{"id":"943ebbab-2d1c-43dc-a46d-e1c3be543b6a","arxiv_id":"2505.12538","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Under limited weather foresight, long-duration storage stockpiles against extreme states, solar PV capacity rises by up to 25% and electricity price duration curves smooth compared with perfect-foresight planning.","lead":"Long-duration storage behaves like insurance when grid planners cannot foresee next month's weather, stockpiling energy before winter in a way perfect-foresight models miss. A new modeling comparison for Germany, Spain and the UK finds this shifts investment toward solar and smooths electricity price curves in fully renewable systems.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'stockpiling effect' and the spring/summer LDES bidding curves are substantially shaped by the Section 2.3 storage-target terminal penalty, not by limited foresight alone; an infinite-horizon or continuation-value test is needed before the headline LF-vs-PF differences can be attributed to…","rationale":"The paper is a serious and transparent attempt to isolate limited foresight in LDES operations and capacity planning. The comparison design (type D versus type C) is appropriate, the SDDP implementation is coupled to open-source code and data, and the paper explicitly flags its main limitations, including the end-of-horizon effect in Sections 2.3, 4.3, and 5.3. The reader's verdict of CONDITIONAL is well calibrated: the central qualitative mechanism is internally consistent, but the load-bearing storage-target assumption is acknowledged to bias dispatch and capacity results. My stress-test pass lands on the same weakest point as the reader: the terminal penalty at the end of June is not a neutral device, because it directly shapes the stockpiling pattern and the spring/summer bidding curves that support the paper's headline claims about LDES as a defensive hedge and about solar gaining value. Both models include the target, but the asymmetry in how PF and LF can satisfy it amplifies the apparent foresight effect. The paper's own Section 4.3 observation that MSVs rise rapidly as a target shortfall becomes more likely, and its Limitations statement that an infinite-horizon formulation would smooth the seasonal bid curves, confirm that this is not a minor modeling detail. A targeted reformulation test would determine whether the quantitative capacity and price-formation findings survive once the artificial cliff is replaced by a continuation value. I therefore recommend no change to the reader's CONDITIONAL verdict.","tokens_in":30441,"tokens_out":6257,"duration_ms":72462,"concrete_test":"Replace the terminal penalty with a continuation value from a multi-year or infinite-horizon formulation (e.g., the discounted infinite-horizon SDDP of Hole et al. 2025) or, minimally, a two-year model in which the second year's June-end storage level feeds a value function instead of a VOLL cliff. If the LF stockpiling peak, the spring/summer MSV curves, and the solar/wind capacity deltas of Figure 5 shift materially under this change, the storage target is driving the headline results. A cheaper diagnostic: move the target month from June to, say, March and re-solve for Germany under No Imports; if the LF stockpiling peak follows the target date rather than the winter climatology, the terminal penalty is the active driver.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing premise is the storage target mechanism in Section 2.3. The LF model chooses an initial storage level e_ini and penalizes any end-of-June shortfall at VOLL (SI.1p-q). The paper concedes this is 'a coarse approximation of the marginal value of storage beyond the last stage' and can create an 'end-of-horizon effect.' Section 4.3 makes the mechanism explicit: in June, the MSV is 100,000 EUR/MWh below the target and 0 above, and even earlier months see MSVs 'increase rapidly as a shortfall against the storage target becomes more likely.' This means the convex spring curves and the stockpiling buildup are partly a hedge against an artificial terminal cliff, not purely against future winter scarcity. The PF model has the same target, but with perfect foresight it can meet it without holding the same precautionary margin, so the LF-PF difference in storage behavior is inflated by the target. If the marginal value of storage beyond June were represented by a continuation value (e.g., an infinite horizon as in Hole et al. [37]), the autumn curves would be less concave, spring/summer bids would not collapse to the VOLL step, and the computed solar/wind capacity shifts (up to +25%/-20%) could change materially. The paper's own Limitations section (5.3) states the same: 'seasonal differences between bidding curves would likely be less pronounced as bids come down for autumn months and come up for spring and early summer months.' Thus the headline mechanism is not yet cleanly separated from the terminal condition.