Abstract Random forest (RF) models for PM2.5 prediction inform exposure assessment and air-quality management. These models are typically evaluated using spatial or random cross-validation (CV), which often yields high R² values. However, how the choice of CV strategy interacts with hyperparameter tuning (e.g., max_features) has received little attention. Using daily PM2.5 data from 157 monitoring stations in northern China, we show that Spatial-CV and Temporal-CV produce substantially different performance (R² of 0.85 vs. 0.37) and, more importantly, lead to opposing optimal hyperparameter settings. The clearest case is max_features: Spatial-CV performs best when almost all features are considered at each split, while Temporal-CV performs best with only a few (20 vs. 3 of the 22 features). The performance gap and the hyperparameter reversal between Spatial-CV and Temporal-CV both reflect spatial interpolation rather than learned physical relationships. A diagnostic Date-ID model built from date and coordinate identifiers alone (no physical predictors) reproduced both patterns and reached the highest Spatial-CV R² (0.95); conversely, reducing station density from 157 to 20 stations progressively weakened both. For the current literature, the implication is direct: a high RF model score (e.g., R²) cannot certify what a model has learned or how far it generalizes. A predictor's assessed value and the model's overall interpretation depend on the data structure, evaluation, and hyperparameters, and can reverse when these change. Satellite aerosol optical depth (AOD), for instance, appears unnecessary under Spatial-CV but helps under Temporal-CV, and featureimportance rankings shift between the optimal hyperparameter configurations of Spatial-and Temporal-CV. This indicates a systematic mismatch between what a CV metric measures andwhat is claimed from it in the literature. We recommend tuning hyperparameters separately for each CV strategy, reporting performance across station densities, and building models using date-coordinate features as a baseline.
Abstract Extreme cold events (ECEs) impose societal risks, yet the local process linking large-scale teleconnections to regional ECEs is unclear. Here we systematically evaluate local atmospheric factors to identify the factor most strongly associated with the linkage between the boreal winter circum-hemisphere teleconnection and regional ECEs. Using a five-part statistical framework, we compare 1000–500 hPa thickness (ΔZ) with 13 other factors. ΔZ exhibits the strongest and most robust statistical association with ECEs among the factors examined: removing ΔZ produces the largest change in the ECE-related cold-tail distribution, P(ECEs | extreme ΔZ) is the highest among all factors, and the ΔZ–ECE relationship remains robust across multiple statistical analyses. The perturbation hypsometric law and complementary dynamical diagnostics provide a physical interpretation of this statistical relationship. Together, these results suggest that atmospheric thickness provides a physically interpretable local bridge from large-scale teleconnection signals to ECEs.
Abstract India is the world’s second-largest rice producer, accounting for $\sim$ 22\% of global production from cultivating over 40 million hectares and exporting more than 15 million tons annually. This makes the country a key player in global food security, but a significant contributor to the global rice methane (CH$_4$) emissions. Current estimates of Indian agricultural CH$_4$ emissions are highly uncertain, exceeding 50\%, and lack spatially explicit details across the region, which poses challenges in implementing sustainable mitigation strategies. This study developed a satellite-based method that resolves seasonal and subnational variability in rice CH$_4$ emissions across India at 10 m resolution. By integrating Sentinel-1 SAR and Sentinel-2 multispectral observations with a semi-empirical, process-based model, we developed the first observationally-driven spatially explicit, multi-season dataset of rice CH$_4$ emissions for 20118--2025 over India. The methodological evaluation achieved 90.12\% pixel-level validation accuracy and demonstrated strong agreement with agricultural census statistics ($R^2$ = 0.896) for rice extent mapping. By utilizing unprecedented details on rice extent, we estimated national annual mean emissions to be 9.73 $\pm$ 2.03 Tg CH$_4$ yr$^{-1}$, which is larger than previously reported. We find a dominant contribution from the Kharif monsoon season (74.7\% of annual emissions), followed by the Summer season (17.5\%) and the Rabi winter season (7.8\%), highlighting untapped mitigation opportunities in rice fields through targeted water-management practices during Kharif rice cultivation. A 18-member model ensemble configuration captures seasonal and spatial variability in emissions driven by temperature-dependent methanogenesis, heterotrophic respiration, and precipitation-modulated flooding dynamics, and enables the quantification of structural uncertainty arising from model parameters and environmental sensitivities. Thus, the study's findings address key structural limitations in existing inventory approaches and provide robust, highly-resolved rice-based methane emission estimates for atmospheric inverse modeling applications, subnational mitigation targeting, and the evaluation of water management strategies. Overall, this study lays the foundation framework for assessing agricultural CH$_4$ emissions and their mitigation efficiency in India, with the potential for adaptation in other regions.
