Model selection and evaluation system
The authors of the present study reviewed past research on CCUS source-sink matching models in the past 10 years with the keywords “CCS,” “CCUS,” and “source-sink matching” using the SCIE Web of Science Database and identified 16 types of models. Each model was renamed in the common format “MX_NAME” according to the time sequence of its publication (i.e., M1-M16) and its original name. When a model was not assigned an original name in the published literature, its name can be determined by combining the associated source-sink matching approach, network characteristics, or the author’s name with the addition of the study area. Thus, 16 models were renamed as follows: M1_SimCCS ; M2_CCSPD [19, 20]; M3_SimCCS_TIME ; M4_InfraCCS_EUR ; M5_ChinaCCS_DSS [23, 24]; M6_PNS ; M7_SCN_US [12, 13]; M8_MARKAL_IND ; M9_MILP_EUR (d’Amore et al., [9, 10]); M10_MINLP_KOR ; M11_MILP_ESP ; M12_ITEAM_CHN [14, 29]; M13_SSM_WANG_CHN ; M14_SSM_FAN_CHN [11, 15]; M15_C3IAM/GCOP (Wei et al., ); and M16_ChinaCCUS_DSS_2.0  (Please refer to Supplementary Table 1 and following List of Abbreviations for the detailed information of the 16 types of CCUS source-sink matching models).
Table 1 shows the evaluation system of the CCUS source-sink matching models, which could be used to comprehensively evaluate one specific model based on the insight of the comprehensive conceptual framework. Moreover, detailed evaluation criteria were proposed in this study, i.e., various levels of capabilities of each key attribute could be expressed with a score of 0–10. This facilitates the integration of the respective evaluation results of the six different attributes into one comprehensive conceptual framework.
Analysis of model evaluation results
The evaluation results of the six key attributes from 16 CCUS source-sink matching models are shown in Fig. 2, with associated analyses from the following three perspectives: comprehensive capability, data processing, and core matching approach.
The previous models show evolutionary characteristics from simple features to the overall enhancement of comprehensive capabilities. Given the trend in the total scores, it was found that the average score of the six models (M1-M6) was 22.8 ± 9.1 and increased from 2008 to 2014. The average score of the ten models (M7-M16) was 32.7 ± 3.7 and increased from 2015 to 2021, indicating that not only the average score has increased by nearly ten points, but the standard deviation has been significantly reduced. CCUS technology has attracted an increasing amount of attention from researchers in recent years. New models developed based on existing models generally have stronger comprehensive capabilities, contributing to the increase in the total score.
In addition, the models have gradually developed from early simple features to current multi-functional integration. Some early classical models, such as SimCCS, which started from the original design concept of the source-sink matching model that committed to the design and optimization of pipeline networks, have made a significant contribution to exploring a prominent issue of CCUS technology – how to make a linkage between different sources and sinks. It should be noted that, limited by the recognition of the role of CCUS technology in the overall mitigation technology system at that time, early models generally lacked consideration of some relevant attributes, especially the mitigation targets, utilization, and integration (synergy). People’s comprehensive recognition of CCUS technology has deepened afterwards, driven by a series of increasingly clear mitigation targets and several CCUS demonstration projects. Model innovation toward a multi-attribute framework of CCUS source-sink matching can demonstrate diversified development for promoting investment decisions, spreading influence to different industries, and driving the exploration of emerging commercial modes.
Reliable, precise, and broad input items continuously improve model data processing capabilities. The emergence of “big data” related high resolution and great collection capability has accelerated various types of well-designed and freely-accessible information, such as databases covering a range of carbon sources, carbon storages, cost estimates, and geographic information, resulting in a corresponding increase in the attribute scores of carbon sources and carbon sinks. For example, the currently developed models generally have a plant-level list of carbon sources for an entire industry at a national scale and have refined the coarse-mesh-based basins into high-resolution grid scales, greatly improving model capability to estimate the mitigation potential accurately. In addition, the accuracy and precision of cost estimates throughout the whole CCUS process have been greatly improved, highlighting the disaggregation of fixed costs and variable costs in the capture, transportation, and storage processes. For example, fixed costs can cover the costs of installing capture equipment (including compression), land purchase, pipeline construction, site selection, drilling, and recompression equipment, whereas variable costs can cover the costs of CO2 capture, transportation, network maintenance, injection, and storage.
