Pith. sign in

REVIEW 4 major objections 5 minor 5 references

Uncertainty in wind and solar projections depends on global and regional climate models

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read This paper shows that in a 31-member regional climate model ensemble over the Alps, projected wind changes are driven mostly by the driving global models except in mountains, while solar change uncertainty shifts from regional models in sum

desk verdict Useful, honest first application of ANOVA uncertainty decomposition to wind/solar projections in EURO-CORDEX, but the headline attribution maps rest on an unbalanced one-replicate design that is convention-dependent; the qualitative story likely holds but needs a balanced or Type II/III reanalysis to be fully convincing. read the letter →

arxiv 2603.20052 v1 pith:PN3EEX4M submitted 2026-03-20 physics.ao-ph stat.AP

Nina Effenberger , Reto Knutti This is my paper · ORCID
classification physics.ao-phstat.AP PACS 92.60.-e
keywords windenergysolarradiationclimateprojectionsuncertaintydecompositionanalysisofvarianceregionalmodelsglobalAlpineregion
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 a practical question: when an energy planner looks at a multi-model spread in future wind and solar resources, is that spread coming from the choice of global climate model (GCM) or from the regional downscaling model (RCM)? Using 31 paired simulations over the Alps and a two-way analysis of variance, the authors attribute the spread in six aspects of the fields—mean, variance, median, 90th percentile, co-occurring low days, and wind-solar correlation—to GCMs, RCMs, and a residual. They find that for projected changes, the GCM determines most of the wind speed uncertainty away from mountains, while the RCM determines most of the summer solar radiation uncertainty and the GCM the winter solar uncertainty. The result matters because it tells modelers whether to spend computing on more global models or more regional models, and it cautions that historical field skill is not the same as change-uncertainty dominance.

What carries the argument

The central object is the ratio R = SS_RCM/SS_GCM from a two-way analysis of variance applied at every grid point and season. The data form an incomplete 6x6 matrix of regional models crossed with driving global models, with one realization per pair; the ANOVA separates the sums of squares into GCM, RCM and residual terms, with the residual absorbing GCM-RCM interactions and internal variability. This ratio is the quantity that defines the maps of where one model family dominates, and it is the basis for the paper's guidance on ensemble design.

What would settle it

Restrict the analysis to the submatrix of pairs that every regional model shares with every global model (or impute the missing pairs) and recompute the R = SS_RCM/SS_GCM maps; if the seasonal and terrain-dependent dominance patterns change substantially, the original split is an artifact of the missing pairs rather than a property of the models.

Watch

Extended reading notes

Core claim

Analysing 31 regional-global model pairs over the Alpine region, the authors decompose the variance of wind speed and solar radiation fields into GCM and RCM contributions using a two-way additive ANOVA. For absolute historical fields, regional models dominate variability over land for both variables, while global models dominate wind over the sea. For climate change signals between 1995–2004 and 2045–2054, the picture inverts for wind: global models dominate the spread in almost all regions, with regional models contributing more only in a narrow mountainous band. For solar radiation change, the dominant source flips with season—regional models in spring and summer, global models in autumn

Load-bearing premise

The entire decomposition hinges on the assumption that GCM and RCM effects add linearly in an incomplete matrix with one run per pair; if interactions or missing-pair patterns are non-negligible, the variance split between the two model types is not uniquely identifiable.

Editorial extensions

If this is right

  • If a study targets wind power changes over flat terrain, sampling many global models is the efficient way to span uncertainty; one or few regional models may suffice.
  • For solar power changes in summer, an ensemble should include many regional models because they dominate the spread in that season.
  • For winter compound events such as low wind and low solar days (energy droughts), global model diversity matters more than regional diversity.
  • Historical field spread and future change spread have opposite drivers for wind over land, so using historical performance to weight models for projections could mislead.
  • The frequent absence of sign agreement across models (fewer than 80% agree) is not evidence of no change; the spread itself is a measure of uncertainty.