Abstract Drought is a hazard that impacts millions of people globally each year and that can have significant impacts on societies, economies, and ecosystems. Limited research has been focused on drought vulnerability in high-income countries partly because many are located in traditionally water-rich regions. However, recent droughts in 2022 and 2023 demonstrate that these countries can also be adversely impacted. This paper reviews current research on drought vulnerability in high-income countries with the aim of assessing the state-of-the-art and identifying research gaps. A systematic literature search identified 144 records, of which 71 studies were retained for full-text review. Of these, 33 had a global scope, 31 focused on Europe or European subregions, and only 6 focused on North America. Oceania and Asia were only represented in global studies. The literature was strongly sectoral, with agriculture being the most studied sector (17 studies), followed by trees/forestry (12 studies), while only a small number of studies assessed vulnerability across multiple sectors. A main finding of the review is that there are substantial differences in drought vulnerability within countries, which are not necessarily related to physical water scarcity, but to differences social, economic, and local and/or regional governance and policies. The findings also highlight the fact that there is no standardized concept for drought vulnerability, making assessments challenging to compare. In addition, the review highlights the lack of cross-sectoral drought vulnerability studies and a limited understanding of cascading impacts and gives recommendations for further research to aid a more comprehensive understanding of the impacts of drought.
Abstract Soil water availability during the dry-to-wet season transition drives ecosystem productivity across the semi-arid Sahel. During this period, soil moisture reflects the balance between water supplied by precipitation and atmospheric demand, but how these processes jointly influence the timing of soil moisture accumulation remains unclear. Here, we analyze trends in soil moisture and precipitation during the dry-to-wet season transition from 2003 to 2024 across the semi-arid zone between the Sahara and the humid tropics. We use daily maximum temperature as an index of evaporative demand in a diagnostic soil moisture model to evaluate how precipitation inputs and temperature-driven dry-down each influence soil moisture dynamics. We find that soil moisture onset is systematically delayed relative to precipitation onset across much of the Sahel, and that trends in soil moisture onset and precipitation onset diverge in an east–west pattern. Including daily maximum temperature alongside precipitation in the diagnostic model substantially improves its ability to reproduce onset timing and helps explain the observed divergence between soil moisture and precipitation onset. These results indicate that increased evaporative demand can offset increases in precipitation during the dry-to-wet season transition and delay the emergence of plant-available water. These findings highlight the importance of atmospheric demand alongside precipitation inputs when assessing shifts in soil water availability across the Sahel, with potential implications for rainfed agricultural systems.
Abstract To meet the Kunming-Montreal Global Biodiversity Framework, spatial planning must upscale area-based conservation, while ensuring equitable governance and local community integration. Historically, a deep divide has persisted between data-driven, top-down systematic conservation planning, often performed at large spatial scales, and participatory, bottom-up local initiatives. Here, we propose a novel framework that bridges this gap by harnessing the power of human and artificial intelligence. By deploying a technique called Reinforcement Learning with Human Feedback in our software CAPTAIN (Conservation Area Prioritization through Artificial INtelligence), our approach can process vast biophysical and socioeconomic data to create fine-tuned conservation strategies that are both data-driven and sensitive to local realities. The methodology outlined here provides a promising avenue for scalable, effective, and equitable biodiversity conservation.
