The subsurface geological data can be classified into several types at various spatial scales: 1) properties of reservoir-seal pairs, geographic sequence, and spatial distribution of sedimentary facies system and lithology with different physical and chemical properties, such as lithology, pressure, temperature, porosity, and permeability, entry pressure, compressibility, thermal conductivity, and other properties; 2) boundary conditions (open, semi-open, or closed systems), tectonic setting (active/inactive faults with/without vertical communication among different geological stratum), and sedimentary facies system (continuity at a regional scale); 3) geo-fluid properties (water salinity, viscosity, phase behavior, density, capillary pressure, solubility, empathy, and dynamic thermal properties) (Dewers et al., 2018). The characterization data of reservoir-seal pairs mainly include spatial distribution of physical and chemical properties, such as porosity, permeability, relative permeability, capillary pressure, geochemistry, minerals, in-situ pressure, temperature, lithology, salinity, rock compressibility, fracture pressure, and mechanical properties of rock. The boundary conditions mainly focus on the open or closed boundary, such as outcrops of aquifer formations, low-permeable facies, impermeable or permeable faults, and other potential leakage pathways. The regional geological data focuses on the sedimentary system, tectonic, and diageneses process at a basin- and sub-basin scales.

The various types and detailed levels of geological data should be assimilated into data collection and evaluation tools. The most common methods to incorporate massive data are geological models, geographic information system (GIS) data, image processing models, data stacks, and other geological models or software. In most scenarios, these geological data can be compiled into GIS systems, such as ArcGIS, MapGIS, MapInfo, QGIS, and reservoir modeling software used by the petroleum industry, such as CMG by Computer Modeling Group, Schlumberger’s Petrel software, landmark, Geostatistical Software Library, and GoCAD (Iea-Ghg, 2009; Jiao and Surdam, 2013; Li et al., 2015). Using reservoir modeling tools, various algorithms based on geological models can calculate storage efficiency coefficients and CO2 storage capacity.

Similar to deterministic and stochastic methods, two types of geological models, homogeneous and heterogeneous, are used in capacity evaluation. Homogeneous models can be generated using the average properties of reservoirs derived from the database. The storage coefficient factor in a homogeneous model can be calculated using the algorithms above. A heterogeneous model can be built considering the spatial distribution of lithology, structural settings, geochemical environment, and mineral composition. The properties can be derived from site characterization or dataset extrapolated from the well-known nearby site by theoretical reservoir engineering analysis. The resolution of the heterogeneous model depends on the resolution of site characterization and data extrapolation. With sufficient data and extrapolation tools, the results of the heterogeneous model can cover a more comprehensive uncertainty range than that of the homogeneous model. However, homogeneous models for large-scale evaluation are more plausible than heterogeneous ones because they are time-saving, efficient, and available in building models, and efficiently perform with limited data, especially in the stochastic analysis that requires large computation capacity.

These site-scale data usually contain well drilling and logging, 2D/3D seismic investigation, micro-gravity investigation, site operating data, geodetic survey, and other sources (Birkholzer and Tsang, 2008; Kim et al., 2014). However, available geological data in the petroleum industry are limited before detailed site characterization. The confidence of geological investigation is always defined by the number of investigation wells and density of seismic investigation in the geological volume of interest. More flexible methods such as the variable grid method can allow various data qualities while preserving the overall spatial trends and patterns (Bauer and Rose, 2015).

The surface data are mainly non-geological and can be grouped into several types: 1) geological data including surface geological characteristics, lithologies, metamorphic rock, igneous rock, stratigraphic contour line, earthquake records, first and secondary tectonic units, outcrops, and others; 2) geographic and geomorphic data including mountains, water system, landform, climate, precipitation, wind energy, solar power, geological hazards, and other features; 3) social and environmental data, including CO₂ emission sources, transportation system, natural preservation park, mining, natural resource, oil and gas field, well information, vegetables, cities, population, industry, economic density area, infrastructure, economic parameters, climate, evaporation, water system, natural preserves, and others. These surface data can be compiled using various data forms of point, polyline, polygon, raster, vector, data stacks, or other types such as ArcGIS, MAPGIS, Access, and mathematical tools developed by various computing languages. The algorithms related to non-geological factors can refer to existing algorithms that perform site suitability and risk analysis. These algorithms include multi-criteria analysis with empirical or statistical criteria, analytic hierarchy process, spatial analysis in GIS systems, numerical modeling, probability analysis, and others (Ellett et al., 2013; Wei et al., 2013). Using knowledge integration and data assimilation of the multiple types of site data at multiple scales and theories of sedimentary basin evolution can improve geology assessments (Popova et al., 2014; Dilmore et al., 2015).