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a multi-stage stochastic capacity expansion model of a fully renewable, sector-coupled energy system with long-duration hydrogen storage, solved with Stochastic Dual Dynamic Programming (SDDP), and compares it against a two-stage perfect-foresight benchmark. The central methodological contribution is a KKT-based derivation that represents LDES dispatch as bidding the efficiency-adjusted expected marginal storage value (MSV), conditional on storage level and time of year. The case study covers Germany, Spain, and the UK under three backup-availability scenarios. The headline findings are that, under limited weather foresight, LDES operates defensively and stockpiles to hedge extreme future states; solar PV gains capacity (up to about 25% in the UK) at the expense of onshore wind; and energy-only price duration curves become smoother because LDES bids reflect the weather-dependent probability of scarce states. The paper is transparent about its simplifications and provides code and data.","tokens_in":30798,"tokens_out":4583,"duration_ms":53031,"significance":"If the results hold, the paper makes a significant contribution to capacity expansion modeling for very high renewable shares: it moves from weather variability to weather uncertainty, gives a clean analytical bridge between stochastic dynamic programming and storage bidding behavior, and produces concrete, testable claims about capacity mixtures and price formation. The derivation in Section 2.4 (Equations 2a-2e) is a genuine strength, and the comparison of model types C and D is a well-designed natural experiment for isolating the role of foresight. The availability of open-source code and data is another positive element. The main residual risk is that the headline stockpiling and bidding-curve results are partly generated by the artificial terminal storage target, a limitation the paper itself identifies.","major_comments":[{"comment":"The storage-target terminal condition is load-bearing for the central claims, not a neutral modeling detail. Under Eq. SI.1p-q, any end-of-June storage shortfall relative to the endogenously chosen initial level is penalized at the value of lost load (100,000 EUR/MWh). Section 4.3 states that in June the MSV is 100,000 EUR/MWh below the target and 0 above it, and that even in earlier months MSVs 'increase rapidly as a shortfall against the storage target becomes more likely.' This means the spring/summer shape of the bidding curves in Figure 6 and the stockpiling trajectories in Figure 4 are driven in part by an artificial terminal cliff, not purely by limited foresight about winter scarcity. The paper itself concedes in Section 5.3 that under an infinite-horizon or continuation-value treatment 'seasonal differences between bidding curves would likely be less pronounced.' Because the perfect-foresight model can meet the same target without the same precautionary margin, the LF-PF differences in storage behavior and the resulting capacity shifts (up to +25% solar, -20% wind in the UK) are inflated by the terminal penalty. A sensitivity analysis with a reduced terminal penalty, or better a continuation-value/infinite-horizon treatment along the lines of the cited Hole et al. work [37], is needed before the headline effects can be attributed to limited foresight itself rather than to the end-of-horizon construction.","section":"Section 2.3, SI.1p-q, Section 4.3"},{"comment":"The SDDP runs are stopped by iteration (15,000) or wall-clock (54 h) limits, with only an ex-post inspection of the lower bound and of capacity trajectories as a convergence guarantee. Table SI.1 shows that two specifications actually stop on the time limit, and SI.3 reports a rolling mean of forward-pass simulations rather than a confidence interval for the upper bound. The paper presents the resulting capacity differences as model outputs, so the absence of a quantitative convergence diagnostic is a concern for the magnitude of the headline shifts. The statement that capacity decisions stabilize after about 3,000 iterations is encouraging, but the refining of the policy for another 12,000 iterations could still change the marginal storage values used in the bidding curves of Section 4.3. At minimum, please report the lower-bound/upper-bound gap or a sensitivity check showing that the capacity results and bidding curves do not materially change if training is extended or restarted with a different random seed.","section":"Section 3.4 and SI.3, Table SI.1"}],"minor_comments":[{"comment":"The notation H(t)\\{H} in Equation (2c) is confusing: H(t) is a set of periods, and writing '\\{H}' suggests a single period. Please write H(t)\\setminus\\{H(t)\\}, or define H as the last period of the stage.","section":"Section 2.4, Eq. (2c)"},{"comment":"The phrase 'fishing constraint' appears to be a typo for 'fixing constraint' or 'coupling constraint'. Please correct it.","section":"SI.2, Eq. (SI.3c)"},{"comment":"The example of '127 EUR/MWh at 80% storage in August' should state explicitly that it refers to the No Imports scenario in Germany and to the MSV in EUR/MWh_H2. The y-axis break makes cross-scenario comparison difficult; consider a log-scale inset for the high-MSV range.","section":"Figure 6 and Section 4.3"},{"comment":"The claim that one-month foresight likely produces a lower bound on the effect of limited foresight is plausible but not demonstrated. A short-horizon sensitivity (e.g., two-week stages for one country) would substantially strengthen this argument, even if only for a subset of scenarios.","section":"Section 2.2 and Section 5.3"},{"comment":"The statement that under perfect foresight 'the maximum dispatch event across all weather years defines the storage size' is a useful intuition but is stated more strongly than the evidence shown. Consider phrasing it as a tendency observed in the simulations.","section":"Section 4.