Pith. sign in

REVIEW 3 major objections 7 minor 68 references

A data-driven solar forecast at 5-km, 10-minute resolution over 7 days makes battery scheduling near-optimal and pushes industrial solar-battery projects past the 12% IRR threshold.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

SunCastNet combines AI weather forecasting with reinforcement-learning battery control to turn high-resolution solar forecasts into large regret reductions and more profitable industrial solar projects.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection End-to-end solar forecast-to-economic value paper that is worth refereeing, but the economic backtest and ERA5-initialization gap need to be closed before the headline numbers can be trusted. the 3 major comments →

arxiv 2509.06925 v1 pith:CD3ORFUF submitted 2025-09-08 physics.geo-ph cs.LG

Data-driven solar forecasting enables near-optimal economic decisions

Zhixiang Dai , Minghao Yin , Xuanhong Chen , Alberto Carpentieri , Jussi Leinonen , Boris Bonev , Chengzhe Zhong , Thorsten Kurth
show 22 more authors
Jingan Sun Ram Cherukuri Yuzhou Zhang Ruihua Zhang Farah Hariri Xiaodong Ding Chuanxiang Zhu Dake Zhang Yaodan Cui Yuxi Lu Yue Song Bin He Jie Chen Yixin Zhu Chenheng Xu Maofeng Liu Zeyi Niu Wanpeng Qi Xu Shan Siyuan Xian Ning Lin Kairui Feng
This is my paper
classification physics.geo-ph cs.LG
keywords solar forecastingreinforcement learningbattery schedulingsolar radiationinvestment backtestinternal rate of returndownscalingindustrial decarbonization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 tries to show that the reason industrial solar-battery systems are hard to operate is not only cost or weather variability but the lack of cheap, high-resolution solar forecasts. It introduces SunCastNet, a four-stage AI pipeline that turns coarse global weather fields into 5-kilometer, 10-minute forecasts of surface solar radiation up to seven days ahead. When these forecasts feed a reinforcement-learning battery scheduler, operational regret—the money lost compared with perfect foresight—drops 76–93% relative to a conservative worst-case baseline, far more than the 43–66% achieved with a lower-resolution operational forecast. In 25-year investment backtests, up to five of ten high-emitting industrial sectors per region cross the 12% internal-rate-of-return threshold, meaning forecast quality alone can move projects from infeasible to profitable.

Core claim

SunCastNet forecasts surface solar radiation downwards at 0.05 degrees and 10-minute resolution up to 7 days ahead, with median relative errors of 13% at 2 days and 20% at 7 days across 2,164 stations in China, 5–10 percentage points better than the GFS operational forecast, and about 20% higher mutual information with ground truth. The paper's central economic claim is that this extra information is what battery operators need: when reinforcement-learning scheduling policies are trained on these forecasts, regret relative to a perfect-information controller falls by 76–93% (50±25% quantiles), compared with 43–66% for GFS-based policies. In 25-year investment backtests, up to five of ten ind

What carries the argument

The central object is SunCastNet, a four-stage AI forecasting chain. Stage one uses a spherical Fourier neural operator to evolve 73 global atmospheric variables at 0.25-degree, 6-hour resolution; stage two uses a modulated adaptive Fourier neural operator to interpolate to hourly fields; stage three applies an AFNO diagnostic to convert key atmospheric fields into hourly surface solar radiation downwards; stage four, CorrDiffSolar, uses residual-corrective diffusion to downscale to 0.05-degree, 10-minute SSRD. The economic argument rides on coupling these forecasts to a reinforcement-learning battery controller, with regret measured against a perfect-information benchmark and against a wors

Load-bearing premise

The economic conclusions assume that 42 confidential industrial demand profiles, 2025 solar and battery costs, and current Chinese peak–valley price spreads remain representative over a 25-year backtest; if demand or prices shift, the IRR crossings and regret numbers are simulated outputs, not realized outcomes.

What would settle it

Re-run the 25-year backtest with demand profiles from a different set of industrial sites or with price spreads from 2010–2020; if fewer than five sectors per region cross the 12% IRR threshold, or if the regret reduction versus the worst-case baseline falls below 76%, the economic claim is refuted. A complementary live test: run SunCastNet-driven RL battery control at an industrial site for one year and compare realized revenue against the perfect-information bound.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Battery operators can exploit cloudy-day warnings: instead of keeping defensive reserves, they can precharge before low-irradiance periods, which is what the regret reductions quantify.
  • Forecast horizon is an economic variable: a 7-day, moderately resolved forecast is worth more than a 2-day, high-resolution one in this setting, so industrial planning cycles should be built around week-ahead forecasts.
  • Forecast information content, not just RMSE, determines value; evaluation metrics such as mutual information and temporal consistency track operational gains better than point error.
  • At about $0.50 per continental 7-day forecast and about 25 minutes per run on one GPU, the pipeline is cheap enough for routine industrial use.
  • In regions with high irradiance variability, extra forecast skill can move solar-plus-storage projects above the 12% IRR threshold, enlarging the economically feasible geography.