Data resolution or accuracy is the smallest difference between adjacent positions/sites that can be recorded at a spatial dimension. The resolution is also the character difference between the interpreted value and the true value. Data uncertainties are the combined effect of site investigation and data interpretation tools; the uncertainty from data compilation tools handling massive data can be neglected. The resolution of site data depends highly on the coverage and details of site characterization tools. In order of decreasing resolution and increasing coverage, these tools inlcude petrophysical and chemistry properties experiments at pore and core scale, well drilling and logging, geophysical investigation (such as profile interpretation crossing multiple wells, 2D/3D seismic investigation, and micro-gravity investigation), and theory of sedimentary process and tectonic activities. Current site characterization is usually conducted by standard petroleum and underground mining technologies. The proximate resolution of site characterization generally ranges from 0.1 m to several hundred meters depending on the characterization approaches and spatial correlation from investigated points with high resolution by well logging and core analysis to low-resolution points by seismic investigation and profile interpreted by data assimilation. The assimilation of various data sources can provide reliable geological models of storage sites (Chen et al., 2020a). However, most capacity assessments are conducted before the stages of detailed site characterization and the stage of contingent resource assessment; the spatial resolution of geological data is much lower than that of seismic investigations. Therefore, the ideal geological model should be built using sufficient data obtained by various site characterization tools and data interpretation tools. These tools contain core characterization, well logging, 2D/3D seismic survey, electromagnetic investigation, micro-gravity investigation, theoretical basin modeling technologies (sedimentary, tectonic, and diagenesis theory), etc.

Nevertheless, data scarcity and imbalanced datasets are common; this decreases the certainties of assessed capacity. The quality of geological data can be defined by the number of investigation wells with well logging or the coverage of 3D/2D seismic investigation in a given area (Pearce et al., 2013; Niemi et al., 2016; Chen et al., 2020a). Although the subsurface data can be refined by high-cost site characterization, the uncertainties of subsurface geological data are incredibly high compared with that of surface data, which low-cost and large-area investigation technologies can obtain. The only suitable method to reduce the uncertainties of the geological model without sufficient detailed site characterization is stochastic approaches using data assimilation and synthesized modeling technologies based on available statistical data (Popova et al., 2014; Dilmore et al., 2015). The geological model can also be integrated into data compilation tools; then, the fluid dynamic analysis can be fulfilled by site performance tools, such as fluid dynamic analysis tools compatible with GIS and geological modeling tools.

The surface geological data can be an extension of sub-surface geological data when the data resolution is insufficient. However, the data precision of surface geological data is much higher than that of sub-surface data in most cases for abundant methods of the data acquisition and existing dataset. Consequently, surface geological data can be compiled independently. The surface geological data, especially the faults, outcrop data, geological sequence, and landform, are always presented using GIS tools by a vector (such as point, polyline, polygon, polyhedron, class, and others), raster data types, and data stacks by other mathematical tools. The spatial resolution is much higher compared with that of sub-surface data.

