{"id":"6f83b14e-4734-4c60-ae20-b7ea17e9e10b","arxiv_id":"2411.10768","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A dynamic land-biosphere reservoir, with land-use emissions reducing land storage capacity one-for-one, raises 2100 atmospheric carbon by about 6%, warming by 0.2°C, and the social cost of carbon by 12-14% in a DICE-2016 model.","lead":"This paper builds interpretable carbon-cycle emulators for economic climate models, comparing three-box, four-box, and dynamic land-use variants inside DICE-2016. It finds that letting deforestation shrink the land carbon sink raises projected warming and the social cost of carbon, and shows that pattern-scaling choices shift regional damage estimates.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (7) sets r=1 for land-use-driven loss of land carbon storage, and the paper itself warns r could differ; because the entire 4PR-X-vs-static divergence (80 GtC, 0.2°C, 12–14% SCC) scales with r, the headline numbers are conditional on an uncalibrated assumption.","rationale":"The reader's weakest_assumption correctly identifies the r=1 assumption in Eq. (7) as the load-bearing element. My reading confirms this: the static 4PR model adds a land box but produces essentially no change in outcomes, so the entire 4PR-X effect is attributable to the time-dependent land capacity, which is driven by the uncalibrated r parameter. The paper itself flags that r could differ from 1, but presents the r=1 results as the headline quantitative findings without a sensitivity analysis. This is an external-validity concern rather than an internal inconsistency: the implementation of Eq. (7) is coherent, and the direction of the effect is physically sensible, but the magnitude is not constrained by the calibration data. The pattern-scaling contribution is an independent demonstration using the Lynch et al. (2017) library and is not affected by this concern. Because the framework and calibration machinery are valuable and the caveat is acknowledged in the text, the appropriate disposition is to keep the reader's CONDITIONAL verdict: the qualitative claim is supported, but the quantitative SCC and warming numbers should not be taken as a central estimate until the r sensitivity is quantified. No new concern beyond the one already identified by the reader was found, so the verdict should remain unchanged.","tokens_in":40403,"tokens_out":7345,"duration_ms":80501,"concrete_test":"Re-solve the Section 4.2 BAU and Section 4.3 optimal-policy cases with Eq. (7) evaluated at r = 0, 0.25, 0.5, 0.75, and 1, holding all other parameters and the land-use emission path fixed, and report 2100 atmospheric carbon, temperature, and SCC relative to 3SR. If the SCC uplift at r = 0.5 falls substantially below the r = 1 value (e.g., below one-third or one-half of the 12–14% claim), the headline numbers should be reframed as a scenario-specific upper bound rather than a central estimate.","verdict_should_be":"UNCHANGED","load_bearing_attack":"All policy-relevant differences between 4PR-X and the static 3SR/4PR emulators flow through Eq. (7), tilde-m_L^{t+1} = tilde-m_L^t - r e_t^L, with r fixed at 1 (Section 2.1). When tilde-m_L declines, the land-to-atmosphere coefficient A_{1,4} = A_{4,1} · tilde-m_A / tilde-m_L increases (Eq. 6 and Eq. 34), so more carbon is rerouted to the atmosphere; this is the mechanism that produces the 0.2°C warming, the 80 GtC atmospheric burden difference, and the 12–14% SCC uplift. The ratio r is not estimated from data. The pulse-decay calibration constrains only atmospheric decay and the ocean/land split of a synthetic pulse (q3 in Section 3.1.1), not how land-use emissions affect long-run land storage capacity. The paper explicitly acknowledges in Section 2.1 that one-to-one correspondence may not be guaranteed and that r could differ from 1, but no sensitivity analysis follows. Because the static 4PR model changes nothing relative to 3SR, the entire quantitative contribution of the land-use channel is a linear scaling of r. If the true r is, for example, 0.5, the warming and SCC effects would plausibly shrink roughly in proportion, potentially halving the headline numbers. The qualitative direction is plausible, but the quantitative headline is not yet established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an open-source framework for building interpretable, physically constrained linear box-model carbon-cycle emulators for use in economic integrated assessment models. Three emulators are constructed and compared within a DICE-2016-style model: a three-reservoir model (3SR), a four-reservoir model with a static land box (4PR), and a four-reservoir model with a dynamic land capacity that declines with land-use emissions (4PR-X). The authors find that 3SR and 4PR give nearly identical results, while 4PR-X produces substantially higher atmospheric carbon, temperature, and social cost of carbon by 2100. A second contribution propagates global-mean temperature projections to regional scales via pattern scaling using the Lynch et al. (2017) CMIP5 pattern library, and illustrates how choices of warming pattern and observational baseline affect regional damage estimates.","tokens_in":40763,"tokens_out":3274,"duration_ms":34363,"significance":"If the quantitative claims are upheld, the paper makes a useful methodological contribution: it provides a transparent, reproducible, and physically interpretable alternative to black-box climate emulators, with code publicly available and validation against the Joos et al. (2013) pulse-decay benchmark as well as ZECMIP and RCP-style tests. The finding that a static land box hardly changes DICE outcomes while a dynamic land-capacity representation materially increases warming and SCC is a policy-relevant result for the IAM community. The pattern-scaling module is a useful plug-and-play addition, clearly separating emulator, pattern, and baseline uncertainties. However, the headline quantitative results for 4PR-X are driven by an uncalibrated parameter, and the accuracy claims in the abstract and Section 3.1.4 are contradicted by the paper's own Appendix A.2, so the current version does not yet establish the magnitude of the claimed effects.","major_comments":[{"comment":"The central quantitative result—approximately 80 GtC higher atmospheric carbon, 0.2°C extra warming, and 12–14% higher SCC in 4PR-X relative to the static models—is generated entirely by the assumption r=1 in Eq. (7), which sets the decline in equilibrium land-biosphere capacity equal to one-for-one land-use emissions. The paper explicitly acknowledges in Section 2.1 that 'a one-to-one correspondence ... may not be guaranteed' and that 'r could be different from 1,' yet no sensitivity analysis is reported. Because all differences between 4PR-X and the static models scale with r, the quantitative headline is conditional on this uncalibrated assumption. I request a sensitivity analysis over a plausible range of r (e.g., 0.25, 0.5, 0.75, 1) for the BAU and optimal mitigation runs, and reporting of how the 2100 atmospheric carbon, temperature, and SCC differences respond. If results shrink substantially for lower r, the abstract and conclusions should be reframed as qualitative.","section":"Section 2.1, Eq. (7); Tables 3-5"},{"comment":"The abstract and Section 3.1.4 claim that the three- and four-box emulators 'reproduce the historical and long-run evolution of atmospheric CO2 and global temperature to within about 5% and 3%, respectively.' This is contradicted by Appendix A.2, which states that the PI-calibrated 3SR and 4PR models 'systematically underestimate atmospheric CO2' under RCP scenarios, and Figure 21 shows these models falling well below the CMIP5 RCP concentration trajectories. The 5%/3% accuracy figure appears to refer to the atmospheric-pulse-decay fit error (Figures 15 and 24), not to historical or scenario concentration accuracy. Please clarify precisely which metric the accuracy claim refers to, and either qualify or remove the broad claim.","section":"Abstract; Section 3.1.4; Appendix A.2, Figure 21"},{"comment":"The 4PR-X model's equilibrium land-biosphere mass in 2015 is 258 GtC (Table 11), far below the Ciais et al. (2014) active land-pool estimate of ~550 GtC cited in Section 3.1.1, and below the 4PR value of 387 GtC. The paper acknowledges this in Section 4.2, noting that the low 2015 land content is 'slightly outside the estimated range.' This implausibly low land pool is a direct consequence of applying r=1 cumulatively to historical land-use emissions, and it contributes to the elevated atmospheric burden in 4PR-X. The paper should discuss whether this initial-condition distortion unduly inflates the projected differences, and whether an r<1 calibration or a cap on the equilibrium-mass decline would yield more realistic land-pool trajectories.","section":"Appendix A.2, Table 11, Section 4.2"}],"minor_comments":[{"comment":"The table header uses '3PR' as the column label, which is inconsistent with the '3SR' nomenclature used throughout the paper.","section":"Table 6"},{"comment":"The caption states that 'the q1 penalty function enforces an approximately equal mass of carbon absorption between the oceans and the land biosphere,' but this is the role of the q3 penalty (reservoir absorption ratios), not q1 (dynamic timescales). Please correct the reference.","section":"Appendix A.1.1, Figure 16 caption"},{"comment":"The comparison of anchoring choices in Table 7 conflates two