Where Pith is reading between the lines

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

  • This is an inference: the same consistency-aware, RL-coupled design could transfer to wind, load, or price forecasting, where a comparable 'always-normal' baseline can look accurate but destroy scheduling value.
  • The paper leaves untested how sensitive the IRR crossings are to the confidential demand profiles and to price-spread changes; a public synthetic-demand benchmark with stressed price spreads would tell whether the five-of-ten result is structural or data-specific.
  • Because mutual information and inconsistency, not RMSE, predicted the operational gains, forecast developers might optimize directly for a differentiable regret surrogate rather than point error, potentially yielding larger economic returns than further RMSE reduction.
  • A direct testable extension: compare a 7-day, 0.25-degree forecast with a 2-day, 0.05-degree forecast in the same RL backtest; the paper's horizon results predict the longer, coarser forecast wins, which would isolate horizon from resolution as the value driver.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The paper introduces SunCastNet, a four-stage data-driven pipeline (SFNO global circulation, ModAFNO temporal interpolation, AFNO solar radiation diagnostic, CorrDiffSolar downscaling) that produces 0.05°, 10-minute horizontal-resolution forecasts of surface solar radiation downwards up to 7 days ahead. The forecast component is validated against 2,164 stations in China and compared with GFS, reporting 5–10% lower relative errors and about 20% higher mutual information. The paper then embeds these forecasts in a reinforcement-learning (RL) battery-management framework and, using 25-year ERA5-driven retrospective forecasts, claims 76–93% operational regret reduction relative to robust decision making (RDM), compared with 43–66% for GFS, and up to five of ten industrial sectors per region crossing a 12% IRR viability threshold. The central claim is that high-resolution, long-horizon solar forecasts translate directly into near-optimal economic decisions.

Significance. If the results hold, the paper makes a strong case that data-driven solar forecasting has decision-relevant value beyond conventional error metrics. The forecast component is externally validated with 2,164 stations, a GFS baseline, and a 2020–2025 robustness check, and the authors release code and sample data. The use of regret against perfect-information and RDM baselines is a sensible way to link forecast quality to operational and investment outcomes. However, the economic headline rests on a simulated backtest with confidential demand data and ERA5-initialized forecasts; transferability to operational decision-making requires additional evidence. The paper also honestly states limitations (China-only evaluation, regulatory constraints), which strengthens its credibility.

major comments (3)
  1. [Results, 'Decision-making under SunCastNet'; Fig. 4 caption] The 25-year economic backtest explicitly uses 'ERA5-driven retrospective forecasts.' ERA5 is a reanalysis whose initial conditions are observationally constrained and not available in real time; operational deployment would initialize SunCastNet from operational NWP analyses (GFS/IFS). If the economic results in Figs. 4d–4g are obtained with ERA5-initialized forecasts, the 76–93% regret reduction and IRR crossings may partly reflect superior initial conditions rather than SunCastNet's architecture. The 2020–2025 robustness check (Fig. S4) still uses the same ERA5-driven setup. Because the central claim is about enabling near-optimal economic decisions operationally, the authors should repeat at least one backtest with operational-analysis-initialized forecasts or quantify the sensitivity of economic metrics to initial-condition degradation.
  2. [Data and materials availability; Tables S1–S2] The 42 industrial electricity-demand profiles are confidential, and the backtest applies 2025 PV/battery costs and time-of-use price spreads across the 25-year period. No sensitivity analysis is provided in the manuscript text. While the regret reduction comparisons are less sensitive because all baselines share the same demand/cost assumptions, the headline IRR crossings (Figs. 4f–g) depend directly on these external parameters being representative over 25 years. The authors should report a sensitivity analysis of IRR and regret to demand-profile variation, price-spread scenarios, and cost degradation, and should release aggregated or anonymized demand profiles to support reproducibility.
  3. [Materials and Methods; Fig. 1b] The economic evaluation framework, including the RL state/action space, reward function, battery degradation model, and training/evaluation split, is not described in the main text, and the SI is not provided in the reviewed manuscript. In particular, it is unclear whether RL policies are trained and evaluated on the same 25-year forecast series (in-sample) or on held-out periods. The authors should move the essential equations and parameter tables into the main text or a fully available SI, and state the train/test split explicitly.
minor comments (7)
  1. [Abstract / Introduction / Discussion] The regret-reduction range is reported inconsistently: 76–93% in the Abstract, 72–93% in the Introduction, and 70–90% in the Discussion. Harmonize these numbers.
  2. [Fig. 2 captions] The spring and summer dates are inconsistent between the main text ('16 January 2020', '20 July 2020') and the figure captions ('17 January 2020', '22 July 2020'). Correct the mismatch.
  3. [Fig. 3a caption] The caption refers to 'daily irradiation' but the text describes errors of 'daily peak SSRD at 12:00 local time.' Clarify which metric is plotted.
  4. [Figs. 3a, 4d–4e] The phrase '50±25% quantiles' is ambiguous. Use 'interquartile range (IQR)' consistently.
  5. [Fig. 1a caption] Typo: 'four-stage sequence(a SunCastNet)' should read 'four-stage sequence of SunCastNet'.
  6. [Abstract] The abstract calls the sectors 'high-emitting', but the list (automobile, electronics, food processing, textiles, pharmaceuticals, etc.) is not exclusively high-emitting. Define the criterion or rephrase.
  7. [Introduction, cost claim] The statement that a forecast costs 'approximately $0.5 per continental-scale forecast' lacks a cost model or energy-price assumption. Add a footnote or supplementary detail.