The non-geological data, especially surface transportation, railway, industries, mining, weather, precipitation map, legislation, economic, and social data, are always presented in GIS tools by vector or raster data types and other mathematical tools using data stacks. The acquisition technologies and resolution of surface non-geological data are considerably different from that of sub-surface geological data. A rigorous statement of accuracy can be used with statistical descriptions of uncertainty and error. For example, in raster-type data, the resolution is the effective size of each grid cell expressed as the length of each cell (or area). The units can be in (arc) degrees, minutes, or seconds in the geographic coordinate system, or meters, kilometers, and other units in the projection coordination system. Data resolution increases with the decreasing size of the cell (Naumova et al., 2006). The spatial resolution of data is shown in the form of scales, which is the smallest distance (or cellular size) that can be represented, such as 1, 250, or 2,500 m in the map of 1:2, 500, 000 scales. Most surface data on non-geological features can be obtained with a spatial resolution ranging from tens of meters to several centimeters by a series of mature technologies, such as remote sensing or multiple spectrum photographic surveys by satellites or unmanned aerial vehicles, interferometric synthetic aperture radar, geodetic surveying, national site investigation, satellite investigation (remote sensing, remote spectrum sensing, and global positioning system), and other satellite-, flight-, and vehicle-based surveying technologies with high precision. Meanwhile, the data resolution of non-geological data is much higher than the general resolution of sub-surface geological data with an average spatial resolution of several meters or tens of meters. The resolutions of surface data are sufficient for capacity evaluation at various scales, such as reservoir and site scales. Most GIS data are spatially analyzed in the form of features, polygons, or raster data. Given specific social, economic, legislative, and regulatory constraints, the uncertainties of CO₂ capacity evaluation mainly come from the available data and algorithms of sub-surface geological features rather than surface features.

This data quality review clarifies that future efforts focused on site characterization and data collection of geological formations with favorable reservoirs may provide a more useful settlement for capacity evaluation and feasibility studies on CO₂ aquifer storage projects.

When statistical data are available, the storage efficiency coefficient can be obtained by applying the commonly established static algorithms using pore volumes of geological formations and storage coefficients under desired conditions, e.g., mass-balance conditions. The statistical algorithms for storage capacity efficiency can be simplified as a product of various components, especially when the correlation coefficients are weak. In general, the statistical distributions of various components are diverse; the logistic-normal and normal distribution functions were mostly chosen to describe geological parameters and the storage efficiency coefficients (Middleton et al., 2020).

Most-capacity evaluation methodologies currently use volumetric-based or mass-balanced approaches for estimating theoretical CO₂ capacity in a geological medium at regional and sub-basin scales (Goodman et al., 2011; Goodman et al., 2016). The US-DOE, Carbon Sequestration Leadership Forum (CSLF), International Energy Agency, Greenhouse Gas R&D Programmer, and the United State Geological Survey (USGS) have independently developed methodologies for capacity assessment of CO₂ storage in open aquifers (Bachu et al., 2007; Bradshaw et al., 2007; Co2Crc, 2008; Zhou et al., 2008; Iea-Ghg, 2009; Kopp et al., 2009a; Kopp et al., 2009b; Goodman et al., 2011; Goodman et al., 2013; Doe-Netl, 2018). These most cited approaches have been applied around the world for basin- and country-scale assessments (Bachu, 2015). These approaches are based on similar assumptions on storage mechanisms, such as free gas trapped by the stratigraphic structures or hydrodynamic systems, solubility, reaction, or residue trapping. The US-DOE method calculates the CO₂ storage capacity based on a volumetric approach with sweeping efficiency by hydrodynamic processes. The CSLF method states that the theoretical capacity is the maximum amount of CO₂ that can be stored in the pore space minus the irreducible water saturation. The USGS method assesses capacity using both residual and buoyant trapping mechanisms in the open part of the aquifer (Brennan et al., 2010; Aminu et al., 2017). Only structural and stratigraphic trappings were considered as key storage mechanisms rather than hydrodynamic trapping (Bachu, 2015; Aminu et al., 2017). Given the same technical schemes and conditions of geological stratum, most methodologies can be equally applied to aquifers or regions of interest; these methodologies and approaches are equivalent through some conversions among various factors, such as storage efficiency coefficients (Brennan et al., 2010; Goodman et al., 2013). However, closed and semi-closed systems’ storage capacities are significantly different from those in open systems (Zhou et al., 2008; Bader et al., 2014; Elenius et al., 2018). The compressibility/expansion-based (or pressure-limited) algorithms for closed and semi-closed systems assume that injected CO₂ displaces natural brine and occupies additional pore volume caused by pore geometry expansion and brine compressibility during the pressure buildup processes; consequently, the assessed results are limited (Zhou et al., 2008).