Abstract Municipal wastewater treatment plants (WWTPs) are important components of urban carbon cycling and have been proposed as potential platforms for ocean alkalinity enhancement (OAE). However, carbonate-system transformations across full-scale treatment processes remain poorly constrained. Here, we present a process-resolved snapshot of total alkalinity (TA), dissolved inorganic carbon (DIC), pH, pCO₂, dissolved CH₄, and dissolved N₂O across five sequential treatment units at a full-scale coastal WWTP in Taiwan. From the influent channel to the chlorination unit, TA and DIC decreased by 94.5% and 84.0%, respectively, accompanied by acidification and persistent CO₂ supersaturation. These patterns were consistent with alkalinity consumption during nitrification and DIC transformation associated with aeration and other internal treatment processes. Dissolved CH₄ declined markedly following aeration, whereas N₂O reached its maximum concentration in the aeration basin and remained elevated in the final effluent. CO₂ exchange estimates obtained using conventional wind-based and limited-fetch parameterizations differed substantially from apparent DIC balances between consecutive treatment stages. This divergence highlights the strong methodological uncertainty associated with estimating CO₂ exchange in confined and mechanically aerated treatment units. Accordingly, the results are interpreted primarily in terms of observed concentration changes and carbonate-system transformations rather than as definitive emission estimates. The marked chemical differences among treatment stages provide empirical constraints for evaluating where and how alkalinity addition might interact with wastewater treatment processes. Repeated observations, direct gas-flux measurements, and alkalinity mass balances will be required to determine the net carbon-removal potential of wastewater-based OAE.
Abstract Marine low clouds strongly affect Earth’s radiation budget, yet climate models continue to underestimate shallow cumulus and misrepresent low-cloud regimes in cold-advection environments. Here, we combine five years of radar observations from the Atmospheric Radiation Measurement Eastern North Atlantic site with 36-h backward trajectories to examine how upstream air-mass pathways shape stratocumulus-related cloud states and vigorous shallow cumulus. Our results show that cold-advection strength alone cannot fully explain the cloud state observed at trajectory arrival; pathway-dependent thermodynamic and dynamical adjustment along the air-mass pathway must also be considered. The Sc-related states exhibit near-surface warming that outpaces moistening, whereas vigorous shallow Cu maintains a more coordinated near-surface warming and moistening pathway. This coordinated adjustment favors a relatively lower lifting condensation level and, together with stronger surface buoyancy flux, provides a favorable environment for active shallow convection. These findings highlight the importance of pathway-dependent warming–moistening adjustment for understanding differences in low-cloud development under cold-advection conditions. They also provide an observational reference for the near-surface thermodynamic conditions and surface forcing accompanying low-cloud evolution in climate-model simulations under cold-advection conditions.
Bottom-up coal mine methane (CMM) inventories rely on static or empirically derived emission factors (EFs), and therefore mine-level emissions are poorly constrained, limiting the use of these inventories for implementing detailed mitigation strategies. Here, we compiled 1418 satellite-detected methane plumes (2019 – 2025) and attributed them to 159 active underground coal mines in Shanxi province, China. We further derived observed mine-level EFs, calculated as mine-level emission rates divided by production data. We then compared these observed EFs with those from the State Administration of Coal Mine Safety (SACMS) and Global Coal Mine Tracker (GCMT), and developed a production-capacity-stratified bootstrap framework to upscale emissions from high-gas and outburst coal mines. Observed EFs were highly heterogeneous, right-skewed, temporally variable and negatively correlated with production capacity. Inventory comparisons revealed distinct biases: SACMS reproduced the overall EF magnitude but systematically underestimated EFs for small-capacity coal mines (production capacity <1.2 Mt yr ^−1 ), whereas GCMT overestimated EFs for medium- (1.2⩽ production capacity <3.0 Mt yr ^−1 ) and large-capacity coal mines (production capacity ⩾3.0 Mt yr ^−1 ). Using the production-capacity-stratified bootstrap upscaling framework, we estimated 2023 CMM emissions from high-gas and outburst coal mines in Shanxi to be 7.0 [5.6 – 8.9] Mt yr ^−1 . Total provincial CMM emissions were estimated at 11.2 [9.3 – 13.6] Mt yr ^−1 . These findings show that satellite-observed plumes can constrain mine-level EFs, reveal inventory biases, and support observation-based provincial CMM estimation.