Circularity Check

0 steps flagged

No significant circularity; SunCastNet's forecast and economic claims are evaluated against independent ground truth and baselines.

full rationale

SunCastNet's derivation chain is not circular. The forecast skill claim is tested against independent CMA station observations (2,164 stations) and against an operational NWP baseline (GFS), rather than against the model's own training targets. The economic evaluation compares RL policies trained on SunCastNet/GFS forecasts against a perfect-information oracle and an RDM baseline, so the regret reductions are not fitted constants and are evaluated on the same metric they optimize; the SunCastNet-vs-GFS contrast controls for the RL-vs-RDM framework. The cited prior modules (SFNO, ModAFNO, AFNO-based diagnostic, CorrDiffSolar) are building blocks with overlapping authorship, but the paper's contribution is their integration and downstream evaluation; the load-bearing evidence is the in-paper comparison against ground truth and GFS, with code release, so the self-citations do not reduce the central claim. The main caveats are non-circular: the 25-year economic backtest is initialized from ERA5 reanalysis rather than operational analyses, which may overstate real-time transferability; the demand data are confidential and no sensitivity analysis is shown; and the economic outcomes are simulation outputs under assumed costs/prices. These are external-validity and correctness risks, not circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central economic claim depends on several external economic inputs (costs, prices, IRR threshold), domain assumptions about reanalysis and satellite ground truth, and an implicit stationarity of 2025 economics across 25 years. These are not derived in the paper, so the reported regret and IRR numbers inherit all of them.

free parameters (4)
  • IRR viability threshold = 12% per year
    Chosen as the commercial viability threshold in the abstract and Results; it is an economic convention, not derived from the data.
  • PV module and battery system costs = 2025 China market prices (refs 76-79)
    IRR backtest uses current cost inputs; 25-year stability is assumed.
  • Time-of-use electricity price spreads by region = regional peak-valley differentials up to about 1.33 RMB/kWh (refs 68-69)
    These price spreads determine battery arbitrage value and thus regret and IRR; they are external market inputs.
  • Battery degradation and cycle-aging parameters = from NREL/PNNL LCOS reports (refs 71-73, 80)
    Battery lifetime economics depend on these coefficients; they are not measured in this work.
axioms (5)
  • domain assumption ERA5 reanalysis is a faithful historical atmosphere for the 25-year retrospective forecasts.
    SunCastNet is initialized/trained with ERA5 and the backtest uses ERA5-driven forecasts; if reanalysis biases matter, the economic results inherit them.
  • domain assumption H8/AHI satellite SSRD product (2016-2020) is valid ground truth at 0.05 degree scale.
    Used for calibration and validation; no independent radiation budget closure is shown.
  • ad hoc to paper 42 industrial demand profiles represent ten sectors over the 25-year backtest.
    Data availability says these reflect actual production in representative enterprises but they are confidential; no statistical representativeness test is shown in the provided text.
  • ad hoc to paper 2025 PV/battery costs and TOU price structures can be applied across the entire 2000-2025 backtest.
    The backtest IRR is computed on historical weather with present-day economics; the main text does not state or defend this stationarity.
  • domain assumption Battery model assumes no electricity sell-back to the grid.
    Explicitly noted as a limitation in the Discussion; it constrains returns and is realistic for Chinese industrial consumers, but different regulatory regimes would change conclusions.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-driven solar forecasting enables near-optimal economic decisions." pith.science (2026). https://pith.science/paper/CD3ORFUF

@misc{pith2026250906925,
  author       = {Pith},
  title        = {Pith review of: Data-driven solar forecasting enables near-optimal economic decisions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CD3ORFUF}},
  note         = {Machine review of arXiv:2509.06925}
}
Share X Bluesky LinkedIn Reddit HN
abstract

Solar energy adoption is critical to achieving net-zero emissions. However, it remains difficult for many industrial and commercial actors to decide on whether they should adopt distributed solar-battery systems, which is largely due to the unavailability of fast, low-cost, and high-resolution irradiance forecasts. Here, we present SunCastNet, a lightweight data-driven forecasting system that provides 0.05$^\circ$, 10-minute resolution predictions of surface solar radiation downwards (SSRD) up to 7 days ahead. SunCastNet, coupled with reinforcement learning (RL) for battery scheduling, reduces operational regret by 76--93\% compared to robust decision making (RDM). In 25-year investment backtests, it enables up to five of ten high-emitting industrial sectors per region to cross the commercial viability threshold of 12\% Internal Rate of Return (IRR). These results show that high-resolution, long-horizon solar forecasts can directly translate into measurable economic gains, supporting near-optimal energy operations and accelerating renewable deployment.