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is honestly written and the analytical core is sound, but the storage-target mechanism is the main threat to the headline attribution. The author's own limitation paragraph largely concedes the point; I would expect either an additional experiment that isolates the effect of the terminal penalty or a reframing of the claims so that the stockpiling effect is presented as conditional on the target-based interannual closure. The convergence diagnostics are secondary but should be addressed with quantitative measures."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is the full multi-stage SDDP capacity expansion model with intra-stage dispatch and hydrogen LDES, compared against a two-stage perfect-foresight benchmark. Previous work either used heuristic bidding rules or heavily reduced scenario trees. The paper also derives LDES bidding curves from the KKT conditions of the SDDP subproblems, which is a clean way to show that limited-foresight bids reflect probability-weighted extreme-state costs. I believe the central mechanism is sound: under limited foresight the LDES acts more defensively, stockpiles earlier, and the resulting price duration curves smooth out. That is a real and useful insight for energy-only market design debates.\n\nThe paper earns credit for being honest about its own weak spots. It ships open code and results data, it tests stagewise independence with autocorrelations, and it includes a limitations section that explicitly concedes the end-of-horizon effect. The comparison design is appropriate, and the KKT derivation of marginal storage value (Equations 2a-2e) holds up.\n\nThe soft spot is exactly where the stress-test note points. The storage target and the VOLL penalty for missing it at the end of June are doing real work. Section 4.3 shows the June MSV is a step function at 100,000 EUR/MWh below the target and zero above, and even earlier months have MSVs that ramp up as a shortfall becomes likely. So the convex spring curves and the stockpiling buildup are partly a hedge against an artificial terminal cliff, not purely against winter scarcity. The PF model, with perfect foresight, meets the same target without holding the same precautionary margin, which inflates the LF-PF storage difference. The paper's own limitations text admits that an infinite-horizon formulation would bring autumn bids down and spring bids up. That means the headline capacity shifts (up to +25% solar, -20% wind in the UK) could change materially. This is not a fatal flaw, but it means the magnitudes should be treated as conditional until someone does a continuation-value or multi-year sensitivity.\n\nTwo smaller concerns: SDDP is stopped by time and iteration limits with only ex-post convergence checks, which is standard practice but not a certified solution. And the single-node, no-interconnection setup likely exaggerates LDES value. Both are acknowledged. The stagewise independence test is fine for the variables actually used; excluding hydro reservoirs is a reasonable move given their autocorrelation.\n\nWho benefits: energy system modelers working on storage and weather uncertainty, and people arguing about whether energy-only markets can survive high renewable shares. Worth a serious referee, definitely. I would not desk-reject. The referee should push for an infinite-horizon or continuation-value sensitivity before the headline numbers are used in policy work. The mechanism will probably survive, but the magnitudes need to be earned.","headline":"Solid first application of full multi-stage SDDP to capacity expansion with LDES; the limited-foresight vs perfect-foresight comparison is well-designed, but the storage-target terminal penalty contaminates the headline stockpiling and bidding-curve results, so the magnitudes need a continuation-value treatment before they can be taken at face value.","tokens_in":31294,"tokens_out":1358,"would_cite":true,"duration_ms":16904,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Limited weather foresight changes long-duration storage from arbitrageur to defensive hedge.","keywords":["long-duration energy storage","weather uncertainty","limited foresight","stochastic programming","capacity expansion","marginal storage value","price formation","renewable electricity markets"],"falsifier":"Re-solve the UK No-Imports case with a discounted infinite-horizon stochastic formulation (or with two linked weather years) and compare the storage-level distributions and bidding curves. If the October-to-full stockpiling and the convex spring bidding curves persist, the limited-foresight mechanism is robust; if they flatten or shift, the paper's