedersen, E. M. Mikkelsen, D. Cronin-Fenton, N. R. Kristensen, T. M. Pham, L. Pedersen, and I. Petersen, “Missing data and multiple imputation in clinical epidemiological research,”Clinical Epidemiology, vol. 9, pp. 157–166, 2017

    2017

  13. [21]

    Recent progress on surface water quality models utilizing machine learning techniques,

    M. He, Q. Qian, X. Liu, J. Zhang, and J. Curry, “Recent progress on surface water quality models utilizing machine learning techniques,” Water, vol. 16, no. 24, 2024, art. no. 3616

    2024

  14. [22]

    Improving remote sensing estimation of secchi disk depth for global lakes and reservoirs using machine learning methods,

    Y. Zhang, K. Shi, X. Sun, Y. Zhang, N. Li, W. Wang, Y. Zhou, W. Zhi, M. Liu, Y. Li, B. Qinet al., “Improving remote sensing estimation of secchi disk depth for global lakes and reservoirs using machine learning methods,”GIScience & Remote Sensing, vol. 59, no. 1, pp. 1367–1383, 2022

    2022

  15. [23]

    Lagos-us landsat: Remotely sensed water quality estimates for u.s. lakes over 4 ha from 1984 to 2020,

    P. J. Hanly, K. E. Webster, and P. A. Soranno, “Lagos-us landsat: Remotely sensed water quality estimates for u.s. lakes over 4 ha from 1984 to 2020,”Scientific Data, vol. 12, p. 1315, 2025, data Descriptor

    1984

  16. [24]

    Remote sensing of lake water clarity: Performance and transferability of both historical algorithms and machine learning,

    H. J. Rubin, D. A. Lutz, B. G. Steele, K. L. Cottingham, K. C. Weathers, M. J. Ducey, M. Palace, K. M. Johnson, and J. W. Chipman, “Remote sensing of lake water clarity: Performance and transferability of both historical algorithms and machine learning,”Remote Sensing, vol. 13...

    2021

  17. [25]

    Deep learning- based remote sensing retrieval of inland water quality: A review,

    Z. Pang, Z. Zhou, J. Fu, W. Jiang, X. Qin, and M. Sun, “Deep learning- based remote sensing retrieval of inland water quality: A review,” Journal of Hydrology: Regional Studies, vol. 61, p. 102759, 2025

    2025

  18. [26]

    Feature-driven hybrid attention learning for accurate water quality prediction,

    X. Yao, Z. Xu, T. Ren, and X.-J. Zeng, “Feature-driven hybrid attention learning for accurate water quality prediction,”Expert Systems with Applications, vol. 276, p. 127160, 2025

    2025

  19. [27]

    Deep representation learning enables cross-basin water quality predic- tion under data-scarce conditions,

    Y. Zheng, X. Zhang, Y. Zhou, Y. Zhang, T. Zhang, and R. Farmani, “Deep representation learning enables cross-basin water quality predic- tion under data-scarce conditions,”npj Clean Water, vol. 8, p. 33, 2025

    2025

  20. [28]

    Sample size determination for prediction models via learning-type curves,

    A. Dayimu, N. Simidjievski, N. Demiris, and J. Abraham, “Sample size determination for prediction models via learning-type curves,”Statistics in Medicine, vol. 43, no. 16, pp. 3062–3072, 2024, epub 2024 May 27. [Online]. Available: https://pubmed.ncbi.nlm.nih.gov/38803150/

    2024

  21. [29]

    Missing data is poorly handled and reported in prediction model studies using machine learning: a literature review,

    S. Nijman, A. Leeuwenberg, I. Beekers, I. Verkouter, J. Jacobs, M. Bots, F. Asselbergs, K. Moons, and T. Debray, “Missing data is poorly handled and reported in prediction model studies using machine learning: a literature review,”Journal of clinical epidemiology, vol. 142, pp...