Figures

Figures reproduced from arXiv: 2509.06925 by Alberto Carpentieri, Bin He, Boris Bonev, Chengzhe Zhong, Chenheng Xu, Chuanxiang Zhu, Dake Zhang, Farah Hariri, Jie Chen, Jingan Sun, Jussi Leinonen, Kairui Feng, Maofeng Liu, Minghao Yin, Ning Lin, Ram Cherukuri, Ruihua Zhang, Siyuan Xian, Thorsten Kurth, Wanpeng Qi, Xiaodong Ding, Xuanhong Chen, Xu Shan, Yaodan Cui, Yixin Zhu, Yue Song, Yuxi Lu, Yuzhou Zhang, Zeyi Niu, Zhixiang Dai.

Figure 1
Figure 1. Figure 1: SunCastNet solar forecasting pipeline coupled with an RL framework for industrial economic evaluation. (a) SunCastNet Framework: The system begins with IFS/GFS (73 atmospheric variables at 0.25°, 6-hour intervals) , processed by a four-stage sequence(a SunCastNet): (i) Weather forecasting with SFNO, which predicts global circulation using 73 input variables to produce 6-hourly fields at 0.25° resolution; (… view at source ↗
Figure 2
Figure 2. Figure 2: Forecast skill of SSRD across seasons and lead times with SunCastNet. (a) Forecasts at 5-km resolution for a typical spring day (17 January 2020), showing SSRD at 12:00 local time (24-hour clock) predicted from forecasts issued at 03:00 with lead times of 1, 2, 3, 5, and 7 days, compared against satellite-derived ground truth (GT). (b) Same as (a) but for a typical summer day (22 July 2020). (c) Forecast s… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of SunCastNet and GFS forecasts of SSRD. (a) Relative error as a function of forecast lead time (2–7 days) of daily irradiation across 2,164 stations in China. Red line and shading denote the median and interquartile range (IQR, 25th–75th percentile) of SunCastNet errors, while blue line and shading denote the corresponding values for GFS. (b) Example time series of SSRD at one station (located … view at source ↗
Figure 4
Figure 4. Figure 4: From forecast consistency to decision outcomes under SunCastNet and GFS. (a–b) One-day forecast inconsistency for SunCastNet and GFS, defined as the difference between the predicted irradiation for day 2 issued on day 1 versus that issued on day 2 itself (red indicates larger inconsistency, blue smaller). (c) Monthly standard deviation of SSRD from satellite ground truth, used as a naïve predictability bas… view at source ↗

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

68 extracted references · 49 canonical work pages · 2 internal anchors

  1. [2]

    Soergel, B.et al.A sustainable development pathway for climate action within the un 2030 agenda.Nat. Clim. Chang.11, 656–664, DOI: 10.1038/s41558-021-01098-3 (2021)

    work page doi:10.1038/s41558-021-01098-3 2030

  2. [3]

    Wang, Y .et al.Global spatiotemporal optimization of photovoltaic and wind power to achieve the paris agreement targets. Nat. Commun.16, 2127, DOI: 10.1038/s41467-025-57292-w (2025)

    work page doi:10.1038/s41467-025-57292-w 2025

  3. [4]

    Lu, X.et al.Combined solar power and storage as cost-competitive and grid-compatible supply for china’s future carbon-neutral electricity system.Proc. Natl. Acad. Sci.118, e2103471118, DOI: 10.1073/pnas.2103471118 (2021)

    work page doi:10.1073/pnas.2103471118 2021

  4. [5]

    Yan, J., Yang, Y ., Elia Campana, P. & He, J. City-level analysis of subsidy-free solar photovoltaic electricity price, profits and grid parity in china.Nat. Energy4, 709–717, DOI: 10.1038/s41560-019-0441-z (2019). 7

    work page doi:10.1038/s41560-019-0441-z 2019

  5. [6]

    & Porporato, A

    Yin, J., Molini, A. & Porporato, A. Impacts of solar intermittency on future photovoltaic reliability.Nat. communications 11, 4781, DOI: 10.1038/s41467-020-18602-6 (2020)

    work page doi:10.1038/s41467-020-18602-6 2020

  6. [7]

    Li, X., Mauzerall, D. L. & Bergin, M. H. Global reduction of solar power generation efficiency due to aerosols and panel soiling.Nat. Sustain.3, 720–727, DOI: 10.1038/s41893-020-0553-2 (2020)

    work page doi:10.1038/s41893-020-0553-2 2020

  7. [8]

    communications6, 10014, DOI: 10.1038/ncomms10014 (2015)