Reading between the lines

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

  • If the additive decomposition is valid, then the residual term—interactions plus internal variability—could be re-partitioned with multi-member initial-condition ensembles, potentially shifting the dominance maps; the paper's fraction is thus a lower-bound attribution of model choice.
  • The seasonally flipping solar result implies that the value of regional downscaling for solar depends on the electricity system's seasonal demand: summer-peaking systems need regional diversity, winter-peaking systems can lean on global models.
  • One could test the mountain-sea contrast by applying the same ANOVA to a flat region (e.g., Northern Germany vs. the Alps) and predicting that the mountain band of RCM dominance for wind shrinks with decreasing topographic roughness.
  • The same decomposition could be applied to next-generation global model ensembles to see whether the dominance patterns persist as GCM resolution and physics improve.
Share X Bluesky LinkedIn Reddit HN

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 quantifies the sources of spread in EURO-CORDEX wind and solar projections over the Alpine region using 31 RCM–GCM pairs. For six aspects (mean, temporal variance, median, 90th percentile, compound drought days, and spatial correlation) and for both historical fields and projected changes, the authors apply a two-way additive ANOVA with GCM and RCM as factors and report the ratio R = SS_RCM/SS_GCM. The main claims are: (i) historical absolute fields are dominated by RCM variability; (ii) projected changes in wind are dominated by GCMs except in mountainous regions; and (iii) projected changes in solar radiation are dominated by RCMs in summer and by GCMs in winter. The paper concludes with ensemble-design guidance for renewable-energy impact studies.

Significance. The question addressed is timely and practically relevant for designing climate model ensembles for energy-system applications. The paper makes good use of the available CORDEX matrix, provides spatially and seasonally resolved maps, and includes a commendable robustness exercise that re-evaluates the main results against ERA5-based model subsets. If the attribution is correct, the results offer concrete, actionable guidance: GCM diversity should be prioritized for wind over flat regions, while RCM diversity matters for summer solar and for mountainous terrain. However, the central attribution rests on a statistical decomposition that is not uniquely identified from the incomplete, one-realization-per-pair design. The manuscript currently provides no uncertainty intervals on variance fractions and no test of the additive-model assumptions, so the headline maps must be regarded as provisional.

major comments (4)
  1. [§2.4, Eqs. (5)–(9); Table 1] The decomposition is convention-dependent. The ANOVA is applied to an incomplete 6×6 matrix (31 of 36 pairs) with a single realization per pair. For unbalanced designs, the sums of squares assigned to main effects depend on the Type I/II/III convention, and the manuscript does not state which convention is used or provide computational details. Consequently, R = SS_RCM/SS_GCM, and in particular the R=1 threshold crossings in Figs. 6–7, are not uniquely identified. Please specify the sum-of-squares type and report sensitivity to Type II vs Type III, or re-analyze a balanced submatrix.
  2. [§2.4, Eqs. (6)–(7); §4] With one realization per pair, SS_res absorbs both GCM–RCM interactions and internal variability. If interactions are non-negligible, the main-effect sums of squares are biased. The manuscript acknowledges this in Section 4 but does not quantify the impact. Please provide a diagnostic, e.g., fit an interaction model on the available pairs or on a balanced subset, and state how large the interaction component would need to be to change the reported R>1/R<1 regions.
  3. [§3.2.1 and Appendix A] The robustness checks only vary the model subset (correlation- and MAPE-based selections). They do not test the sensitivity of the ANOVA results to the unbalanced one-replicate design, the missingness pattern, or the sum-of-squares convention. Thus the claim that the results are 'robust' is too broad. Please add a sensitivity analysis that varies the ANOVA specification, not only the data selection.
  4. [§2.4.1 and Figure 2] The paper acknowledges that internal variability is substantial and that the variance fractions still partially include internal variability. Since there is only one realization per GCM–RCM pair, the estimated f_GCM, f_RCM, and R are noisy; no uncertainty intervals are provided. Without intervals, the spatial threshold maps in Figs. 6–7 may overstate confidence. Please provide bootstrap or other uncertainty quantification, or at minimum a clear statement of the expected sampling uncertainty.
minor comments (5)
  1. [Eq. (4)] The quantity labeled MAE is not mean absolute error; the equation is a signed sum of differences without an absolute value or normalization. Please correct the notation and clarify whether winter solar changes are analyzed with a signed or absolute metric.
  2. [§2.2, solar radiation preprocessing] The exclusion of zero solar radiation values to avoid zero winter medians is described, but it is not stated whether this exclusion is applied consistently to all aspects, seasons, and future-period data. Please clarify, as this could affect the comparability of variance decompositions.
  3. [Figures 6 and A.10–A.13] Several figure captions describe wind-speed panels as 'solar radiation changes.' Please correct the captions.
  4. [Eq. (3) and surrounding text] The notation y^h_t and y^f_t is not formally defined. Since the ANOVA acts on grid-point values, clarify what the time index t represents in the MSPE expression.
  5. [§2.3] The phrase 'allowing correlations to be analyzed consistently with MSPE in the ANOVA framework' is unclear. The Fisher z-transformation is a variance-stabilizing transformation for correlation coefficients and is not directly related to MSPE.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the variance decomposition is descriptive and externally benchmarked via ERA5 subsets.

