pith. sign in

arxiv: 2601.08663 · v3 · submitted 2026-01-13 · 💻 cs.CE

Efficient Parameter Calibration of Numerical Weather Prediction Models via Evolutionary Sequential Transfer Optimization

Heping Fang , Bingdong Li , Peng Yang This is my paper

Pith reviewed 2026-05-16 14:39 UTC · model grok-4.3

classification 💻 cs.CE
keywords numerical weather predictionparameter calibrationevolutionary optimizationtransfer optimizationsequential calibrationsurrogate modelinghypervolume indicator
0
0 comments X

The pith

SEETO transfers knowledge across similar weather calibration tasks to reach better results with far fewer expensive evaluations.

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

The paper argues that sequential calibration of Numerical Weather Prediction models can be made much more efficient by detecting physical similarities between tasks rather than treating each one in isolation. It introduces a representation extractor that turns high-dimensional meteorological fields into latent vectors for measuring task similarity. On top of that similarity, SEETO reuses high-quality solution populations from matching past tasks as a warm start and builds an adaptive ensemble surrogate that blends source and target data with dynamic weights. Experiments on ten tasks show that this approach improves hypervolume by 6 percent on average when limited to 20 evaluations, while standard methods need substantially more runs to catch up. Readers should care because each calibration run is computationally heavy, so cutting the number of required simulations directly lowers the cost of keeping forecast models accurate.

Core claim

By mapping meteorological fields to latent representations, SEETO quantifies similarity between calibration tasks and applies bi-level transfer: superior populations from similar source tasks provide a warm start at the solution level, while an ensemble surrogate built from source data assists the search at the model level with adaptive weighting that balances old and new information, yielding higher hypervolume under a tight budget of twenty evaluations than isolated evolutionary methods.

What carries the argument

The meteorological state representation extractor that produces latent vectors for task similarity, driving the bi-level adaptive knowledge transfer inside the SEETO evolutionary algorithm.

If this is right

  • Fewer than twenty expensive model runs suffice to reach higher hypervolume than isolated methods achieve with many more runs.
  • Warm-start populations from similar past tasks accelerate convergence on new calibration problems.
  • Adaptive weighting in the ensemble surrogate prevents harmful transfer when source and target differ.
  • Sequential calibration across multiple NWP tasks becomes practical under realistic computational limits.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same representation-based transfer could be applied to other expensive simulation tuning problems that share underlying physics, such as climate model parameter adjustment.
  • An online version might update the similarity database after each new calibration, gradually improving efficiency across an entire forecast system.
  • Safeguards against negative transfer would be needed when the extractor flags similarity but the underlying dynamics have changed.

Load-bearing premise

The extractor must correctly identify physical similarities so that transferred solutions and models improve rather than degrade performance on the target task.

What would settle it

Run SEETO on a calibration task whose extracted representation signals strong similarity to a source task, yet the true optimal parameters differ markedly; if hypervolume then falls below the non-transfer baseline, the similarity measure is not reliable.

Figures

Figures reproduced from arXiv: 2601.08663 by Bingdong Li, Heping Fang, Peng Yang.

Figure 1
Figure 1. Figure 1: Flowchart of the NWP Model Parameter Calibration. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The framework of the SEETO. optimization process from scratch for each incoming target task. Consequently, constructing a source task data archive is imperative. As historical calibration tasks accumulate, the repository of transferable knowledge expands, facilitating the rapid adaptation of the optimization algorithm to newly ar￾riving target tasks. The archived data for each source task primarily consist… view at source ↗
Figure 3
Figure 3. Figure 3: HV evolution trajectories across 60 WRF evaluations. The subfigures with [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison of forecast fields and error maps for Task 1. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Parameter sensitivity of λ [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Parameter sensitivity of c. number of channels representing the meteorological element dimensions, denoted as λ, and the regulation parameter c, which governs the rate of transition from source domain knowledge to target domain local knowledge. The richness of the meteorological state representation di￾rectly dictates the accuracy of task similarity metrics. We cate￾gorized the input channel configurations… view at source ↗
read the original abstract