The basis for capacity estimation is essentially the integration of the production of the volume of storage formation, storage efficiency coefficient, and average CO₂ density at reservoir conditions ( ρ C O 2   ) (Doe-Netl, 2018), as follows: G C O 2 = ∫ ρ C O 2 ⋅ φ t o t ⋅ E s   ⋅ d V = ∭ ρ C O 2 ⋅ A ⋅ h g ⋅ φ t o t ⋅ E s   ⋅ d x ⋅ d y ⋅ d z , (1) G C O 2 = ρ C O 2 ⋅ V b u l k ⋅ E s = ρ C O 2 ⋅ A ⋅ h g ⋅ φ t o t ⋅ E s , (2) where V b u l k   is total pore volume of geological formation for assessment [L³]; ρ C O 2 is CO₂ density under reservoir conditions [M/L³], A is the surface area for reservoir [M/L³],   φ t o t is the porosity of reservoir [-], E s is storage efficiency coefficient [-], which is defined as the proportion of available pore volume accessible for storage. E s reflects the fraction of a given geological volume in which CO₂ can be effectively stored (Gorecki et al., 2009b). For high-resolution evaluation, the resource can use discrete methods by dividing the site of interest into cellular or grid aggregation (Eq. 1). The cellular size is determined by the detailed level of geological data and the resolution requirement of evaluation.   G C O 2 = ∑ i = 1 N ∑ j = 1 M ∑ k = 1 L G C O 2   i , j , k = ∑ i = 1 N ∑ j = 1 M ∑ k = 1 L ( A i , j ⋅ h i , j , k ⋅ φ i , j , k ⋅ ρ C O 2   i , j , k ⋅ E s   i , j , k ) , (3) where G C O 2 [M] is the mass estimate of CO₂ capacity; A [L²] is the surface geographic area defining the geological volume for storage;   G C O 2   i , j , k [M] is the CO₂ capacity of the cellular i, j, k being assessed within the region; h i , j , k is the gross thickness of the cell i, j, k [L]; φ i , j , k [-] is the total porosity of the assessed formation volume; ρ C O 2   i , j , k is the density of CO2 evaluated at storage conditions [M.L ⁻³]; N ,   M ,   L is the maximum index number for i ,   j ,   k dimension, respectively; and E s   i , j , k is the storage efficiency coefficient in cells i ,   j ,   k and is the product of several factor components [-]. In general, the storage coefficients increase with decreasing evaluation scale, and the uncertainties of CO₂ source decrease with reducing evaluation scale.

When the grid or cellular sizes differ, the variable grid or cellular methods are more flexible methods that allow for capacity assessment with different spatial sizes with different data quality (Bauer and Rose, 2015). Storage efficiency coefficients also depend on storage mechanisms acting at different spatial scales (cellular size) and temporal scales. At national and regional (basin) scales, the empirical and analytical methods can speedily obtain reasonable resolution storage efficiency coefficients (Iea-Ghg, 2009). At the site scale, detailed geological models and reservoir modeling tools can be used for storage efficiency coefficients and storage capacity with more storage mechanisms and higher resolutions.

Multiple physical and chemical coupling processes must be embodied in complex geological models and assessment tools to apply more trapping mechanisms accurately. Using existing algorithms and tools, CO2 capacity evaluation with detailed site characterization can achieve very high resolution (Wei et al., 2015a; Rezk and Foroozesh, 2019; Wen and Benson, 2019). Advanced approaches with more data can provide more reliable capacity results considering more storage mechanisms and more detailed site characterization data over a desired period, especially with monitoring and production history data (Rezk and Foroozesh, 2019; Wen and Benson, 2019). However, no single approach can simulate all these coupling processes of trapping mechanisms reliably at once, nor is such a model necessary for practical purposes.