full rationale

The paper's central claims are derived by applying a two-way additive ANOVA (Eqs. 5–8) directly to the 31 RCM–GCM pair outputs, with no fitted parameters and no target quantity being used to define the inputs. The ratio R = SS_RCM/SS_GCM (Eq. 9) is a descriptive summary of the variance partition, not a prediction forced by construction. The robustness checks select subsets based on independent ERA5 evaluation (Section 2.3 and Appendix A), providing external benchmark support rather than circular self-confirmation. Self-citations to Effenberger et al. [2025] concern drought methodology and emulators, but are not load-bearing for the GCM/RCM attribution. The manuscript openly acknowledges the ANOVA linear-separability assumption, incomplete matrix, and absorbed internal variability (Section 4), which are limitations affecting identifiability and robustness, but they do not amount to circular reasoning. No equation is shown to reduce to its own input, no fitted parameter is relabeled as a prediction, and no load-bearing uniqueness claim is imported from the authors' prior work. Accordingly, no significant circularity is found.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The paper contributes no fitted constants; its main burden is statistical assumptions about the ANOVA design, not physics. The key unverified premises are additivity of GCM/RCM effects, estimability from an incomplete one-realization matrix, and adequacy of 10-year averaging against internal variability.

free parameters (1)
  • Drought threshold = 20th percentile (per season and dataset)
    Chosen by hand in §2.2 to define compound wind/solar lows; controls the 'droughts' aspect and therefore combined-aspect variance results.
assumptions (5)
  • domain assumption GCM and RCM effects are additive and independent in Yij = mu + alpha_i + beta_j + epsilon_ij (Eq. 5), so SS_GCM and SS_RCM are interpretable as separable contributions.
    Central to every figure; interactions are pushed into the residual, so if non-additivity is large, R = SS_RCM/SS_GCM misstates the true drivers.
  • ad hoc to paper The incomplete 31/36 RCM–GCM matrix can be analyzed by two-way ANOVA without imputation and without biasing the sum-of-squares decomposition.
    Table 1 has missing pairs; no type I/II/III specification is given, so the decomposition is not uniquely defined for unbalanced designs.
  • domain assumption Averaging over 10-year periods sufficiently damps internal variability so that remaining variance is dominated by model differences.
    Section 2.4.1 acknowledges internal variability is substantial even for 10- or 20-year mean changes (Figure 2), so this premise is load-bearing but imperfect.
  • domain assumption Single realization per GCM–RCM pair; internal variability is folded into residual and main effects rather than separated.
    Without multiple realizations, ANOVA cannot separate internal noise from model response; authors note this, but the central ratios still mix the two.
  • ad hoc to paper Zero solar radiation values are excluded when computing percentiles to avoid zero winter medians.
    Preprocessing choice in §2.2 affects winter solar aspects; not central to wind claims but changes the analyzed distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Uncertainty in wind and solar projections depends on global and regional climate models." pith.science (2026). https://pith.science/paper/PN3EEX4M