    2022

  22. [30]

    Missing data imputation of high-resolution temporal climate time series data,

    E.Afrifa-Yamoah,U.A.Mueller,S.M.Taylor,andA.J.Fisher,“Missing data imputation of high-resolution temporal climate time series data,” Meteorological Applications, vol. 27, no. 1, p. e1873, 2020

    2020

  23. [31]

    Maximum likelihood from incomplete data via the em algorithm,

    A. P. Dempster, N. M. Laird, and D. B. Rubin, “Maximum likelihood from incomplete data via the em algorithm,”Journal of the royal statistical society: series B (methodological), vol. 39, no. 1, pp. 1–22, 1977

    1977

  24. [32]

    Multiple imputation for nonresponse in surveys. new york, ny: Johnwiley & sons,

    D. Rubin, “Multiple imputation for nonresponse in surveys. new york, ny: Johnwiley & sons,” 1987

    1987

  25. [33]

    Recurrent neural networks for multivariate time series with missing values,

    Z. Che, S. Purushotham, K. Cho, D. Sontag, and Y. Liu, “Recurrent neural networks for multivariate time series with missing values,” Scientific reports, vol. 8, no. 1, p. 6085, 2018

    2018

  26. [34]

    Long-term missing value imputation for time series data using deep neural networks,

    J. Park, J. Müller, B. Arora, B. Faybishenko, G. Pastorello, C. Varadharajan, R. Sahu, and D. Agarwal, “Long-term missing value imputation for time series data using deep neural networks,”Neural Comput. Appl., vol. 35, no. 12, p. 9071–9091, dec 2022. [Online]. Available: https...

    2022 doi

  27. [35]

    mice: Multivariate impu- tation by chained equations in r,

    S. Van Buuren and K. Groothuis-Oudshoorn, “mice: Multivariate impu- tation by chained equations in r,”Journal of statistical software, vol. 45, pp. 1–67, 2011

    2011

  28. [36]

    Missforest—non-parametric missing value imputation for mixed-type data,

    D. J. Stekhoven and P. Bühlmann, “Missforest—non-parametric missing value imputation for mixed-type data,”Bioinformatics, vol. 28, no. 1, pp. 112–118, 2012

    2012

  29. [37]

    Missing data imputa- tion using optimal transport,

    B. Muzellec, J. Josse, C. Boyer, and M. Cuturi, “Missing data imputa- tion using optimal transport,” inInternational Conference on Machine Learning. PMLR, 2020, pp. 7130–7140

    2020

  30. [38]

    The midas touch: Accurate and scalable missing-data imputation with deep learning,

    R. Lall and T. Robinson, “The midas touch: Accurate and scalable missing-data imputation with deep learning,”Political Analysis, vol. 30, no. 2, p. 179–196, 2022

    2022

  31. [39]

    Miwae: Deep generative modelling and imputation of incomplete data sets,

    P.-A. Mattei and J. Frellsen, “Miwae: Deep generative modelling and imputation of incomplete data sets,” inInternational conference on machine learning. PMLR, 2019, pp. 4413–4423

    2019

  32. [40]

    Miracle: Causally-aware imputation via learning missing data mechanisms,

    T. Kyono, Y. Zhang, A. Bellot, and M. van der Schaar, “Miracle: Causally-aware imputation via learning missing data mechanisms,”Ad- vances in Neural Information Processing Systems, vol. 34, pp. 23806– 23817, 2021

    2021

  33. [41]

    Gain: Missing data imputation using generative adversarial nets,

    J. Yoon, J. Jordon, and M. Schaar, “Gain: Missing data imputation using generative adversarial nets,” inInternational conference on machine learning. PMLR, 2018, pp. 5689–5698

    2018

  34. [42]

    Imputation of missing streamflow data at multiple gauging stations in Benin Republic,

    R. Mbuvha, P. J. Y. Adounkpe, W. T. Mongwe, M. C. M. Houngnibo, N. Newlands, and T. Marwala, “Imputation of missing streamflow data at multiple gauging stations in Benin Republic,” arXiv preprint, 2022

    2022

  35. [43]

    Multiple data imputation methods advance risk analysis and treatability of co-occurring inorganic chemicals in groundwater,

    A. U. Mahmood, M. Islam, A. V. Gulyuk, E. Briese, C. A. Velasco, M. Malu, N. Sharma, A. Spanias, Y. G. Yingling, and P. Westerhoff, “Multiple data imputation methods advance risk analysis and treatability of co-occurring inorganic chemicals in groundwater,”Environmental Scienc...