The configuration of physical parameterization schemes in Numerical Weather Prediction (NWP) models plays a critical role in determining the accuracy of the forecast. However, existing parameter calibration methods typically treat each calibration task as an isolated optimization problem. This approach suffers from prohibitive computational costs and necessitates performing iterative searches from scratch for each task, leading to low efficiency in sequential calibration scenarios. To address this issue, we propose the SEquential Evolutionary Transfer Optimization (SEETO) algorithm driven by the representations of the meteorological state. First, to accurately measure the physical similarity between calibration tasks, a meteorological state representation extractor is introduced to map high-dimensional meteorological fields into latent representations. Second, given the similarity in the latent space, a bi-level adaptive knowledge transfer mechanism is designed. At the solution level, superior populations from similar historical tasks are reused to achieve a "warm start" for optimization. At the model level, an ensemble surrogate model based on source task data is constructed to assist the search, employing an adaptive weighting mechanism to dynamically balance the contributions of source domain knowledge and target domain data. Extensive experiments across 10 distinct calibration tasks, which span varying source-target similarities, highlight SEETO's superior efficiency. Under a strict budget of 20 expensive evaluations, SEETO achieves a 6% average improvement in Hypervolume (HV) over two state-of-the-art baselines. Notably, to match SEETO's performance at this stage, the comparison algorithms would require an average of 64% and 28% additional evaluations, respectively. This presents a new paradigm for the efficient and accurate automated calibration of NWP model parameters.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes the SEquential Evolutionary Transfer Optimization (SEETO) algorithm for efficient calibration of physical parameterization schemes in Numerical Weather Prediction (NWP) models. It introduces a meteorological state representation extractor to quantify physical similarity between tasks in latent space and a bi-level adaptive transfer mechanism that reuses superior populations from similar historical tasks for warm-start initialization while constructing an ensemble surrogate with adaptive weighting between source and target data. Experiments across 10 calibration tasks with varying similarities report that, under a strict budget of 20 expensive evaluations, SEETO yields a 6% average Hypervolume improvement over two state-of-the-art baselines, which would require 28% and 64% additional evaluations on average to reach equivalent performance.

Significance. If the reported efficiency gains hold under rigorous validation, SEETO would represent a meaningful practical advance for sequential NWP calibration workflows by reducing the prohibitive cost of restarting optimization from scratch on each new task. The approach directly addresses a recurring bottleneck in operational weather modeling where parameter tuning must be repeated across related meteorological regimes.

major comments (2)
  1. [Experiments] Experiments section: the headline performance claims (6% HV gain, 28%/64% extra evaluations needed) are presented without statistical significance tests, variance estimates across runs, or explicit analysis of negative-transfer cases, leaving the central efficiency result only partially supported despite the abstract's concrete numbers.
  2. [Method] Method section (bi-level adaptive transfer): the adaptive weighting coefficients and ensemble surrogate construction rely on the assumption that the meteorological state representation extractor reliably captures physical similarity; no ablation or sensitivity study is provided to quantify how extractor accuracy affects transfer quality when source-target similarity is low.
minor comments (2)
  1. [Abstract] Abstract: define HV (Hypervolume) on first use and clarify the two baselines by name rather than referring to them generically.
  2. [Method] Notation: ensure consistent use of symbols for the latent representations and weighting coefficients across equations and text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review of our manuscript. We address the major comments point by point below and outline the revisions we plan to make.

read point-by-point responses

  1. Referee: Experiments section: the headline performance claims (6% HV gain, 28%/64% extra evaluations needed) are presented without statistical significance tests, variance estimates across runs, or explicit analysis of negative-transfer cases, leaving the central efficiency result only partially supported despite the abstract's concrete numbers.

    Authors: We acknowledge this limitation in the current version. The experiments were conducted with multiple runs, but the variance and significance were not explicitly reported. In the revision, we will provide standard deviations across runs, conduct statistical significance tests (e.g., Wilcoxon rank-sum test) to validate the 6% HV improvement, and include an analysis of potential negative transfer cases by examining performance on low-similarity task pairs. These changes will be made to strengthen the empirical support for our claims. revision: yes

  2. Referee: Method section (bi-level adaptive transfer): the adaptive weighting coefficients and ensemble surrogate construction rely on the assumption that the meteorological state representation extractor reliably captures physical similarity; no ablation or sensitivity study is provided to quantify how extractor accuracy affects transfer quality when source-target similarity is low.