Current dynamic approaches rely highly on geological models and numerical algorithms with limited resolution and inconclusive factors before detailed site-specific data and history matching (Bachu, 2015). By contrast, the analytic and simplified numerical approaches are more flexible and applicable with precious data of limited site characterization and computation resources. Accordingly, the static approaches with statistical and analytical algorithms have been used broadly and routinely in large-scale capacity assessments compared with the dynamic methods because of more flexible and applicable with limited site characterization. (Cslf, 2008; Doe-Netl, 2018).

Analytical algorithms using several assumptions can quickly obtain storage capacity. These analytical algorithms include those for multiphase flow, semi-analytical algorithms of multiphase (two-phase) flow, solute-transport models of multiple phases and multiple species, coupled multiphase-reaction-temperature algorithms, coupled geomechanics-flow algorithms, and others (Claridge, 1972; Nordbotten et al., 2005; Okwen et al., 2010; Szulczewski et al., 2014; González-Nicolás et al., 2015; Ganjdanesh and Hosseini, 2018; Middleton et al., 2020; De Simone and Krevor, 2021). The analytical algorithms are preferred because they require a relatively small amount of data based on the idealized or conceptual model and can provide quick assessments with acceptable resolutions, especially for basin or sub-basin scale evaluations with limited data. However, the analytical algorithms must use several assumptions to solve the equations mathematically but miss important mechanisms of the CO₂ storage, such as heterogeneity of aquifer formation, injection strategy, buoyancy, mobility ratio, multi-phase dissolution, rock-brine-CO₂ interaction, and others. Consequently, the usage of these algorithms should be careful under certain conditions.

These approaches can also be grouped into deterministic methods and stochastic methods in the perspective of data of site characterization. The geological formation has substantial spatial heterogeneity of physical and chemical properties due to a complex history of sedimentary, tectonic, and diagenetic processes, and the heterogeneity causes significant uncertainties in capacity and site performance assessment (Burruss et al., 2009; Lv et al., 2015; Han and Kim, 2018; Jayne et al., 2019; Wen and Benson, 2019). The data of site characterization under the development stage are still sparse and have extremely high uncertainties. Providing limited data with high uncertainties, the only appropriate and reasonable way to describe the capacity uncertainty is using deterministic approaches based on the statistical data of site properties (Popova et al., 2014). The stochastic methods can be implemented using the statistical properties of site properties as input parameters. These statistical distributions can be in the logistic normal, normal distribution, and other forms (Popova et al., 2014). Statistical data from underground resource recovery projects are helpful to determine the storage efficiency coefficients. Organizations and researchers have established several global databases, including a large volume of reservoir data on geological formations with different lithologies and depositional environments, structures, and traps to determine storage coefficients based on examination of worldwide existing CO₂ storage projects and properties data on hydrocarbon reservoirs (Gorecki et al., 2009d; Iea-Ghg, 2009). For high-resolution evaluation, provided that each cellular with storage potential has various parametric distribution functions for each storage efficiency, the individual p values of different storage factors are multiplied to determine the distribution of storage efficiency coefficient   E s   i , j , k for cellular i, j, k.

Multiple numerical tools using different algorithms have been used worldwide, such as TOUGH2, ECLIPSE, GEM, CO₂-PENS, STARS, NUFT, TRANSTOUGH, MODFLOW, FLOTRAN, SIMUSCOPP, STOMP, MORES, finite element heat, and mass transfer code (FEHM), novel reservoir monitoring, modeling, and simulation (NORMS), MATLAB reservoir modeling tools (MRST), and other tools (Ennis-King and Paterson, 2007; Pruess and Spycher, 2007; Nordbotten et al., 2012; Ranjith et al., 2013; Teletzke and Lu, 2013; Celia et al., 2015; Møll Nilsen et al., 2015; Rezk and Foroozesh, 2019; Wen and Benson, 2019). With a comprehensive geological model based on site characterization, numerical simulation can determine the distribution range of storage efficiency coefficients (Yoshida et al., 2016). Numerical algorithms are capable of providing more flexible and precise results than statistical and analytical algorithms. However, the uncertainty that stems from numerical tools and numeric algorithms incorporating various storage mechanisms persist.