Perceived tradeoffs between ecosystem services (ES) delivered by constructed ponds and wetlands in agricultural landscapes may limit their widespread uptake for environmental management. These nature-based solutions (NBS) can mitigate downstream eutrophying effects of agricultural nutrient runoff and contribute to carbon (C) storage. However, they can also be significant greenhouse gas (GHG) sources. Here, we report on water chemistry, dissolved GHG concentrations, total methane (CH _4 ) emissions, and sediment properties measured over three years at 40 Swedish free-surface constructed agricultural wetlands (ponds). Large temporal variations in inlet water chemistry highlighted the influence of seasonality and land management. Inlet phosphorus (P) concentrations were positively correlated with water column dissolved CH _4 and sediment P concentrations; a clear tradeoff in nutrient retention vs. climate impact. A more nuanced pattern occurred for nitrogen (N) where inlet concentrations correlated positively with dissolved nitrous oxide (N _2 O) concentrations. However, these wetlands mitigated downstream dissolved N levels as suggested by lower outflow concentrations. All waterbodies were supersaturated with CO _2 . Higher CO _2 and N _2 O concentrations occurred during fall/winter than summer. The opposite pattern occurred for CH _4 fluxes and dissolved CH _4 although relative contributions from ebullition were greater during fall/winter. Sediment C concentrations were unrelated to any measured parameters, suggesting it would be difficult to purposefully design wetlands for C sequestration. Although there are tradeoffs between mitigating downstream eutrophication and climate impacts, this should not preclude the use of constructed free surface wetlands as multifunctional NBS for ES delivery in agricultural landscapes.
Tropical cyclones (TCs) cause substantial disaster losses worldwide. Forecast skill for TC track and intensity has been improved by enhanced observations, high-resolution numerical models, advanced data assimilation methods, and applications of machine-learning methods. Yet these improvements have not consistently translated into reduced losses, in part because disaster outcomes depend on many other elements, including communications between weather agency and the public. Therefore, it is not straightforward to observe the damage reduction by improved TC forecast, and the impact of the improvement of TC track and intensity forecast on damage has not been comprehensively quantified using real-world data. In this study, we examine 32 TCs that made landfall in Japan between 2006 and 2023 and quantify how errors in Joint Typhoon Warning Center operational TC track and intensity forecasts relate to flood-induced damage. To our knowledge, this is the first nationwide assessment for Japan linking operational TC forecast accuracy to observed TC-induced damages. Our multi-linear regression analysis reveals that along-track forecast error—distance between forecast and actual positions along the direction of travel—is positively associated with building and household damages ( p $ \lt $ 0.05), implying greater residential and structural impacts when forecasted TC positions deviate farther along the track. Conversely, landfall timing and intensity errors show no statistically significant association with flood damage. Despite the unclear causal relationships, these findings imply that further reductions in track error, particularly its along-track component, may contribute to mitigating TC-related flood losses.
Abstract Nature-based Solutions (NbS) are increasingly promoted for climate adaptation to compound hazards, but systematic evidence of their effectiveness is still scarce. We address this gap with a systematic review of 38 studies. Each study was sorted by what its measurement design can support, not by what its authors claim. This produced four groups. Direct studies attribute a measured hazard reduction to the intervention, proxy studies infer protection from ecosystem condition, prioritization studies map suitable locations without estimating performance, and qualitative studies describe effectiveness without measuring it. We detect trends in geographic distribution, hazard combinations, intervention types, methodological approaches, and strength of evidence. Six studies directly measured attributable hazard reduction, and the remaining 32 studies fell into the proxy, prioritization, or qualitative classes, none of which can separate an ecosystem that buffers a hazard from one that receives it. The evidence concentrates on coastal flooding, with urban heat second. The most common hazard combination, compound drought and heat, yielded no direct evidence of hazard reduction, and only one of the five studies addressing cascading hazards did so. Research priorities include direct assessment of vegetation-based NbS under compound drought and heat, attributable measurements for cascading hazards, and long-term monitoring of hazard outcomes. As NbS are scaled, confidence in them should be scaled with the strength of the evidence base.