    Authors: We agree that an ablation study would better substantiate the method. We will add an ablation experiment removing or degrading the meteorological state representation extractor and measure the resulting impact on optimization performance. Additionally, a sensitivity study will be included to show how variations in extractor accuracy (simulated by noise injection in latent space) affect the adaptive weighting and overall hypervolume, with particular focus on low-similarity scenarios. This will be incorporated in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper proposes the SEETO algorithm for efficient NWP parameter calibration using a meteorological state representation extractor and bi-level adaptive knowledge transfer from historical tasks. Its central claims are empirical efficiency results under a fixed 20-evaluation budget, directly measured via hypervolume improvements on 10 distinct calibration tasks with varying source-target similarities. No load-bearing step reduces by construction to fitted parameters, self-citations, or renamed inputs; the method is defined via standard evolutionary operators and external data, with experiments providing independent validation rather than self-referential definitions.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The approach rests on the unproven assumption that latent representations faithfully encode physical task similarity and on the introduction of new algorithmic components whose behavior is validated only empirically.

free parameters (1)
  • adaptive weighting coefficients
    The ensemble surrogate uses dynamic weights to balance source and target knowledge; these are adjusted during search and not derived from first principles.
axioms (1)
  • domain assumption Latent meteorological state representations accurately reflect physical similarity between calibration tasks
    Invoked to justify reuse of populations and surrogates from source tasks.
invented entities (1)
  • Meteorological state representation extractor no independent evidence
    purpose: Maps high-dimensional meteorological fields to latent vectors for task similarity measurement
    New component introduced to enable the transfer mechanism; no independent evidence of correctness provided.

pith-pipeline@v0.9.0 · 5589 in / 1252 out tokens · 26947 ms · 2026-05-16T14:39:53.640071+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages

  1. [1]

    Automated weather forecasting and field monitoring using gru-cnn model along with iot to support precision agriculture,

    T. Akilan and K. Baalamurugan, “Automated weather forecasting and field monitoring using gru-cnn model along with iot to support precision agriculture,”Expert systems with applications, vol. 249, p. 123468, 2024

    work page 2024

  2. [2]

    A review of high impact weather for aviation meteorology,

    I. Gultepe, R. Sharman, P. D. Williams, B. Zhou, G. Ellrod, P. Minnis, S. Trier, S. Griffin, S. S. Yum, B. Gharabaghiet al., “A review of high impact weather for aviation meteorology,”Pure and applied geophysics, vol. 176, no. 5, pp. 1869–1921, 2019

    work page 1921

  3. [3]

    Optimisation of energy man- agement in commercial buildings with weather forecasting inputs: A review,

    D. Lazos, A. B. Sproul, and M. Kay, “Optimisation of energy man- agement in commercial buildings with weather forecasting inputs: A review,”Renewable and Sustainable Energy Reviews, vol. 39, pp. 587– 603, 2014

    work page 2014

  4. [4]

    Artificial intelligence for modeling and understanding extreme weather and climate events,

    G. Camps-Valls, M.- ´A. Fern ´andez-Torres, K.-H. Cohrs, A. H ¨ohl, A. Castelletti, A. Pacal, C. Robin, F. Martinuzzi, I. Papoutsis, I. Prapas et al., “Artificial intelligence for modeling and understanding extreme weather and climate events,”Nature Communications, vol. 16, no. 1, p. 1919, 2025

    work page 1919

  5. [5]

    Artificial intelligence and numerical weather prediction models: A technical survey,

    M. Waqas, U. W. Humphries, B. Chueasa, and A. Wangwongchai, “Artificial intelligence and numerical weather prediction models: A technical survey,”Natural Hazards Research, vol. 5, no. 2, pp. 306– 320, 2025

    work page 2025

  6. [6]

    Can deep learning beat numerical weather prediction?