    Jerez, S.et al.The impact of climate change on photovoltaic power generation in europe.Nat. communications6, 10014, DOI: 10.1038/ncomms10014 (2015)

    doi:10.1038/ncomms10014 2015

  8. [9]

    & Hall, J

    Huang, J., Li, W., Guo, L., Hu, X. & Hall, J. W. Renewable energy and household economy in rural china.Renew. Energy 155, 669–676 (2020)

    work page 2020

  9. [10]

    & Rahim, N

    Sobri, S., Koohi-Kamali, S. & Rahim, N. A. Solar photovoltaic generation forecasting methods: A review.Energy conversion management156, 459–497, DOI: 10.1016/j.enconman.2017.11.019 (2018)

    work page doi:10.1016/j.enconman.2017.11.019 2017

  10. [11]

    & Arif, M

    Ahmed, R., Sreeram, V ., Mishra, Y . & Arif, M. A review and evaluation of the state-of-the-art in PV solar power forecasting: Techniques and optimization.Renew. sustainable energy reviews124, 109792, DOI: 10.1016/j.rser.2020.109792 (2020)

    work page arXiv 2020

  11. [12]

    K.et al.Forecasting of photovoltaic power generation and model optimization: A review.Renew

    Das, U. K.et al.Forecasting of photovoltaic power generation and model optimization: A review.Renew. Sustain. Energy Rev.81, 912–928, DOI: 10.1016/j.rser.2017.08.017 (2018). 13.Miller, S. D., Rogers, M. A., Haynes, J. M., Sengupta, M. & Heidinger, A. K. Short-term solar irradiance forecasting via satellite/model coupling.Sol. Energy168, 102–117, DOI: 10.1...

    work page doi:10.1016/j.rser.2017.08.017 2017

  12. [14]

    G., Reichert, A

    Coville, A., Zivin, J. G., Reichert, A. & Reitmann, A.-K. Quality signaling and demand for renewable energy technology: Evidence from a randomized field experiment.J. Dev. Econ.176, 103514 (2025)

    work page 2025

  13. [15]

    Cui, C., Lonergan, K. E. & Sansavini, G. Policy-driven transformation of global solar PV supply chains and resulting impacts.Nat. Commun.16, 6742, DOI: 10.1038/s41467-025-61979-5 (2025)

    work page doi:10.1038/s41467-025-61979-5 2025

  14. [16]

    A.et al.Solar energy resource availability under extreme and historical wildfire smoke conditions.Nat

    Corwin, K. A.et al.Solar energy resource availability under extreme and historical wildfire smoke conditions.Nat. Commun.16, 245, DOI: 10.1038/s41467-024-54163-8 (2025)

    doi:10.1038/s41467-024-54163-8 2025

  15. [17]

    arXiv preprint arXiv:2202.11214(2022)

    Pathak, J.et al.Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators. arXiv preprint arXiv:2202.11214(2022)

    Pith/arXiv arXiv 2022

  16. [18]

    adi2336 (2023)

    Lam, R.et al.Learning skillful medium-range global weather forecasting.Science382, 1416–1421, DOI: 10.1126/science. adi2336 (2023)

    doi:10.1126/science 2023

  17. [19]

    Bi, K.et al.Accurate medium-range global weather forecasting with 3d neural networks.Nature619, 533–538, DOI: 10.1038/s41586-023-06185-3 (2023)

    doi:10.1038/s41586-023-06185-3 2023

  18. [20]

    Rasp, S.et al.Weatherbench 2: A benchmark for the next generation of data-driven global weather models.J. Adv. Model. Earth Syst.16, e2023MS004019, DOI: 10.1029/2023MS004019 (2024)

    doi:10.1029/2023ms004019 2024

  19. [21]

    Chen, K.et al.FengWu: Pushing the skillful global medium-range weather forecast beyond 10 days lead.arXiv preprint arXiv:2304.02948DOI: 10.48550/arXiv.2304.02948 (2023)

    doi:10.48550/arxiv.2304.02948 2023

  20. [22]

    Commun.16, 6658, DOI: 10.1038/ s41467-025-62024-1 (2025)

    Sun, X.et al.A data-to-forecast machine learning system for global weather.Nat. Commun.16, 6658, DOI: 10.1038/ s41467-025-62024-1 (2025)

    work page 2025

  21. [23]

    Price, I.et al.Probabilistic weather forecasting with machine learning.Nature637, 84–90, DOI: 10.1038/ s41586-024-08252-9 (2025)

    work page 2025

  22. [24]

    Kochkov, D.et al.Neural general circulation models for weather and climate.Nature632, 1060–1066, DOI: 10.1038/ s41586-024-07744-y (2024)

    work page 2024

  23. [25]

    Nguyen, T., Brandstetter, J., Kapoor, A., Gupta, J. K. & Grover, A. Climax: A foundation model for weather and climate. ICML(2023)

    work page 2023

  24. [26]