The core matching approach and logical architecture become more sophisticated and complete. The MILP is the core optimization and solving method adopted by most models, whereas the pinch analysis or graphic processing methods are adopted by a few of the models (Supplementary Table 1). The pinch analysis or graphic processing methods can be introduced to address the CCUS planning problem. The graphical tool known as the CCUS planning composite curve could provide valuable insights and performance targets to facilitate the CCUS retrofit planning stage. For example, a minimum viable duration of connectivity could be specified for CCUS to be worthwhile. However, geographical distances and pipeline costs between various sources and sinks are neglected. All sources and sinks are assumed to be in close geographical proximity due to inherent simplifications and lower expandability of pinch approaches. The MILP method generally blends a direction-setting objective function and several constraints that the model needs to follow, thereby simplifying the CCUS source-sink matching process into a typical linear programming problem. Early models tend to take the minimum cost as the final optimization goal, while multiple optimization goals that consider the trade-off between mitigation potential and total cost have been included in recent models. In addition, there are increased constraint terms incorporated in MILP over time, reflecting the continuous improvement of model capacity to better describe the CO2 flow process in line with CCUS technology development. For example, early models seldom use injection capabilities as a key constraint to explore the impact of the actual injection rate, and rarely consider the role of economies of scale in reducing the total cost by introducing a pipeline thickness. However, these important constraints have now been incorporated into most models to different degrees.
The evolutionary characteristics of different attributes
In the absence of a clear mitigation goal of carbon neutrality, the mitigation contribution of CCUS provided by IAMs in the IPCC or IEA assessment reports could generally be regarded as a top-down target for overall mitigation potential. However, this approach has an over-simplified assumption about the underlying technology, which can easily cause CCUS development to deviate from real situations, further leading to significant uncertainties about its contribution to carbon neutrality. Other approaches with a bottom-up account of mitigation potential can make relatively reasonable inferences based on existing technology and industrial conditions but fail to accurately carry out future potential assessments because of the incomplete information of carbon sources or carbon storages. Some models developed in recent years can better integrate with global or national mitigation targets. For example, the M15_C3IAM/GCOP model  proposed a global source-sink matching layout from the perspective of limiting the global temperature rise by 2 °C, targeting the final optimization goal that achieves the 92 Gt mitigation contribution of CCUS at a minimal cost. The M16_ChinaCCUS_DSS_2.0 model  assessed the effect of CCUS in the overall realization of the optimal route for China’s carbon neutrality by connecting the CCUS source-sink matching model with the China TIMES model. This coupling could also provide solutions for studying the impact of technological competition on the layout of CCUS at a national scale.
The dynamic evolution of mitigation targets, particularly for the feasibility and technological competitiveness of CCUS retrofit, is now being valued by more models. In addition, a dynamic mitigation target represents the process of introducing CCUS infrastructure construction that is consistent with the changing mitigation targets and capital investments over time. For example, the M3_SimCCS_TIME model  focused on multiple periods of matching processes by introducing a time parameter that could index all model variables in the MILP formula. In this way, the amount of CO2 that was captured, transported, or stored in different periods varied, thus creating a powerful capability for a dynamic design of CCUS infrastructure over time. The dynamic mitigation targets have also been included in models such as M4_InfraCCS_EUR , M5_ChinaCCS_DSS [23, 24], and M10_MINLP_KOR  in different ways. Overall, more models are attempting to break free from the previous static models associated with a one-time investment of all CCUS infrastructures, as well as a time-invariant way of CO2 management by setting mitigation targets from a dynamic perspective. This is important for optimizing investment decisions with dynamic economic parameters (such as for discount rate and carbon price).
Carbon sources and carbon sinks
The temporal and spatial scale of carbon sources covered by CCUS source-sink matching models has widened in recent years. For example, early models were intended for the traditional thermal power industry as the carbon sources, whereas new models have taken a broader adoption of plant-level emission sources of steel, cement, and chemical industries. Early models usually selected basins or reservoirs as the candidate storage sites with a relatively coarse spatial resolution, insufficient considerations of injection capacity constraints, and the exclusion of offshore sequestration, resulting in a non-equivalent matching between carbon sources and sinks on relevant spatial scales . In general, models developed in recent years are dedicated to enriching databases, providing a considerable amount of information for carbon sources and sinks. For example, the M7_SCN_US model [12, 13] designed a cost-effective CCUS supply chain network in the United States, fully considering 3317 industry-wide sources and 2489 sinks (high-resolution grid for saline aquifers and oil/gas fields). The M15_C3IAM/GCOP model  collected 4220 industry-wide carbon clusters and 794 sinks from 85 countries and regions and was committed to the global 92 Gt emission reduction goal contributed by CCUS. The M14_SSM_FAN_CHN model [11, 15] developed an optimization model that does not consider cost constraints to explore the emission reduction potential of 607 existing coal-fired power plants in China, using a total of 3853 high-resolution grids for saline aquifers and oil fields as the carbon sinks. The marine geological survey database has rarely been fully utilized to identify potential carbon sinks, although some exploration has been conducted by the M9_MILP_EUR [9, 10] and M10_MINLP_KOR  models.