    M. G. Schultz, C. Betancourt, B. Gong, F. Kleinert, M. Langguth, L. H. Leufen, A. Mozaffari, and S. Stadtler, “Can deep learning beat numerical weather prediction?”Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, vol. 379, no. 2194, 2021

    work page 2021

  7. [7]

    Objective calibration of numerical weather prediction models,

    A. V oudouri, P. Khain, I. Carmona, O. Bellprat, F. Grazzini, E. Avgous- toglou, J. Bettems, and P. Kaufmann, “Objective calibration of numerical weather prediction models,”Atmospheric research, vol. 190, pp. 128– 140, 2017

    work page 2017

  8. [8]

    Automatic model calibration: A new way to improve numerical weather forecasting,

    Q. Duan, Z. Di, J. Quan, C. Wang, W. Gong, Y . Gan, A. Ye, C. Miao, S. Miao, X. Lianget al., “Automatic model calibration: A new way to improve numerical weather forecasting,”Bulletin of the American Meteorological Society, vol. 98, no. 5, pp. 959–970, 2017

    work page 2017

  9. [9]

    A description of the advanced research wrf version 4,

    W. C. Skamarock, J. B. Klemp, J. Dudhia, D. O. Gill, Z. Liu, J. Berner, W. Wang, J. G. Powers, M. G. Duda, D. M. Barkeret al., “A description of the advanced research wrf version 4,”NCAR tech. note ncar/tn-556+ str, vol. 145, no. 10.5065, 2019

    work page 2019

  10. [10]

    Sensitivity of localized heavy rainfall in northern japan to wrf physics parameterization schemes,

    Y . Hiraga and R. Tahara, “Sensitivity of localized heavy rainfall in northern japan to wrf physics parameterization schemes,”Atmospheric Research, vol. 314, p. 107802, 2025

    work page 2025

  11. [11]

    Simlob: Learning representations of limit order book for financial market simulation,

    Y . Li, Y . Wu, M. Zhong, S. Liu, and P. Yang, “Simlob: Learning representations of limit order book for financial market simulation,” IEEE Transactions on Artificial Intelligence, pp. 1–16, 2025

    work page 2025

  12. [12]

    Alleviating nonidentifiability: a high- fidelity calibration objective for financial market simulation with mul- tivariate time series data,

    C. Wang, J. Ren, and P. Yang, “Alleviating nonidentifiability: a high- fidelity calibration objective for financial market simulation with mul- tivariate time series data,”IEEE Transactions on Computational Social Systems, 2025

    work page 2025

  13. [13]

    Towards calibrating financial market simulators with high-frequency data,

    P. Yang, J. Ren, F. Wang, and K. Tang, “Towards calibrating financial market simulators with high-frequency data,”Complex System Modeling and Simulation, 2025

    work page 2025

  14. [14]

    Multi-objective evolutionary optimization for hardware-aware neural network pruning,

    W. Hong, G. Li, S. Liu, P. Yang, and K. Tang, “Multi-objective evolutionary optimization for hardware-aware neural network pruning,” Fundamental Research, vol. 4, no. 4, pp. 941–950, 2024

    work page 2024

  15. [15]

    Reducing idleness in financial cloud services via multi-objective evolutionary reinforcement learning based load balancer,

    P. Yang, L. Zhang, H. Liu, and G. Li, “Reducing idleness in financial cloud services via multi-objective evolutionary reinforcement learning based load balancer,”Science China Information Sciences, vol. 67, no. 2, p. 120102, 2024

    work page 2024

  16. [16]

    Subset selection by pareto optimiza- tion,

    C. Qian, Y . Yu, and Z.-H. Zhou, “Subset selection by pareto optimiza- tion,”Advances in neural information processing systems, vol. 28, 2015

    work page 2015

  17. [17]

    It’s morphing time: Unleashing the potential of multiple llms via multi-objective optimization,

    B. Li, Z. Di, Y . Yang, H. Qian, P. Yang, H. Hao, K. Tang, and A. Zhou, “It’s morphing time: Unleashing the potential of multiple llms via multi-objective optimization,”IEEE Transactions on Evolutionary Computation, 2025

    work page 2025

  18. [18]