Sub-Saharan African (SSA) cities have high air and noise pollution levels, yet limited information exists on children's joint exposures at home and school locations - the two most important environments where children spend much of their time. This study characterised schoolchildren's exposure to multiple air pollutants and environmental noise and sound sources within the Accra School Health and Environment Study (ASHES). ASHES involved 1034 children aged 8-12 years from 90 public (73%) and private schools in Accra, Ghana, a major SSA city. Annual mean concentrations of fine particulate matter (PM2.5), nitrogen dioxide (NO2), and black carbon (BC), as well as environmental noise (L day, L night, L den) and sound sources, were derived from land use regression models and linked to geocoded school and home locations for 919 children with valid data. A time-location weighted total exposure estimate was calculated for the air and noise pollution metrics. We also examined children's responses to a noise annoyance survey in relation to their noise exposure levels at home. For air pollutants, median home exposures were 33.8 μg m-3 (PM2.5), 55.5 μg m-3 (NO2), and 5.2 × 10-5m-1 (BC), and corresponding median school exposures were 29.9 μg m-3, 53.5 μg m-3, and 4.9 × 10-5m-1. The time location-weighted total air pollutant exposures for most children surpassed the respective World Health Organization (WHO) guideline. Similarly, median environmental noise levels (in dBA) at home (L den = 66.3 and L night = 54.2) and at school (L day = 62.4) exceeded WHO thresholds for road-traffic noise and Ghana's noise standard for areas with educational facilities, respectively. Pollution levels at home and school were moderately correlated (r [PM2.5] = 0.56; r [NO2] = 0.33; r [L den]= 0.47). On average, children attending public schools or living in lower income areas had higher exposures to both air and noise pollution than their private school (fee-paying) or higher-income counterparts. In contrast, nature-based sounds were more common in higher income areas. Among children exposed to L den noise levels between 65 and 70 dBA, 35% reported being highly annoyed to road traffic noise. Schoolchildren in Accra experience air and noise pollution levels exceeding health-based guidelines and standards. Exposures are unequally distributed, highlighting environmental inequalities with implications for child health, development, and learning.
As science surrounding climate prediction has improved in recent decades, focus has shifted to making predictive information accessible to end users through decision support tools (DSTs). Involving and prioritizing stakeholders in the creation of these tools is widely considered best practice; however, successfully co-producing tools within the structure of grant-funded academic research projects is challenging. We present a case study of one such effort, the Dashboard for Agricultural Water use and Nutrient management (DAWN), a platform which provides DSTs derived from coupled climate-agriculture forecasts for the subseasonal-to-seasonal period. We explore the development of DAWN’s Climate Data Viewer, reviewing the challenges and opportunities involved in its creation and highlighting three key themes that emerged: the selection of relevant and useful outputs; the communication of uncertainty and model skill; and the balance between comprehensiveness and functional simplicity.
The Paris Agreement’s ratchet mechanism asks each successive nationally determined contribution to represent a progression beyond its predecessor, with Article 4.3 leaving the metric, baseline, and accounting conventions for assessing progression at the discretion of each Party. This perspective shows that this architectural latitude permits substantive regression of the absolute emissions constraint across consecutive NDCs, exercised through distinct mechanisms in major parties to the Agreement. India’s 2035 NDC raises each of its three quantitative targets above the 2030 NDC and satisfies the textual progression requirement. Under matched GDP-growth assumptions across 5%–8% CAGR, however, the marginal 2030–2035 period permits faster annual absolute emissions growth than the 2030 NDC’s 2020–2030 operative period, and India’s required pace of intensity decline drops to half the rate the previous NDC required. Two related forms of the same latitude appear in the 2035 NDCs of China and the European Union, exercised respectively through target-reference calibration and accounting-boundary choices. The architecture as currently assessed cannot distinguish these exercises from genuine progression. The Global Stocktake’s technical assessment phase will need to develop cross-vintage absolute-emissions diagnostics across Parties if the ratchet is to function as its design assumes.
Here we develop and analyze a new global crop-specific fertilizer application rates timeseries dataset for the period 1980–2022. We use three inputs—reported fertilizer use by crops, crop-specific harvested areas and total national-level fertilizer consumption—together with a simple three-step imputation process to fill in missing data, followed with extensive comparisons against independent datasets. Using this new dataset, we find that from circa 1980–2020, global consumption of N-fertilizers in crops (excluding pastures and forests) increased by about 68%, while P- and K-fertilizers rose by about 52% and 69%, respectively. More than ten-fold gaps in N-fertilizer applications however exists between regions with the highest and lowest application rates. Recent increases in N- and P-consumption was driven more by expansion of crop harvested areas than increasing application rates. In the case of potassium, both increases in crop area and fertilizer rates contributed to the rising use. Lately, N- and P-application rates have declined in East Asia and Western Europe while their use has increased in regions with under-utilization, pointing to more recent policy successes, and a general trend in global fertilizer use convergence.