    D.et al.Challenges and benchmark datasets for machine learning in the atmospheric sciences: Definition, status, and outlook.Artif

    Dueben, P. D.et al.Challenges and benchmark datasets for machine learning in the atmospheric sciences: Definition, status, and outlook.Artif. Intell. for Earth Syst.1, e210002, DOI: 10.1175/aies-d-21-0002.1 (2022)

    work page doi:10.1175/aies-d-21-0002.1 2022

  25. [27]

    T.et al.Overcoming the disconnect between energy system and climate modeling.Joule6, 1405–1417, DOI: 10.1016/j.joule.2022.05.010 (2022)

    Craig, M. T.et al.Overcoming the disconnect between energy system and climate modeling.Joule6, 1405–1417, DOI: 10.1016/j.joule.2022.05.010 (2022)

    doi:10.1016/j.joule.2022.05.010 2022

  26. [28]

    Bai, M.et al.Solarseer: Ultrafast and accurate 24-hour solar irradiance forecasts outperforming numerical weather prediction across the usa.arXiv preprint arXiv:2508.03590(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [29]

    Reichstein, M.et al.Deep learning and process understanding for data-driven earth system science.Nature566, 195–204, DOI: 10.1038/s41586-019-0912-1 (2019). 8

    doi:10.1038/s41586-019-0912-1 2019

  28. [30]

    G.et al.Can deep learning beat numerical weather prediction?Philos

    Schultz, M. G.et al.Can deep learning beat numerical weather prediction?Philos. Transactions Royal Soc. A379, 20200097, DOI: 10.1098/rsta.2020.0097 (2021)

    arXiv 2020

  29. [31]

    Lehmann, F.et al.Finetuning a weather foundation model with lightweight decoders for unseen physical processes.arXiv preprint arXiv:2506.19088(2025)

    Pith/arXiv arXiv 2025

  30. [32]

    Carpentieri, A.et al.Data-driven surface solar irradiance estimation using neural operators at global scale.arXiv preprint arXiv:2411.08843DOI: 10.48550/arXiv.2411.08843 (2024)

    Pith/arXiv arXiv doi:10.48550/arxiv.2411.08843 2024

  31. [33]

    Commun.15, 510, DOI: 10.1038/s41467-023-44666-1 (2024)

    Xia, P.et al.Accurate nowcasting of cloud cover at solar photovoltaic plants using geostationary satellite images.Nat. Commun.15, 510, DOI: 10.1038/s41467-023-44666-1 (2024)

    work page doi:10.1038/s41467-023-44666-1 2024

  32. [34]

    Bodnar, C.et al.A foundation model for the earth system.Nature641, 1180–1187, DOI: 10.1038/s41586-025-09005-y (2025)

    doi:10.1038/s41586-025-09005-y 2025

  33. [35]

    Allen, A.et al.End-to-end data-driven weather prediction.Nature641, 1172–1179, DOI: 10.1038/s41586-025-08897-0 (2025)

    doi:10.1038/s41586-025-08897-0 2025

  34. [36]

    Ben Bouallègue, Z.et al.The rise of data-driven weather forecasting: A first statistical assessment of machine learning- based weather forecasts in an operational-like context.Bull. Am. Meteorol. Soc.105, E864–E883, DOI: 10.1175/ BAMS-D-23-0162.1 (2024)

    work page 2024

  35. [37]

    & King, R

    Stengel, K., Glaws, A., Hettinger, D. & King, R. N. Adversarial super-resolution of climatological wind and solar data. Proc. Natl. Acad. Sci.117, 16805–16815, DOI: 10.1073/pnas.1923328117 (2020)

    work page doi:10.1073/pnas.1923328117 2020

  36. [38]

    Hess, P., Drüke, M., Petri, S., Strnad, F. M. & Boers, N. Physically constrained generative adversarial networks for improving precipitation fields from earth system models.Nat. Mach. Intell.4, 828–839, DOI: 10.1038/s42256-022-00540-1 (2022)

    doi:10.1038/s42256-022-00540-1 2022

  37. [39]

    InInternational conference on machine learning, 2806–2823 (PMLR, 2023)

    Bonev, B.et al.Spherical fourier neural operators: Learning stable dynamics on the sphere. InInternational conference on machine learning, 2806–2823 (PMLR, 2023)

    work page 2023

  38. [40]

    & Cohen, Y

    Leinonen, J., Bonev, B., Kurth, T. & Cohen, Y . Modulated adaptive fourier neural operators for temporal interpolation of weather forecasts.arXiv preprint arXiv:2410.18904DOI: 10.48550/arXiv.2410.18904 (2024)

    Pith/arXiv arXiv doi:10.48550/arxiv.2410.18904 2024

  39. [41]

    energy136, 78–111, DOI: 10.1016/j.solener.2016.06

    Antonanzas, J.et al.Review of photovoltaic power forecasting.Sol. energy136, 78–111, DOI: 10.1016/j.solener.2016.06. 069 (2016)

    work page doi:10.1016/j.solener.2016.06 2016

  40. [42]