The integral modeling of multiple-phases fluid properties, CO₂ plume behavior, pressure spreading, and reactive-transport process, mechanic process at various temporal and spatial scales depend greatly on storage mechanisms, appropriate geological model and gridding, cross-scaling of geological properties, upscaling methodology, and result interpretation, but less on numerical modeling algorithms (Nordbotten et al., 2012; Teletzke and Lu, 2013; Thanh and Sugai, 2021). Uncertainty modeling, which uses statistical data and stochastic tools to improve the predicted results, may measure the uncertainties of CO₂ capacity to a certain extent, but it is inadequate for describing the overall uncertainties. History matching using time-lapse monitoring is essential to enhance predictions on the site’s long-term performance and CO₂ behavior underground (Nordbotten et al., 2012; Jenkins et al., 2015; Chen et al., 2020a). Furthermore, site characterizations and experiments at multiple scales ranging from pore scale to site-scale reveal the basic parameters of the storage process. These basic parameters depend on characteristics and upscaling methodology at a smaller scale, such as pore geometry, capillary pressure, rock and fluid properties, interfacial tension, wettability, pore geometry, molecular diffusion, hydrodynamic dispersion, water salinity, surface minerals, as well as the mineralization and precipitation process (Pruess et al., 2004; Middleton et al., 2012a; Yoshida et al., 2016). The appropriate geological model and gridding that reflects the cross-scaling of complex geological properties by site characterization, complex properties with multiple phases fluid, algorithms reflecting various trapping mechanisms, and heterogeneous reservoir properties are the keys to resolving the uncertainties in numerical simulation (Middleton et al., 2012a; Bouquet et al., 2016; Yoshida et al., 2016). Meanwhile, essential questions relating to CO₂ storage cannot be predicted convincingly to a satisfactory accuracy with existing numerical simulation tools, even for highly idealized problems (Nordbotten et al., 2012).

Authority must be verified by applying sensitivity analyses or stochastic analysis of key variables based on field or statistical data, especially for complex reservoir-seal systems (Pawar et al., 2017; Alcalde et al., 2018; Jin and Durlofsky, 2018). A quick way to simulate an entire reservoir is the application of ROMs, which can understand complex processes with acceptable computational efficiency (Pawar et al., 2015; Chen et al., 2020b; Middleton et al., 2020). The development of ROMs requires a series of simulations or calculations of detailed component models for reservoirs, wellbores, caprock, faults, and aquifers; then, ROMs can be integrated to predict site performance, economic feature, and geological risk (Pawar et al., 2015; Chen et al., 2020b).

The ROMs for CO₂ injection in heterogeneous reservoirs are used to quickly estimate site performance and CO₂ capacity based on values of key dimensionless scaling groups (Stauffer et al., 2011; Harp et al., 2016; Pawar et al., 2017; Jin and Durlofsky, 2018). This algorithm combines simplifications of full-order flow simulation, linearization of a nonlinear system, projection into a low-dimensional sub-space using proper orthogonal decomposition, or other ways to reduce the complexity of computation and storage mechanisms (Jin and Durlofsky, 2018). The ROMs link basic parameters, storage mechanisms, and site performance assessment. They are more efficient for high-effort and quick simulations than conventional simulations; nevertheless, they cannot decrease the uncertainties similar to numerical simulation.

Algorithms should make the best of limited subsurface data. For example, based on known geological theory and site characterization data, geological interpretation tools can be used to generate a spatial distribution of reservoir parameters reflecting the correlativity. Then, the proper algorithm can be selected to assess the geological storage efficiency coefficient (Popova et al., 2014). However, suppose the data scarcity varies in different regions. In that case, the variable grid or cellular methods with hybrid algorithms are more flexible methods that allow for capacity assessment with various data quality while still preserving the overall spatial trends.

Hybrid algorithms can integrate different algorithm components and related datasets in a comparative way for the assessment of site performance and capacity. The hybrid algorithms can start with volumetric calculations underpinned by deterministic statistical models with limited site data and progressing through probabilistic analyses, and dynamic storage assessments using reservoir simulation with dynamic heterogeneous reservoir models that compile and assimilate detailed site characterization.