differences: the baseline period (1961-1990 for model climatologies vs 1991-2020 for ERA5) and the use of model versus observed climatology. The statement that the anchor choice can shift 2100 regional means by 'up to 2.5-3°C' should be decomposed into these two effects, since the baseline-period difference alone can explain part of the spread.","section":"Section 5.2, Table 7"},{"comment":"There is a typo in the paragraph above Figure 12: 'model undertainty' should be 'model uncertainty.'","section":"Figure 12 text"}],"recommendation":"major_revision","confidential_remarks":"The paper's main contribution is the 4PR-X mechanism, but the quantitative magnitude rests entirely on the uncalibrated r=1 assumption. The authors should be pushed to provide sensitivity analysis or empirical grounding for r before the central claim can be accepted. The accuracy claim inconsistency with Appendix A.2 is also important to resolve. The pattern-scaling part is more straightforward and could be separated if the carbon-cycle part needs longer revision. No concerns about novelty or attribution; the prior Folini et al. (2024) work is clearly cited."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this for the framework, not for the headline SCC numbers. The paper builds a clean, open-source toolbox for linear multi-reservoir carbon-cycle emulators, calibrates the 3SR and 4PR versions to the Joos pulse-decay data, and then evaluates them inside DICE-2016. That part is solid and reproducible: the calibration is transparent, the penalties are explained, and the code is public. The pattern-scaling half is also useful. It separates emulator, ESM-pattern, and baseline-anchor uncertainty, and shows those can flip regional winners and losers. That is a real contribution.\n\nThe new piece is 4PR-X, a land box whose equilibrium capacity shrinks with land-use emissions. I agree with the stress-test note: the entire divergence from the static models—80 GtC, 0.2°C, 12–14% SCC—flows through Eq. (7) with r fixed at one. The paper itself says one-to-one correspondence may not be guaranteed and r could differ, but no sensitivity is run. The static 4PR changes nothing, so the headline effect is essentially an assumption rather than an estimated quantity. The direction is plausible: deforestation removes a sink, so more carbon stays in the atmosphere. But the magnitude is uncalibrated. A referee should ask for a sensitivity sweep over r, or the paper should present 4PR-X as a scenario rather than as a quantitative estimate.\n\nThere is also a smaller accuracy-language problem. The abstract says the three- and four-box models reproduce historical and 500-year trajectories within 5% and 3%; that claim seems to come from pulse-decay fit errors, while Appendix A.2 shows PI-calibrated static models systematically underestimate CO2 under RCP scenarios. The wording overstates what the validation demonstrates.\n\nThe citation pattern is fine—they engage the climate-science literature properly—and I do not see a serious methodology problem in the calibration or the economic solution. This is honest, clear work; it just needs the central assumption labeled and tested.\n\nVerdict: send to peer review. It deserves referee time, but conditional on adding r-sensitivity and rewriting the accuracy claim. I would bring it to a reading group and cite the framework, while treating the land-use numbers as illustrative pending calibration.","headline":"A transparent, reproducible emulator framework whose static-box calibration deserves publication, but whose headline land-use SCC results rest on an acknowledged, uncalibrated r=1 assumption and should be treated as a scenario until sensitivity analysis is added.","tokens_in":41307,"tokens_out":2640,"would_cite":true,"duration_ms":41782,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Representing deforestation-driven loss of land carbon storage raises the optimal social cost of carbon by up to 17 US$ per tonne of CO2 within a DICE-2016-style model.","keywords":["climate emulators","carbon cycle box models","integrated assessment models","social cost of carbon","land-use change","pattern scaling","DICE-2016"],"falsifier":"Estimate r directly from data: compare cumulative land-use emissions against observed losses of global vegetation and soil carbon stocks; an r of, say, 0.3 instead of 1 would cut the 0.2 °C and 17 USD effects proportionally. Equivalently, drive the 4PR-X emulator with historical land-use emissions from 1850 onward and check whether the simulated atmospheric CO2 trajectory remains within the observed ice-core and Mauna Loa record; a systematic divergence would falsify the one-for-one sink-loss mechanism.","tokens_in":1764,"feed_emoji":"🌍","tokens_out":2771,"duration_ms":77402,"temperature":0.7,"pith_summary":"This