@misc{pith2026260320052,
  author       = {Pith},
  title        = {Pith review of: Uncertainty in wind and solar projections depends on global and regional climate models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PN3EEX4M}},
  note         = {Machine review of arXiv:2603.20052}
}
read the original abstract

Ensembles of regional-global climate model combinations show substantial spread in projected wind and solar resources. Using 31 RCM-GCM pairs, we quantify the sources of this spread with a spatially and seasonally resolved variance decomposition, separating contributions from RCMs and GCMs. For both wind speed and solar radiation, RCMs dominate the variability in the absolute historical fields. In contrast, projected changes in wind speed are largely controlled by the driving GCMs, except in mountainous regions where RCM-induced variance becomes larger than that induced by GCMs. For solar radiation, contributions are strongly season-dependent, with RCMs dominating in summer and GCMs in winter. Our findings support that GCM and RCM variability together define the uncertainty of wind and solar climate projections. This provides guidance for designing climate model ensembles that better support uncertainty-aware energy system decisions under climate change.

Figures

Figures reproduced from arXiv: 2603.20052 by the authors.

Figure 1
Figure 1. Geographic region analyzed in this study, covering the Alps and surrounding areas in Central Europe. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Spatial mean changes for the SMHI/MPI-M-MPI-ESM-LR ensemble between the 1995–2004 and [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Results of the two-way ANOVA on historical solar radiation data between 1995-2004 across the 31 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Results of the two-way ANOVA on historical wind speed data between 1995-2004 across the 31 RCM–GCM [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Results of the two-way ANOVA on combined aspects of solar radiation and wind speed data between 1995- [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Results of the two-way ANOVA on changes of wind speed data between 1995-2004 and 2045-2054. The fig [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Results of the two-way ANOVA on changes of solar radiation data between 1995-2004 and 2045-2054. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Results of the two-way ANOVA on changes of combined aspects between 1995-2004 and 2045-2054. The [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Average shifts in solar and wind energy metrics. Kernel density estimations (KDEs) of the ensemble [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

5 extracted references · 1 linked inside Pith

  1. [2]

    Bridging CORDEX and CMIP6: Machine Learning Downscaling for Wind and Solar Energy Droughts in Central Europe.arXiv preprint arXiv:2512.07429,

    Nina Effenberger, Maxim Samarin, Maybritt Schillinger, and Reto Knutti. Bridging CORDEX and CMIP6: Machine Learning Downscaling for Wind and Solar Energy Droughts in Central Europe.arXiv preprint arXiv:2512.07429,

    arXiv

  2. [2010]

    Marine Herrmann, Samuel Somot, Sandro Calmanti, Clotilde Dubois, and Florence Sevault. Representation of spatial and temporal variability of daily wind speed and of intense wind events over the Mediterranean Sea using dynamical downscaling: impact of the regional climate model configuration.Natural Hazards and Earth System Sciences, 11 (7):1983–2001,

    1983

  3. [2011]

    All datasets, except one, appear in at least one of the subsets

    12 Appendices Appendix A Subset experiments Table 3: EURO-CORDEX RCM–GCM simulation pairs were selected based on correlation and mean absolute per- centage error relative to ERA5.✓marks models with high correlation for solar radiation,✓those with high correlation for wind speed.✓indicates low MAPE for solar radiation, and✓low MAPE for wind speed. All data...

    1995

  4. [2015]

    Climate change impacts on net load under technological uncertainty in European power systems.arXiv preprint arXiv:2512.13461,

    10 Luna Bloin-Wibe, Erich Fischer, Leonard Göke, Reto Knutti, Francesco De Marco, and Jan Wohland. Climate change impacts on net load under technological uncertainty in European power systems.arXiv preprint arXiv:2512.13461,

    arXiv

  5. [2020]

    Assessing future wind energy potential under climate change: The critical role of multi-model ensembles in robustness assessment.arXiv preprint arXiv:2505.24463,

    Andrea Lira-Loarca, Francesco Ferrari, and Andrea Mazzino. Assessing future wind energy potential under climate change: The critical role of multi-model ensembles in robustness assessment.arXiv preprint arXiv:2505.24463,

    arXiv

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.