    2024

  36. [44]

    Multiple imputations by chained equations for recovering miss- ing daily streamflow observations: A case study of Langat River Basin in Malaysia,

    F. B. Hamzah, F. Mohamad Hamzah, S. F. Mohd Razali, and A. El- Shafie, “Multiple imputations by chained equations for recovering miss- ing daily streamflow observations: A case study of Langat River Basin in Malaysia,”Hydrological Sciences Journal, vol. 67, no. 1, pp. 137–149, 2022

    2022

  37. [45]

    Out-of-sample tests of forecasting accuracy: An analysis and review,

    L. J. Tashman, “Out-of-sample tests of forecasting accuracy: An analysis and review,”International Journal of Forecasting, vol. 16, no. 4, pp. 437–450, 2000

    2000

  38. [46]

    Projecting stream water quality using weighted regression on time, discharge, and season (wrtds): An example with drought conditions in the delaware river basin,

    C. T. Green, R. M. Hirsch, H. I. Essaid, and W. E. Sanford, “Projecting stream water quality using weighted regression on time, discharge, and season (wrtds): An example with drought conditions in the delaware river basin,”Science of the Total Environment, vol. 999, p. 180286, 2025

    2025

  39. [47]

    Probts: Benchmarking point and distributional forecasting across diverse prediction horizons,

    J. Zhang, X. Wen, Z. Zhang, S. Zheng, J. Li, and J. Bian, “Probts: Benchmarking point and distributional forecasting across diverse prediction horizons,”arXiv preprint arXiv:2310.07446, 2024, neurIPS 2024 Datasets and Benchmarks Track; version v5 (Oct. 21, 2024). [Online]. Ava...

    2024 arXiv

  40. [48]

    Adaptive machine learning for forecasting in wind energy: A dynamic, multi-algorithmic approach for short and long-term predictions,

    M. AlShafeey and C. Csaki, “Adaptive machine learning for forecasting in wind energy: A dynamic, multi-algorithmic approach for short and long-term predictions,”Heliyon, vol. 10, no. 15, p. e34807, 2024, pMCID: PMC11333901

    2024

  41. [49]

    The coefficient of determination r-squared is more informative than smape, mae, mape, mse and rmse in regression analysis evaluation,

    D. Chicco, M. J. Warrens, and G. Jurman, “The coefficient of determination r-squared is more informative than smape, mae, mape, mse and rmse in regression analysis evaluation,”PeerJ Computer Science, vol. 7, p. e623, 2021, pMCID: PMC8279135. [Online]. Available: https://pubmed...

    2021

  42. [50]

    Recommending training set sizes for classification,

    P. Koshute, J. Zook, and I. McCulloh, “Recommending training set sizes for classification,” 2021

    2021

  43. [51]

    The learning-curve sam- pling method applied to model-based clustering,

    C. Meek, B. Thiesson, and D. Heckerman, “The learning-curve sam- pling method applied to model-based clustering,”Journal of Machine Learning Research, vol. 2, pp. 397–418, 2002

    2002

  44. [52]

    How much complexity is war- ranted in a rainfall-runoff model?

    A. J. Jakeman and G. M. Hornberger, “How much complexity is war- ranted in a rainfall-runoff model?”Water Resources Research, vol. 29, no. 8, pp. 2637–2649, 1993

    1993

  45. [53]

    Automatic time series forecasting: Theforecastpackage for r,

    R. J. Hyndman and Y. Khandakar, “Automatic time series forecasting: Theforecastpackage for r,”Journal of Statistical Software, vol. 27, no. 3, pp. 1–22, 2008

    2008

  46. [54]

    Analysis of the effectiveness of ARIMA, SARIMA, and SVR models in time series forecasting: A case study of wind farm energy production,

    K. Szostek, D. Mazur, G. Drałus, and J. Kusznier, “Analysis of the effectiveness of ARIMA, SARIMA, and SVR models in time series forecasting: A case study of wind farm energy production,”Energies, vol. 17, no. 19, p. 4803, 2024

    2024

  47. [55]

    Statsmodels: Econometric and statistical modeling with python,

    S. Seabold and J. Perktold, “Statsmodels: Econometric and statistical modeling with python,” inProceedings of the 9th Python in TABLE III Per-lake training/test errors. The table also lists the 30 lakes used in the study. Lake Train MAE (m) Test MAE (m) Train nMAE Test nMAE An...

    2010

  48. [56]

    Darts: User-friendly modern machine learning for time series,

    J. Herzen, F. Lässig, S. G. Piazzetta, T. Neuer, L. Tafti, G. Raille, T. Van Pottelbergh, M. Pasieka, A. Skrodzki, N. Huguenin, M. Dumonal, J. Kościsz, D. Bader, F. Gusset, M. Benheddi, C. Williamson, M. Kosinski, M. Petrik, and G. Grosch, “Darts: User-friendly modern machine ...

    2022

  49. [57]

    Scikit-learn: Machine learning in Python,

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vander- plas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duch- esnay, “Scikit-learn: Machine learning in Python,”Journal of Machine Lea...

    2011

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.