headline results are driven by the terminal target rather than by weather uncertainty.","tokens_in":30230,"feed_emoji":"🔋","tokens_out":10142,"duration_ms":95936,"temperature":0.7,"pith_summary":"The paper asks whether capacity-expansion models that assume perfect weather foresight misrepresent the role of long-duration energy storage (LDES). It argues they do: when operators cannot see months ahead, LDES stops acting as an arbitrageur and instead stockpiles energy in autumn and early winter as a defensive hedge against extreme weather states. A stochastic limited-foresight model of fully renewable systems in Germany, Spain, and the UK is compared with a perfect-foresight benchmark, and the paper finds that solar PV gains system value (roughly 25% more capacity in the UK) while onshore wind loses some, and that LDES bids smooth electricity price duration curves. If the claim is right, planning studies that assume perfect foresight overstate wind and understate solar, and extreme price disparity in energy-only markets is partly an artifact of that assumption.","feed_headline":"Limited weather foresight turns long-duration storage into a hedge","feed_subtitle":"Stochastic planning raises solar capacity by 25% and smooths price spikes in a wind-heavy system.","key_machinery":"The load-bearing object is the marginal storage value (MSV), the derivative of the expected future cost-to-go with respect to the storage level at the end of a dispatch stage. The recursive identity $\\lambda^{\\mathrm{MSV}}_{s,t,h} = \\sum_{j=h}^{H(t)}(\\mu^e_{s,t,j}-\\bar\\mu^e_{s,t,j}) + \\mathbb{E}[\\lambda^{\\mathrm{MSV}}_{s,t+1,1}]$ expresses the MSV in a given month and hour as current-period storage-bound effects plus the conditional expectation of next month's MSV, so today's value of stored energy already contains the weather-dependent probability of all future scarcity events. At the margin, discharging sets the electricity price at $\\lambda_{t,h}=(\\eta^f_s)^{-1}\\lambda^{\\mathrm{MSV}}_{s,t,h}$ and charging at $\\lambda_{t,h}=\\eta^h_s\\lambda^{\\mathrm{MSV}}_{s,t,h}$. The model approximates the expected cost-to-go functions with a cutting-plane algorithm, then evaluates their slopes over the storage domain in steps of 10 GWh to produce monthly bidding curves; those curves are concave in autumn, convex in spring, and their height is set by the cost of the outside option (load shedding or hydrogen imports).","core_discovery":"The central claim is that weather uncertainty changes both the operation and the system value of long-duration storage. In a fully renewable, sector-coupled system with limited foresight, the optimal LDES policy is defensive: the storage builds toward full capacity by October, holds high levels through winter, and only then returns to its target level, so that the distribution of storage trajectories lies well above the perfect-foresight distribution. The bidding logic that produces this behavior is the marginal storage value: a storage unit at the margin bids the efficiency-adjusted expected value of stored energy, and that expectation is the probability of future extreme scarcity (when storage would be empty and load must be shed or hydrogen imported) times the cost of that state. Perfect foresight removes the probability term, allowing aggressive, year-specific dispatch that perfect-foresight models mistake for optimal operation. The paper's quantitative claims include up to 25% higher solar PV capacity and 20% lower onshore wind in the United Kingdom under limited foresight, little change in LDES capacities, and smoother price duration curves because LDES bids vary continuously with storage level and season.","pith_inferences":["Editorial extension: because the paper's operator minimizes expected cost and is risk-neutral, the smoothing of price duration curves is a central-case result; a risk-averse operator would weight tail states more heavily, so the hedging and price-smoothing effects would likely be stronger, not weaker.","Editorial extension: the results imply that the system value of a generation technology under limited foresight depends on the predictability of its output in the storage-filling season, not only on its annual energy; technologies with narrow weather variance should command a premium in planning models.","Editorial extension: the bidding-curve framework could be reused to value interconnectors, demand response, or firm low-carbon backup, since each lowers the effective cost of the outside option and should flatten the autumn bidding curves.","Editorial extension: a testable prediction is that systems with higher wind variance and weaker interconnection will show larger solar-for-wind substitution and higher optimal autumn storage levels; comparing weather-year variances across countries would give a direct check."],"forward_implications":["Capacity-expansion studies that assume perfect foresight likely overestimate onshore wind and underestimate solar PV; in the wind-dominated UK case, limited foresight raises solar capacity by about 25% and cuts onshore wind by about 20%.","Limited-foresight LDES trajectories are systematically higher in autumn and early winter than perfect-foresight