The mathematical theories, evaluation procedures, and data requirements for the above capacity algorithms vary greatly. The available algorithms and tools for storage capacity estimation can be grouped into several large class sets as empirical, semi-analytical, ROMs, numerical simulations, and hybrid algorithms. The precise and detailed comparisons of existing algorithms have been carried out (Pruess et al., 2004; Bachu, 2008; Goodman et al., 2013). It illustrates that currently available simulation codes could model the complex phenomena with quantitatively similar results but significant discrepancy from fluid properties and discretization approaches (Pruess et al., 2004; Nordbotten et al., 2012).

Future work should focus on site characterization, data collection, advanced data assimilation, and highly effective algorithms that reflect the effects of storage mechanisms to reduce uncertainties in the capacity evaluation and provide the uncertainty ranges of each dataset, algorithm, and integrated method. Among these, the data quality of site characterization is of priority.

The CO₂ storage project aims to find a suitable site with favorite storage volumes and injectivity. These characteristics can be estimated based on storage cost using available storage volume and injectivity parameters (Mathias et al., 2015). The algorithms might strongly correlate with those for storage mechanisms (Mathias et al., 2015).

CO₂ capacity is constrained by geological volume and related risk, which allow the areal and vertical spread of CO₂ plume without significant impacts; consequently, a crucial task is to specify the influence volume and surface area that can be assigned for CO₂ geological storage. The primary risk is leakage of CO₂ and brine with/without dissolved CO₂ into overlying strata, protected aquifers, shallow soil zones, and the atmosphere, and other health, safety, and environmental (HSE) impacts. Considerable experience has been gained on managing site performance and long-term risk containment and identifying key uncertainties that need to be targeted (Pawar et al., 2015). Potential leakages involve wellbores, active faults, fractures, assigned boundary impact site performance, long-term containment migration, HSE risks, public perception, and market risks. Neither permeable pathways nor reactivation by CO₂ injection should happen. The safety of storage sites depends on the integrity of cap-rock with closed faults/fracture networks and abandoned wells that have a certain possibility of occurrence (Zoback and Gorelick, 2012).

As one part of site suitability algorithms, many algorithms can be applied similarly. Algorithms such as Bayesian network, CO₂-PENS, multi-criteria method, fault tree, certification framework, QPAC-CO₂, NRAP, and other algorithms and related tools have been developed for quantitative and qualitative risk assessment applications (Price and Oldenburg, 2009; Tanaka et al., 2011; Zhang et al., 2011; Aktouf and Bentellis, 2016; Li and Liu, 2016; Dean and Tucker, 2017; Xia and Wilkinson, 2017; Hnottavange-Telleen, 2018). These approaches can also predict the behavior of the CO₂ storage process and corresponding risk (risk probability and consequence).

Social, regulation, legislation, and environmental constraints mainly stem from the requirements of technical schemes and risk management of stored CO₂. The legislation and regulatory frameworks aim to protect and minimize the impact on environmental, economic, and social aspects, underground and surface resources, such as freshwater, minerals, vegetables, surface water system, national reserve parks, industrial centers, municipalities, and cities (Aminu et al., 2017). The legislative system sets prohibitions and permissions for CO₂ geological storage projects and defines the rights and obligations of stakeholders. The regulatory and legislative constraints include various rules that limit the injection activities, such as maximum bottom-hole injection pressure (e.g., 1.25 times initial pressure and less than fracture pressure), minimum total dissolved solids (TDS) of brine (TDS > 10 g/L), geological volumes or area of review permitted by administrative organizations, storage duration, conflicts with different mining rights, and relevant regions of influence (Bachu, 2015). The social and economic constraints mean that the storage sites should avoid potential negative effects on the surface or underground activities, such as clandestine mining activities, oil and gas reservoirs, geothermal utilization on natural reserves, water sources, and metropolitan and crucial industrial areas. Depending on the combined effect of the factors above, the cellular or rock block conflicted with vital activities or features might not be able to obtain permission from administrative organizations as assigned geological volume to inject any CO₂ (Dixon et al., 2015). The algorithms handling these restrictions can be integrated into storage schemes and risk assessment algorithms addressing risk probability and risk consequence.