    Causal inference based large-scale multi-objective optimization,

    B. Li, Y . Yang, P. Yang, G. Li, K. Tang, and A. Zhou, “Causal inference based large-scale multi-objective optimization,”IEEE Transactions on Evolutionary Computation, 2025

    work page 2025

  19. [19]

    Knee point- based multiobjective optimization for the numerical weather prediction model in the greater beijing area,

    H. Wang, H. Mo, Z. Di, R. Liu, Y . Lang, and Q. Duan, “Knee point- based multiobjective optimization for the numerical weather prediction model in the greater beijing area,”Geophysical Research Letters, vol. 50, no. 23, p. e2023GL104330, 2023

    work page 2023

  20. [20]

    Calibration of wrf model parameters using multiobjective adaptive surrogate model-based optimization to improve the prediction of the indian summer monsoon,

    S. Chinta and C. Balaji, “Calibration of wrf model parameters using multiobjective adaptive surrogate model-based optimization to improve the prediction of the indian summer monsoon,”Climate Dynamics, vol. 55, no. 3, pp. 631–650, 2020

    work page 2020

  21. [21]

    Parameter calibration to improve the prediction of tropical cyclones over the bay of bengal using machine learning–based multiobjective optimization,

    H. Baki, S. Chinta, C. Balaji, and B. Srinivasan, “Parameter calibration to improve the prediction of tropical cyclones over the bay of bengal using machine learning–based multiobjective optimization,”Journal of Applied Meteorology and Climatology, vol. 61, no. 7, pp. 819–837, 2022

    work page 2022

  22. [22]

    Efficient execution of the wrf model and other hpc applications in the cloud,

    H. A. Duran-Limon, J. Flores-Contreras, N. Parlavantzas, M. Zhao, and A. Meulenert-Pe ˜na, “Efficient execution of the wrf model and other hpc applications in the cloud,”Earth Science Informatics, vol. 9, no. 3, pp. 365–382, 2016

    work page 2016

  23. [23]

    (2022) WRF Version 4.4 Benchmark Data

    University Corporation for Atmospheric Research. (2022) WRF Version 4.4 Benchmark Data. Accessed: December 26, 2025. [Online]. Available: https://www2.mmm.ucar.edu/wrf/users/benchmark/ v44/benchdata v44.html

    work page 2022

  24. [24]

    Surrogate-assisted evolutionary computation: Recent advances and future challenges,

    Y . Jin, “Surrogate-assisted evolutionary computation: Recent advances and future challenges,”Swarm and Evolutionary Computation, vol. 1, no. 2, pp. 61–70, 2011

    work page 2011

  25. [25]

    Evolutionary transfer optimization-a new frontier in evolutionary computation research,

    K. C. Tan, L. Feng, and M. Jiang, “Evolutionary transfer optimization-a new frontier in evolutionary computation research,”IEEE Computational Intelligence Magazine, vol. 16, no. 1, pp. 22–33, 2021

    work page 2021

  26. [26]

    Ensemble of domain adaptation-based knowledge transfer for evolutionary multitasking,

    W. Lin, Q. Lin, L. Feng, and K. C. Tan, “Ensemble of domain adaptation-based knowledge transfer for evolutionary multitasking,” IEEE Transactions on Evolutionary Computation, vol. 28, no. 2, pp. 388–402, 2023

    work page 2023

  27. [27]

    Solving expensive dynamic multi-objective problem via cross-problem knowledge transfer,

    Z. Cheng, X. Xue, H. Tang, and L. Feng, “Solving expensive dynamic multi-objective problem via cross-problem knowledge transfer,” inIn- ternational Conference on Neural Information Processing. Springer, 2024, pp. 264–278

    work page 2024

  28. [28]

    Solution transfer in evolutionary optimization: An empirical study on sequential transfer,

    X. Xue, C. Yang, L. Feng, K. Zhang, L. Song, and K. C. Tan, “Solution transfer in evolutionary optimization: An empirical study on sequential transfer,”IEEE Transactions on Evolutionary Computation, vol. 28, no. 6, pp. 1776–1793, 2023

    work page 2023

  29. [29]