Network optimization is an additional upgrade process for the preliminary results provided by CCUS source-sink matching models, helps make the transportation network more efficient and practical. Innovations in network optimization are considered more difficult than other attributes because of the complex network structures. Thus, a series of assumptions are needed to simplify the optimization process and ensure the model is tractable, such as a straight-line connection between a single source and a single sink and using a given distance constraint. However, too simple assumptions may cause the networks to be unrealistic and inefficient. It is worth mentioning that the M1_SimCCS model  was still one of the few models that could satisfactorily describe topographical conditions on the network layout. Specifically, the SimCCS model has established a rasterized-weighted cost surface to estimate the real cost of pipeline construction crossing through each grid by using a set of topographical factors including terrain, topography, land use, intersections, land ownership, the right-of-way of pipelines, and population density. Furthermore, the SimCCS model could determine the location, length, and size of different trunk pipelines and branch pipelines based on a vector-based candidate network combined with a pre-optimization algorithm. Similarly, the M10_MINLP_KOR model  followed the concept of cost surfaces to obtain the minimum cost path and applied several subdivided penalty factors to show dramatic changes in geographic conditions in a more generic way using ArcGIS software. Generally, the cost surface processing method demonstrates a strong ability to characterize the impact of topographic conditions on the costs and the optimization of CCUS infrastructures. However, other issues, such as an understanding of the aggregation benefits from having trunk pipelines with different thicknesses, and the definition and identification of CCUS clusters or hubs, still lack effective solutions.
To offset the cost associated with CO2 capture, transportation, and storage, there is growing interest in finding profitable end-use opportunities for the captured CO2. In this perspective, some studies focus on the potential contribution of carbon capture and utilization (CCU), instead of trapping CO2 underground as a terminal destination [13, 32, 33]. However, some studies have also found that CCU could, at most, be seen as supplementing carbon capture and storage (CCS) to a small extent, and it is highly improbable to allow long-term sequestration owing to the limited scale and rate of utilization. Thus, even if the CO2 utilization is analysed separately in this review, it is still considered a commercially viable method under the entire CCUS chain towards CO2 sequestration. Apart from some EOR approaches, most existing models have insufficiently considered the CO2 utilization process. The associated chemical, biological, and mineralization utilization still cannot be identified in previous models. Some models can identify early development opportunities of CCUS hubs combined with intensive EOR utilization from the spatial distribution of source-sink matching. For example, the M16_ChinaCCUS_DSS_2.0 model  explored the differentiated layout of CCUS for China’s power system at low, medium, and high deployment levels, and concluded that the northern regions with abundant oil and coal reserves should push forward with the implementation of EOR and ECBM in practical applications. The M12_ITEAM_CHN model [14, 29] evaluated the synergetic benefits of providing large-scale mitigation potential and additional water supply through EWR utilization to cope with the huge water consumption of China’s coal chemical industry. It was found that EWR could be conducted at a relatively low cost and mitigate 269 Mt CO2 from the separation process, as well as generate 404 Mt groundwater for further desalination and utilization.
Recently, more consideration has been given to the integration and synergy capabilities of various models. In addition to the continuous enhancement of layout optimization capabilities related to the evolution of transportation networks, some models can also comprehensively consider investment decisions or social risks. However, the complementation and synergy between CCUS (fossil-based energy), hydrogen energy, and other renewable energy sources within an integrated carbon management framework are less studied. The M9_MILP_EUR model [9, 10] proposed a spatially explicit MILP approach for the economic optimization of the European CCUS supply chain and explicitly incorporated a social risk assessment module in the modelling framework, as well as risk mitigation measures in the pipeline design. The M11_MILP_ESP model  also adopted the MILP method as the matching framework to explore the conditions and infrastructures for the large-scale deployment of CCUS in Spain by 2040. More importantly, this model investigated the CO2 emission threshold that drives necessary investment decisions, combining the overall layout of multiple regional networks with spatial characteristics of the Spanish peninsula. The M12_ITEAM_CHN model [17, 29] investigated the relationship between the standardized cost and CO2 emission reductions by integrating CCUS source-sink matching and investment decisions, finding that 22–58% of coal-fired power plants are cost-competitive with onshore wind power.