paper argues that the way a climate emulator represents the land biosphere is not a neutral modeling detail: it changes the optimal climate policy. Within a DICE-2016-style integrated assessment model, the authors compare a three-box carbon cycle, a four-box version with a static land reservoir, and a four-box version whose land storage capacity shrinks one-for-one with land-use emissions. The static land box leaves outcomes essentially unchanged, but the shrinking-sink version leaves nearly 80 GtC more carbon in the atmosphere by 2100, adds roughly 0.2 °C of warming, and raises the optimal social cost of carbon by up to 17 USD per tCO2 (roughly 12–14%). The same framework, through pattern scaling, also quantifies how the choice of regional warming pattern and present-day baseline propagates into local damages. The authors conclude that omitting deforestation-driven loss of land carbon storage causes systematic underpricing of carbon.","feed_headline":"Deforestation lifts optimal carbon price by up to 17 US$","feed_subtitle":"Dynamic land-sink carbon cycle adds 0.2 °C warming by 2100 and raises SCC by 12–14%.","key_machinery":"The central object is the generalized linear multi-reservoir carbon-cycle operator A (a mass-conserving, equilibrium-constrained box model) with a time-dependent land-biosphere equilibrium mass. The decisive element is the update $\\tilde{m}^L_{t+1} = \\tilde{m}^L_t - r\\, e^L_t$ with $r=1$, which shrinks the land reservoir's carbon holding capacity in lockstep with land-use emissions, forcing a larger share of emitted carbon to remain in the atmosphere. Calibration fits the operator's fluxes and equilibrium masses to the multi-model mean of the 100 GtC pulse-decay benchmark, with penalty terms for dynamic timescales, equilibrium masses, and ocean-to-land uptake ratios; a second layer rescales eigenvalues to emulate fast and slow extremes. For spatial downscaling, the workhorse is the linear pattern-scaling relation $\\Delta T^z = \\Delta T^{AT}\\, \\beta^z$ combined with an observational climatology to obtain absolute regional temperatures.","core_discovery":"Representing the land biosphere as a dynamic reservoir whose equilibrium carbon mass falls in proportion to land-use emissions (the 4PR-X model) materially alters the climate-economy outcome relative to the static three- and four-box emulators. Under business-as-usual and optimal mitigation, the 4PR-X emulator projects atmospheric carbon about 6% higher (~80 GtC) and global-mean temperature about 0.2 °C higher by 2100, raising the optimal social cost of carbon by 11.9–13.9% in 2020–2050 (up to 17 USD per tCO2). The fourth reservoir alone does nothing: the static 4PR model reproduces 3SR behavior almost exactly. The result is driven by an explicit mechanism: each ton of deforestation carbon emits CO2 and simultaneously removes one ton of permanent land storage capacity, so more carbon remains airborne; the paper explicitly flags that the one-to-one relationship (r=1) is an assumption.","pith_inferences":["If the r=1 assumption is relaxed to a value calibrated against observed biomass-loss data, the SCC uplift would likely shrink but remain positive for any r>0; the paper's headline numbers should be read as an upper-bound estimate of the deforestation feedback.","The same shrinking-sink mechanism could be ported to other compact climate models used for policy (impulse-response or reduced-complexity emulators), where land-use emissions are currently treated purely as an atmospheric source rather than as a reduction in future uptake capacity.","A testable extension is to couple the dynamic land reservoir with a simple carbon-cycle non-linearity such as saturating CO2 fertilization; under high emissions that non-linearity would reinforce the 4PR-X effect, making the 0.2 °C a lower bound at high concentration.","The pattern-scaling uncertainty decomposition suggests that spatial integrated assessment models should treat the warming pattern and the baseline climatology as separate, hedgeable uncertainty sources rather than pooling them into one damage-function risk."],"forward_implications":["DICE-type integrated assessment models that omit deforestation-driven loss of land carbon storage underestimate future atmospheric carbon, warming, and the optimal carbon price needed to offset them.","Policy evaluations of carbon capture and storage are overly optimistic unless land-use change is controlled: the 4PR-X model shows that CCS alone leaves temperatures about 7% higher in 2100 if deforestation continues.","Adding a static land-biosphere box to a three-box carbon cycle changes atmospheric carbon, temperature, and SCC by negligible amounts; the improvement