    Earth & Environ.6, 124, DOI: 10.1038/s43247-025-02042-5 (2025)

    Mardani, M.et al.Residual corrective diffusion modeling for km-scale atmospheric downscaling.Commun. Earth & Environ.6, 124, DOI: 10.1038/s43247-025-02042-5 (2025)

    doi:10.1038/s43247-025-02042-5 2025

  41. [43]

    Letu, H.et al.A new benchmark for surface radiation products over the east asia–pacific region retrieved from the himawari-8/ahi next-generation geostationary satellite.Bull. Am. Meteorol. Soc.103, E873–E888 (2022)

    work page 2022

  42. [44]

    Kalnay, E.et al.The ncep/ncar 40-year reanalysis project.Bull. Am. Meteorol. Soc.77, 437–472, DOI: 10.1175/ 1520-0477(1996)077<0437:TNYRP>2.0.CO;2 (1996)

    work page 1996

  43. [45]

    McGovern, A.et al.Using artificial intelligence to improve real-time decision-making for high-impact weather.Bull. Am. Meteorol. Soc.98, 2073–2090 (2017). 46.Winkler, R. L. Evaluating probabilities: Asymmetric scoring rules.Manag. Sci.40, 1395–1405 (1994). 47.Jose, V . R. R. Percentage and relative error measures in forecast evaluation.Oper. Res.65, 200–2...

    work page 2073

  44. [48]

    & van Sark, W

    Visser, L., AlSkaif, T., Khurram, A., Kleissl, J. & van Sark, W. Probabilistic solar power forecasting: An economic and technical evaluation of an optimal market bidding strategy.Appl. Energy370, 123573 (2024)

    work page 2024

  45. [49]

    A., Martins, G

    Campos, R. A., Martins, G. L. & Rüther, R. Assessing the influence of solar forecast accuracy on the revenue optimization of photovoltaic+ battery power plants in day-ahead energy markets.J. Energy Storage48, 104093 (2022)

    work page 2022

  46. [50]

    Bertsimas, D., Brown, D. B. & Caramanis, C. Theory and applications of robust optimization.SIAM review53, 464–501 (2011)

    work page 2011

  47. [51]

    & Rennen, G

    Ben-Tal, A., Den Hertog, D., De Waegenaere, A., Melenberg, B. & Rennen, G. Robust solutions of optimization problems affected by uncertain probabilities.Manag. Sci.59, 341–357 (2013)

    work page 2013

  48. [52]

    & Choi, D

    Kim, S. & Choi, D. G. A sample robust optimal bidding model for a virtual power plant.Eur. J. Oper. Res.316, 1101–1113 (2024)

    work page 2024

  49. [53]

    Jiang, D. R. & Powell, W. B. Optimal hour-ahead bidding in the real-time electricity market with battery storage using approximate dynamic programming.INFORMS J. on Comput.27, 525–543 (2015). 9

    work page 2015

  50. [54]

    Brown, D. B. & Smith, J. E. Unit commitment without commitment: A dynamic programming approach for managing an integrated energy system under uncertainty.Oper. Res.(2025)

    work page 2025

  51. [55]

    & Chen, Y

    Wu, Y ., Zhang, X., Li, M. & Chen, Y . Evaluation of the global forecast system (gfs) model for surface solar radiation prediction.Renew. Energy203, 25–38, DOI: 10.1016/j.renene.2023.01.067 (2023)

    work page doi:10.1016/j.renene.2023.01.067 2023

  52. [56]

    & Mauzerall, D

    Li, X., Wagner, F., Peng, W., Yang, J. & Mauzerall, D. L. Reduction of solar photovoltaic resources due to air pollution in china.Proc. Natl. Acad. Sci.114, 11867–11872, DOI: 10.1073/pnas.1711462114 (2017)

    work page doi:10.1073/pnas.1711462114 2017

  53. [57]

    Ceferino, L., Lin, N. & Xi, D. Stochastic modeling of solar irradiance during hurricanes.Stoch. Environ. Res. Risk Assess. 36, 2681–2693 (2022)

    work page 2022

  54. [58]

    Global forecast system (gfs) documentation

    NCEP, N. Global forecast system (gfs) documentation. https://www.emc.ncep.noaa.gov/emc/pages/numerical_forecast_ systems/gfs/documentation.php (2021). Accessed: 2025-09-03

    work page 2021

  55. [59]

    A.et al.Atmospheric radiative transfer modeling: A summary of the aer codes.J

    Clough, S. A.et al.Atmospheric radiative transfer modeling: A summary of the aer codes.J. Quant. Spectrosc. Radiat. Transf.91, 233–244 (2005)

    work page 2005

  56. [60]

    Balancing Forecast Accuracy and Switching Costs in Online Optimization of Energy Management Systems

