Remote Sensing Perspective on Monitoring and Predicting Underground Energy Sources Storage Environmental Impacts: Literature Review

Abstract

Geological storage is an integral element of the green energy transition. Geological formations, such as aquifers, depleted reservoirs, and hard rock caverns, are used mainly for the storage of hydrocarbons, carbon dioxide and increasingly hydrogen. However, potential adverse effects such as ground movements, leakage, seismic activity, and environmental pollution are observed. Existing research focuses on monitoring subsurface elements of the storage, while on the surface it is limited to ground movement observations. The review was carried out based on 191 research contributions related to geological storage. It emphasizes the importance of monitoring underground gas storage (UGS) sites and their surroundings to ensure sustainable and safe operation. It details surface monitoring methods, distinguishing geodetic surveys and remote sensing techniques. Remote sensing, including active methods such as InSAR and LiDAR, and passive methods of multispectral and hyperspectral imaging, provide valuable spatiotemporal information on UGS sites on a large scale. The review covers modelling and prediction methods used to analyze the environmental impacts of UGS, with data-driven models employing geostatistical tools and machine learning algorithms. The limited number of contributions treating geological storage sites holistically opens perspectives for the development of complex approaches capable of monitoring and modelling its environmental impacts.

1. Introduction

The just energy transition is a complex and long-term multidisciplinary process that requires strategic planning, large investments, and adaptation to socio-political changes. It has accelerated in recent years due to climate change phenomena and unstable political conditions such as war in Ukraine [1,2], affecting global fuel prices and introducing significant volatility in the market. The situation raised the discussion of the importance of energy diversification and the transition towards green energy sources. The process should ensure stable supplies to meet current demand. Such opportunities are offered by geological storage, which refers to all activities of storing materials in underground geological formations. It includes storage of energy sources and carriers such as natural gas, oil, or hydrogen, but also trapping and long-term disposal (carbon capture and storage). Underground gas storage (UGS) as part of the energy sector allows countries to create and maintain crucial fuel reserves to secure their energy stability. In temperate climate zones, such deposits are mainly used to meet the demand for heat energy. The reservoirs are usually replenished in summer when gas consumption is lower and are used in autumn and winter [2]. Therefore, many UGS facilities operate in a seasonal cycle with the possibility of increasing the number of cycles [3].

Geological storage, including also carbon capture and storage (CCS), is in line with the United Nations’ (UN) Sustainable Development Goals (SDGs). Goal 7 of “Affordable and Clean Energy” fosters common access to electricity, clean cooking fuels, technologies, and an increasing share of renewable energy sources. It should be achieved by moving from conventional sources toward green energy sources, including hydro, solar, wind, bio, hydrothermal and hydrogen energy [4,5]. Next, target 2 of the 12th SDG “Ensure sustainable consumption and production patterns” states that by 2030 sustainable management and use of natural resources shall be achieved. Moreover, the release of chemicals and waste should be managed to minimize the adverse effects on humans and the environment [6]. Finally, SDG 13 targets global warming and climate change [7]. Conventional energy sources, namely fossil fuels, are known to have a harmful environmental impact [8,9]. During energy production, carbon dioxide (CO₂) is emitted, contributing to the greenhouse effect and global warming. To mitigate the effect and meet the SDGs, CO₂ is injected into underground geological formations in the CCS process [10].

In addition to the positive effects of geological storage, including their role in reducing greenhouse gas emissions, the negative environmental effects of their operation should also be discussed. Many conceptual and pilot projects have been carried out over the last 20 years. The studies focused on site characterization and feasibility analysis [11,12,13,14], risk assessment [15,16], construction [17,18], site development [19,20,21,22,23], storage technologies, in and ex situ tests [24]. Underground storage facilities require the monitoring of various parameters to verify the integrity of a reservoir. These include monitoring seismic activity, leakage detection, geochemical analysis of stored material and host medium, and ground movements. The influence of geological storage extends beyond the subsurface sphere and interacts with the entire environment in a complex way. It is necessary to consider underground gas storage sites from a wider perspective and then select optimal monitoring methods that ensure sustainable management and safe operation. One of the promising options discussed in this paper is the use of remotely sensed data. Airborne and spaceborne sensors provide information on land, surface water, and atmosphere with high temporal and spatial resolution that allow monitoring UGS sites on a large scale. However, the number of studies on geological storage monitoring based on remote sensing (RS) is yet limited. Combined with advanced data-driven methods and machine learning (ML) approaches, RS could provide a comprehensive tool for cost-effective and accurate research and prediction of the environmental impacts of geological energy storage. Our review focuses primarily on the underground gas storage, but it also incorporates underground hydrogen storage (being a particular type of UGS) and CCS, as there are various analogies in the observed impacts and monitoring methods.