comes only when the land reservoir's capacity responds to land-use emissions.","Pattern-scaling choices carry real economic weight: anchoring regional warming to different present-day climatologies can shift 2100 regional mean temperatures by up to roughly 3 °C, which can flip a region from a relative winner to a relative loser in a hump-shaped damage function.","Emulator calibration to present-day rather than pre-industrial conditions changes the trajectory but preserves the qualitative ordering, with present-day initialization yielding higher accumulation and higher SCC."],"supporting_citations":[{"why":"Supplies the DICE-2016 model and the three-box carbon-cycle structure that serves as the validation anchor and baseline economy.","marker":"Nordhaus (2017)"},{"why":"Provides the 100 GtC multi-model pulse-decay trajectories used as the calibration target for the carbon-cycle emulators.","marker":"Joos et al. (2013)"},{"why":"Supplies the calibration-validation protocol and the CDICE emulator against which the 3SR benchmark is checked and extended.","marker":"Folini et al. (2024)"},{"why":"Provides the pre-industrial equilibrium carbon masses for atmosphere, ocean, and land biosphere used in the equilibrium-mass penalty.","marker":"Ciais et al. (2014)"},{"why":"Provides the two-layer energy-balance model and parameter values used to convert CO2 concentrations to temperature.","marker":"Geoffroy et al. (2013a)"},{"why":"Supplies the CMIP5 pattern library used for pattern scaling and the associated inter-model spread in regional warming.","marker":"Lynch et al. (2017)"},{"why":"Supplies the hump-shaped regional damage function used to translate local absolute temperatures into productivity losses.","marker":"Krusell and Smith (2022)"}],"fun_headline_variants":["Deforestation carbon-sink loss lifts optimal CO2 price up to $17","Dynamic land sink adds 0.2°C warming and raises SCC by 14%","Fourth box does nothing; land-use change drives carbon price up","Land-use emulator: 0.2°C extra warming, 14% higher SCC by 2100","One-to-one land sink loss: deforestation raises SCC to $17/tCO2"],"cache_read_input_tokens":43392,"weakest_assumption_plain":"The whole policy effect rests on the assumption that each ton of carbon emitted through land-use change permanently removes one ton of carbon storage capacity from the land biosphere in the emulator; if the true loss is smaller or delayed, the extra warming and higher carbon price shrink accordingly.","fun_headline_variants_meta":{"raw":{"variants":["Deforestation carbon-sink loss lifts optimal CO2 price up to $17","Dynamic land sink adds 0.2°C warming and raises SCC by 14%","Fourth box does nothing; land-use change drives carbon price up","Land-use emulator: 0.2°C extra warming, 14% higher SCC by 2100","One-to-one land sink loss: deforestation raises SCC to $17/tCO2"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000928,"raw_usage":{"total_tokens":3977,"prompt_tokens":952,"completion_tokens":3025,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":568,"completion_tokens_details":{"reasoning_tokens":2924}},"tokens_in":568,"tokens_out":3025,"duration_ms":21247,"temperature":1.0,"reasoning_tokens":2924,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:20:03.474945+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Estimate r directly from data: compare cumulative land-use emissions against observed losses of global vegetation and soil carbon stocks; an r of, say, 0.3 instead of 1 would cut the 0.2 °C and 17 USD effects proportionally. Equivalently, drive the 4PR-X emulator with historical land-use emissions from 1850 onward and check whether the simulated atmospheric CO2 trajectory remains within the observed ice-core and Mauna Loa record; a systematic divergence would falsify the one-for-one sink-loss mechanism.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the DICE-2016 model and the three-box carbon-cycle structure that serves as the validation anchor and baseline economy."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the 100 GtC multi-model pulse-decay trajectories used as the calibration target for the carbon-cycle emulators."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the calibration-validation protocol and the CDICE emulator against which the 3SR benchmark is checked and extended."},{"cited_title":"Sabine, G","cited_arxiv_id":null,"evidence_quote":"Provides the pre-industrial equilibrium carbon masses for atmosphere, ocean, and land biosphere used in the equilibrium-mass penalty."},{"cited_title":"Smith, Anthony A","cited_arxiv_id":null,"evidence_quote":"Supplies the hump-shaped regional damage function used to translate local absolute temperatures into productivity losses."}],"review_version":1}