    Evolutionary sequential transfer optimization for objective- heterogeneous problems,

    X. Xue, C. Yang, Y . Hu, K. Zhang, Y .-M. Cheung, L. Song, and K. C. Tan, “Evolutionary sequential transfer optimization for objective- heterogeneous problems,”IEEE Transactions on Evolutionary Compu- tation, vol. 26, no. 6, pp. 1424–1438, 2021

    work page 2021

  30. [30]

    Global and local search experience-based evolutionary sequential transfer optimization,

    C. Cao, K. Zhang, X. Xue, K. C. Tan, J. Wang, L. Zhang, P. Liu, and X. Yan, “Global and local search experience-based evolutionary sequential transfer optimization,”IEEE Transactions on Evolutionary Computation, 2024

    work page 2024

  31. [31]

    Towards robustness and explainability of automatic algorithm selection,

    X. Wu, J. Wu, Y . Zhou, L. Feng, and K. C. Tan, “Towards robustness and explainability of automatic algorithm selection,” inForty-second International Conference on Machine Learning, 2025

    work page 2025

  32. [32]

    Multi-agent dynamic algorithm configuration,

    K. Xue, J. Xu, L. Yuan, M. Li, C. Qian, Z. Zhang, and Y . Yu, “Multi-agent dynamic algorithm configuration,”Advances in Neural Information Processing Systems, vol. 35, pp. 20 147–20 161, 2022

    work page 2022

  33. [33]

    Learnable evolutionary search across heterogeneous problems via kernelized autoencoding,

    L. Zhou, L. Feng, A. Gupta, and Y .-S. Ong, “Learnable evolutionary search across heterogeneous problems via kernelized autoencoding,” IEEE Transactions on Evolutionary Computation, vol. 25, no. 3, pp. 567–581, 2021

    work page 2021

  34. [34]

    Multiobjective many-tasking evolutionary optimization using diversified gaussian-based knowledge transfer,

    Q. Lin, Q. Wang, B. Chen, Y . Ye, L. Ma, and K. C. Tan, “Multiobjective many-tasking evolutionary optimization using diversified gaussian-based knowledge transfer,”IEEE Transactions on Evolutionary Computation, 2024

    work page 2024

  35. [35]

    Multiproblem surrogates: Transfer evolutionary multiobjective optimization of com- putationally expensive problems,

    A. T. W. Min, Y .-S. Ong, A. Gupta, and C.-K. Goh, “Multiproblem surrogates: Transfer evolutionary multiobjective optimization of com- putationally expensive problems,”IEEE Transactions on Evolutionary Computation, vol. 23, no. 1, pp. 15–28, 2017

    work page 2017

  36. [36]

    Scalable transfer evolu- tionary optimization: Coping with big task instances,

    M. Shakeri, E. Miahi, A. Gupta, and Y .-S. Ong, “Scalable transfer evolu- tionary optimization: Coping with big task instances,”IEEE transactions on cybernetics, vol. 53, no. 10, pp. 6160–6172, 2022. 14

    work page 2022

  37. [37]

    Multisource selective transfer framework in multiobjective optimization problems,

    J. Zhang, W. Zhou, X. Chen, W. Yao, and L. Cao, “Multisource selective transfer framework in multiobjective optimization problems,” IEEE Transactions on Evolutionary Computation, vol. 24, no. 3, pp. 424–438, 2019

    work page 2019

  38. [38]

    Data-driven evolutionary computation under continuously streaming environments: A drift-aware approach,

    Y .-T. Zhong and Y .-J. Gong, “Data-driven evolutionary computation under continuously streaming environments: A drift-aware approach,” IEEE Transactions on Evolutionary Computation, 2025

    work page 2025

  39. [39]

    Multiobjective multi- tasking optimization based on incremental learning,

    J. Lin, H.-L. Liu, B. Xue, M. Zhang, and F. Gu, “Multiobjective multi- tasking optimization based on incremental learning,”IEEE Transactions on Evolutionary Computation, vol. 24, no. 5, pp. 824–838, 2019

    work page 2019

  40. [40]