The Indian Government has ambitious goals for the renewable energy (RE) transition in the country, especially through the expansion of wind and solar. New monitoring stations have come up in recent years to support this countrywide expansion. Despite all these, there still remains a gap in the observational network in terms of spatial distribution and long-term data availability when compared against global RE leaders. For example, in the highest wind-resource state/province in 2024, the maximum number of wind farms having at least one observing station within a 25 km radius in India is 11, whereas the corresponding numbers for China, the contiguous United States, and Germany are 84, 162, and 206, respectively. Additionally, only a handful of stations in India have more than 90% data record in the past decade from 2014–2024, a stark contrast to the above-mentioned countries. While some wind and solar observations over India are available freely, majority of them are not, thereby posing a hindrance to the advancement of RE research in the country, especially on understanding factors affecting grid integration, compound wind-solar droughts, energy mix and energy reserve, resilience of RE systems, long-term projections and planning. The present study provides a comparative assessment of India’s observational infrastructure for RE applications. Overall, India’s current observational datasets provide an important foundation but are not yet commensurate with the country’s rapidly expanding RE ambitions. Strengthening observation-network density, long-term data continuity, and data accessibility will be essential to support energy planning and services in India across a multitude of scales.
Human judgment drives reservoir water systems globally, but computational models struggle to capture it. Data-driven approaches replicate patterns without revealing decision logic, and optimization requires objectives that are hard to fully specify. Large language models (LLMs) may offer an alternative, as they replicate human decision patterns in economic and cognitive tasks and show emerging multi-step reasoning under uncertainty. We test this by embedding an LLM as an operator of Folsom Lake reservoir in California, tasked with monthly water supply allocation decisions over a 21 year simulation. Benchmarking against historical operations, a neural network regressor, and dynamic programming reveals that the LLM exhibits risk-averse hedging behavior without task-specific training. The LLM’s performance falls between naive release policies and formal optimization, spanning a range comparable to varying risk aversion in stochastic dynamic programming. Models that generate intermediate chain-of-thought traces consistently outperform those that output decisions directly, and sensitivity and ablation analyses partially validate the LLM’s self-reported rationale. These results demonstrate that LLM generative agents can approximate human-like water management decisions from contextual information alone, bridging quantitative optimization and qualitative judgment for decision support, policy exploration, and human–water system research.
Mediterranean Sea surface temperature (SST) has warmed consistently over the past century, with a clear intensification in recent decades. Using multiple observational SST products spanning more than 170 years, we jointly analyse the low-frequency SST, T ( t ), and its temporal derivative, d T /d t , to characterize the interplay between historical multidecadal modulation and long-term warming in Mediterranean SST evolution. During the satellite era (1982–2025), the basin-mean SST exhibits a linear warming trend of 0.34 ± 0.06 °C/decade and pronounced warm-season amplification. After removing the component linearly associated with global mean SST, an Atlantic Multidecadal Variability-like modulation remains evident in Mediterranean residual variability, whereas multidecadal reversals become less apparent in the total SST record. Change-point and phase-space diagnostics indicate an early twentieth-century shift toward higher low-frequency SSTs and a late twentieth-century shift toward sustained positive warming rates, placing recent decades in a warmer and faster-warming region of the ( T ( t ), d T /d t ) phase space. Overall, the observational record suggests that long-term warming is becoming increasingly dominant in shaping Mediterranean SST variability.
Concrete accounts for approximately 7% of global anthropogenic CO $ _2$ emissions, nearly two-thirds of which are hard-to-abate limestone calcination emissions that are partially offset through natural carbonation. This study presents a dynamic cradle-to-grave carbon footprint assessment of six scenarios for hollow concrete blocks, evaluating carbon capture, carbonation curing and mineral carbonation alongside their impact on natural CO $ _2$ uptake. Natural carbonation offsets $11\%$ – $37\%$ of calcination emissions, rising to $25\%$ – $55\%$ , depending on modelling assumptions, with carbonation curing. Static accounting overestimates this benefit by $4\%$ – $10\%$ . Compared with conventional concrete, carbon capture, utilization, and storage scenarios can reduce emissions by $52\%$ – $59\%$ , with carbon capture responsible for $32\%$ – $43\%$ and the elimination of steam curing for $16\%$ – $21\%$ . Subsequent carbonation curing and mineral carbonation components themselves yield only marginal emission savings of $1\%$ – $6\%$ , but reduce the reliance on geological CO $ _2$ storage by $36\%$ – $48\%$ relative to a capture-only scenario. These findings show that carbon capture and eliminating steam curing drive the emission reductions, while carbonation curing and mineral carbonation offer alternatives to CO $ _2$ storage.