trajectories, and the stockpiling difference is largest in wind-dominated systems and nearly disappears when cheap hydrogen imports provide backup.","When LDES sets the electricity price, its bid is the efficiency-adjusted expected marginal storage value, so prices embed the weather-dependent probability of extreme states and price duration curves are smoother than in perfect-foresight models.","LDES capacities themselves change little between the two models; the main adjustments are more solar, less onshore wind, and more battery energy capacity (about 86% more in the UK).","The derived monthly bidding curves give a concrete starting point for heuristic LDES bidding rules in large-scale planning models that cannot solve a full stochastic program."],"supporting_citations":[{"why":"Supplies the multi-stage stochastic programming algorithm used to solve the limited-foresight model.","marker":"[36]"},{"why":"Provides the storage bidding formulation and the energy-only price formation setting the paper extends to weather uncertainty.","marker":"[32]"},{"why":"Recent capacity-expansion implementation with an infinite-horizon stochastic model, used as the alternative to the paper's terminal storage target.","marker":"[37]"},{"why":"Hydroelectric bidding literature that defines the water value or marginal storage value concept behind the LDES bidding curves.","marker":"[34]"},{"why":"Documents how extreme events and interannual variability set storage requirements in 100% renewable systems.","marker":"[13]"},{"why":"Shows Dunkelflaute-driven storage needs in Europe and supplies the compressor-demand treatment used in the case study.","marker":"[15]"},{"why":"Establishes that single-year capacity choices can be highly suboptimal under weather variability, motivating the stochastic comparison.","marker":"[29]"},{"why":"Provides sequential multi-year storage trajectories under perfect foresight that the paper contrasts with limited-foresight stockpiling.","marker":"[19]"}],"fun_headline_variants":["Weather uncertainty turns long-duration storage into a defensive hedge","Limited foresight boosts solar by 25% and trims wind, says stochastic model","Stockpiling storage under weather uncertainty: a hedge against extremes","Probabilistic storage bids smooth prices and favor solar over wind"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the value of storage beyond the modeled year can be represented by a single end-of-June target level chosen by the planner, with any shortfall penalized at the value of lost load; if that artificial target is replaced by a proper continuation value, the stockpiling pattern and the spring and summer bidding curves could change, and with them the headline capacity results.","fun_headline_variants_meta":{"raw":{"variants":["Weather uncertainty turns long-duration storage into a defensive hedge","Limited foresight boosts solar by 25% and trims wind, says stochastic model","Stockpiling storage under weather uncertainty: a hedge against extremes","Probabilistic storage bids smooth prices and favor solar over wind"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000811,"raw_usage":{"total_tokens":3568,"prompt_tokens":967,"completion_tokens":2601,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":583,"completion_tokens_details":{"reasoning_tokens":2526}},"tokens_in":583,"tokens_out":2601,"duration_ms":20808,"temperature":1.0,"reasoning_tokens":2526,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:32:04.543293+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-solve the UK No-Imports case with a discounted infinite-horizon stochastic formulation (or with two linked weather years) and compare the storage-level distributions and bidding curves. If the October-to-full stockpiling and the convex spring bidding curves persist, the limited-foresight mechanism is robust; if they flatten or shift, the paper's headline results are driven by the terminal target rather than by weather uncertainty.","supporting_citations":[{"cited_title":"Price formation without fuel costs: The interaction of demand elasticity with storage bidding","cited_arxiv_id":null,"evidence_quote":"Provides the storage bidding formulation and the energy-only price formation setting the paper extends to weather uncertainty."},{"cited_title":"Capacity planning of renewable energy systems using stochastic dual dy- namic programming","cited_arxiv_id":null,"evidence_quote":"Recent capacity-expansion implementation with an infinite-horizon stochastic model, used as the alternative to the paper's terminal storage target."},{"cited_title":"Optimal Bidding Strategies for Hydro-Electric Producers: A Literature Survey","cited_arxiv_id":null,"evidence_quote":"Hydroelectric bidding literature that defines the water value or marginal storage value concept behind the LDES bidding curves."},{"cited_title":"Designing low-carbon power systems for Great Britain in 2050 that are robust to the spatiotemporal and inter-annual variability of weather","cited_arxiv_id":null,"evidence_quote":"Establishes that single-year capacity choices can be highly suboptimal under weather variability, motivating the stochastic comparison."}],"review_version":1}