    Genov, E.et al.Balancing forecast accuracy and switching costs in online optimization of energy management systems. arXiv preprint arXiv:2407.03368(2024). 61.Keisler, R. Forecasting global weather with graph neural networks.arXiv preprint arXiv:2202.07575(2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  57. [62]

    & Laine, S

    Karras, T., Aittala, M., Aila, T. & Laine, S. Elucidating the design space of diffusion-based generative models.Adv. neural information processing systems35, 26565–26577 (2022)

    work page 2022

  58. [63]

    The development and evaluation process followed at ECMWF to upgrade the Integrated F orecasting System (IFS)(European Centre for Medium Range Weather Forecasts, 2018)

    Buizza, R.et al. The development and evaluation process followed at ECMWF to upgrade the Integrated F orecasting System (IFS)(European Centre for Medium Range Weather Forecasts, 2018)

    work page 2018

  59. [64]

    F orecast.37, 1069–1084 (2022)

    Zhou, X.et al.The development of the ncep global ensemble forecast system version 12.Weather. F orecast.37, 1069–1084 (2022). 65.Lin, S.-J., Putman, W. & Harris, L. The gfdl finite-volume cubed-sphere dynamical core.GFDL Tech. Note(2017). 66.Hersbach, H.et al.The era5 global reanalysis.Q. journal royal meteorological society146, 1999–2049 (2020)

    work page 2022

  60. [67]

    & Alsafasfeh, Q

    Chang, F., Chen, T., Su, W. & Alsafasfeh, Q. Control of battery charging based on reinforcement learning and long short-term memory networks.Comput. & Electr. Eng.85, 106670 (2020)

    work page 2020

  61. [68]

    C&i (industrial/commercial) energy storage: Guangdong pearl river delta sees tou dif- ferential of 1.2 rmb/kwh (1.49 peak, 0.289 valley)

    InfoLink. C&i (industrial/commercial) energy storage: Guangdong pearl river delta sees tou dif- ferential of 1.2 rmb/kwh (1.49 peak, 0.289 valley). https://www.infolink-group.com/energy-article/ energy-storage-topic-is-ci-storage-demand-boom-on-widening-peak-valley-electricty-price-difference (2023). Peak–valley pricing in Pearl River Delta mid-2023

    work page 2023

  62. [69]

    By february 2025, in 21 provinces, peak–valley electricity price differentials exceed 0.6 rmb/kwh; guangdong peaks at 1.3339 rmb/kwh

    News, B. By february 2025, in 21 provinces, peak–valley electricity price differentials exceed 0.6 rmb/kwh; guangdong peaks at 1.3339 rmb/kwh. https://m.bjx.com.cn/mnews/20250212/1426600.shtml (2025). Nationwide TOU price spread data, Feb 2025

    work page arXiv 2025

  63. [70]

    & Staffell, I

    Schmidt, O., Melchior, S., Hawkes, A. & Staffell, I. Projecting the future levelized cost of electricity storage technologies. Joule3, 81–100, DOI: 10.1016/j.joule.2018.12.008 (2019)

    work page doi:10.1016/j.joule.2018.12.008 2018

  64. [71]

    & Litvinov, E

    Xu, B., Zhao, J., Zheng, T. & Litvinov, E. Factoring the cycle aging cost of batteries participating in electricity markets. IEEE Transactions on Power Syst.33, 2248–2259, DOI: 10.1109/TPWRS.2017.2733339 (2018). 72.National Renewable Energy Laboratory (NREL). Levelized cost of storage analysis. Tech. Rep. NREL/PR-6A20-75721, NREL (2020)

    work page arXiv 2017

  65. [73]

    Grid energy storage technology cost and performance assessment

    Pacific Northwest National Laboratory (PNNL). Grid energy storage technology cost and performance assessment. Tech. Rep., PNNL (2022)

    work page 2022

  66. [74]

    & Klimov, O

    Schulman, J., Wolski, F., Dhariwal, P., Radford, A. & Klimov, O. Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347(2017). 75.Renou, L. & Schlag, K. H. Minimax regret and strategic uncertainty.J. Econ. Theory145, 264–286 (2010). 76.PV Tech. China n-type topcon module prices q1–q3 2025 (2025). Industry market report, accessed 2025. 77....

    Pith/arXiv arXiv 2017

  67. [79]

    Annual technology baseline (atb) 2024

    National Renewable Energy Laboratory (NREL). Annual technology baseline (atb) 2024. Tech. Rep., NREL (2024). Available at: https://atb.nrel.gov. 10

    work page 2024

  68. [80]

    Levelized cost of storage (lcos) and battery degradation methodology

    National Renewable Energy Laboratory (NREL) & Pacific Northwest National Laboratory (PNNL). Levelized cost of storage (lcos) and battery degradation methodology. Tech. Rep., NREL and PNNL (2024). Guidance for storage cost modeling. Competing interests:Ten co-authors (Z.D., B.B., A.C., J.L., T.K., J.S., R.C., Y .Zh., R.Zh. and F.H.) are employees of NVIDIA...

    2024

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.