    Regularized evolu- tionary multitask optimization: Learning to intertask transfer in aligned subspace,

    Z. Tang, M. Gong, Y . Wu, W. Liu, and Y . Xie, “Regularized evolu- tionary multitask optimization: Learning to intertask transfer in aligned subspace,”IEEE Transactions on Evolutionary Computation, vol. 25, no. 2, pp. 262–276, 2020

    work page 2020

  41. [41]

    Surrogate-assisted evolutionary framework with adaptive knowledge transfer for multi-task optimiza- tion,

    S. Huang, J. Zhong, and W.-J. Yu, “Surrogate-assisted evolutionary framework with adaptive knowledge transfer for multi-task optimiza- tion,”IEEE transactions on emerging topics in computing, vol. 9, no. 4, pp. 1930–1944, 2019

    work page 1930

  42. [42]

    Fixing the double penalty in data-driven weather forecasting through a modified spherical harmonic loss function,

    C. Subich, S. Z. Husain, L. Separovic, and J. Yang, “Fixing the double penalty in data-driven weather forecasting through a modified spherical harmonic loss function,”arXiv preprint arXiv:2501.19374, 2025

    work page arXiv 2025

  43. [43]

    Multiobjective adaptive surrogate modeling-based optimization for parameter estimation of large, complex geophysical models,

    W. Gong, Q. Duan, J. Li, C. Wang, Z. Di, A. Ye, C. Miao, and Y . Dai, “Multiobjective adaptive surrogate modeling-based optimization for parameter estimation of large, complex geophysical models,”Water Resources Research, vol. 52, no. 3, pp. 1984–2008, 2016

    work page 1984

  44. [44]

    Self-supervised classification of weather systems based on spatiotemporal contrastive learning,

    L. Wang, Q. Li, and Q. Lv, “Self-supervised classification of weather systems based on spatiotemporal contrastive learning,”Geophysical Research Letters, vol. 49, no. 15, p. e2022GL099131, 2022

    work page 2022

  45. [45]

    Atmodist: Self-supervised representation learning for atmospheric dynamics,

    S. Hoffmann and C. Lessig, “Atmodist: Self-supervised representation learning for atmospheric dynamics,”Environmental Data Science, vol. 2, p. e6, 2023

    work page 2023

  46. [46]

    The era5 global reanalysis,

    H. Hersbach, B. Bell, P. Berrisford, S. Hirahara, A. Hor ´anyi, J. Mu ˜noz- Sabater, J. Nicolas, C. Peubey, R. Radu, D. Scheperset al., “The era5 global reanalysis,”Quarterly journal of the royal meteorological society, vol. 146, no. 730, pp. 1999–2049, 2020

    work page 1999

  47. [47]

    CMA Land Data Assimilation System (CLDAS-V2.0),

    National Meteorological Information Center, “CMA Land Data Assimilation System (CLDAS-V2.0),” 2017, accessed: December 26,

    work page 2017

  48. [48]

    Available: https://data.cma.cn/data/cdcdetail/dataCode/ NAFP CLDAS2.0 RT.html

    [Online]. Available: https://data.cma.cn/data/cdcdetail/dataCode/ NAFP CLDAS2.0 RT.html

    work page

  49. [49]

    Platemo: A matlab platform for evolutionary multi-objective optimization [educational forum],

    Y . Tian, R. Cheng, X. Zhang, and Y . Jin, “Platemo: A matlab platform for evolutionary multi-objective optimization [educational forum],”IEEE Computational Intelligence Magazine, vol. 12, no. 4, pp. 73–87, 2017

    work page 2017

  50. [50]

    Weather- bench 2: A benchmark for the next generation of data-driven global weather models,

    S. Rasp, S. Hoyer, A. Merose, I. Langmore, P. Battaglia, T. Russell, A. Sanchez-Gonzalez, V . Yang, R. Carver, S. Agrawalet al., “Weather- bench 2: A benchmark for the next generation of data-driven global weather models,”Journal of Advances in Modeling Earth Systems, vol. 16, no. 6, p. e2023MS004019, 2024

    work page 2024