The aim of the review is to evaluate the current state in studies of UGS facilities with a particular focus on RS monitoring and data-driven modelling of selected environmental impacts. We seek to answer the following research questions.

• How do geological storage facilities influence the environment, and what impacts are considered to be the most critical in terms of sustainability and long-term stability?
• What surface monitoring techniques are used at UGS facilities, and how do the capabilities and limitations of these techniques affect the effectiveness in detecting environmental impacts?
• What are the critical gaps and future research directions for integrating multi-source remote sensing data and data-driven models to create a holistic monitoring framework for UGS sites?

To address these questions, our study starts with systematizing knowledge on geological storage, with a focus on salt caverns, its environmental impacts, and critical assessment of established monitoring and modelling methods. Based on this review, we follow with an analysis of current trends, challenges, and likely future developments related to the progress and proliferation of open remotely sensed data and data-driven processing methods with regard to their applications in the field of UGS monitoring.

The review was prepared in a narrative way to guide the reader through the current status and emerging trends of RS applications in monitoring and data-driven predictive modelling of the environmental impacts of UGS with particular attention to natural gas (NG), hydrogen (H₂) and CCS. The paper is structured as follows. Section 2 describes the review methodology. Section 3 summarizes the findings of the review, including the fundamentals of geological storage, namely the geological types and the materials stored with the aim of organizing the associated environmental impacts. Then, surface monitoring methods are presented followed by modelling and prediction methods used for analysis and assessment of the impacts of UGS operation. Finally, current trends and perspectives are discussed in Section 4.

2. Review Methodology

The review was carried out using the Web of Science (WoS) database. Underground geological storage is used for various purposes and due to analogies in development and operation, the search query was constructed using terms corresponding to the following types of energy and CO₂ storage: UGS, underground hydrogen storage (UHS), CCS, and compressed air energy storage (CAES). Keywords were combined with the Boolean OR operator to ensure that all contributions are related to at least one of the storage types mentioned above. The phrase monitoring was then added to retrieve contributions related to all types of monitoring (Figure 1). The query was limited to 500 most relevant results according to WoS. A total of 325 publications were rejected after the initial screening. During the review, further contributions were removed manually from the collection, as they focus of subsurface monitoring, such as geophysical, geochemical, seismic and in situ methods. Contributions on given subjects were excluded from the analysis. The focus in the paper is put on remote sensing techniques, particularly those involving airborne and satellite sensors, which are primarily used to monitor surface impacts. The collection includes publications published in the first part of 2025. The review is based on 191 publications related to geological storage and 57 related to the topics of mining, geostatistics, and ML. The topics were aggregated and linked based on geological formation and stored material, then presented in chart form. The environmental impacts were summarized in a table according to UGS types, listing the primary advantages and disadvantages of each. Additionally, the monitoring methods were classified by acquisition level and sensor type. A summary of relevant contributions concerning passive and active remote sensing for monitoring UGS was provided in tabular format.

3. Review of Geological Storage

The available literature reports projects developed and tested in various geological formations for the purpose of UGS, UHS, CCS, and CAES. The links between the types of rock formation and the stored materials are shown in Figure 2 and are discussed in Section 3.1. An overview of the environmental impacts of UGS is presented in Section 3.2.

3.1. Types and Stored Materials

Geology is fundamental for the development of a storage site, because it strictly limits the location of potential sites [25]. The three basic types of geological storage are depleted reservoirs and deposits, salt caverns, and aquifers (Figure 3). They vary in size, available capacity, and operation schemes. Each facility can be described with basic parameters including total capacity, working gas and cushion gas volume. Total capacity refers to the maximum volume of gas that can be stored in the reservoir. It is further divided into the working gas, which is the volume of gas that can be injected and withdrawn from the field during operation, and the cushion gas. The latter refers to the amount of gas that remains permanently in the storage reservoir to maintain adequate pressure and ensure efficient operation [26]. It is required for deliverability during withdrawal periods and is generally not available for commercial use.

Depleted deposits and reservoirs are the most common type of underground storage worldwide [27,28,29]. These refer to mined out deposits or non-economically recoverable reservoirs [30]. Oil and gas reservoirs are naturally predisposed to receive and store hydrocarbons. They are used primarily for the storage of the same type of material, but in [30], the opportunities and challenges of using depleted deposits for hydrogen were discussed. The geomechanics of UHS in depleted reservoirs in sandstones were reviewed in [31]. There are also cases of UGS and CCS in depleted oil and gas reservoirs [32,33]. Depleted reservoirs have great total capacity, but the cushion gas volume takes approx. 50% of the total volume [26,34].

Another type of UGS is the cavern underground storage [35,36,37]. According to Blanco-Martín et al. [38], the leaching of salt caverns for the purpose of gas storage began in the late 1940s. In contrast to depleted fields, these magazines are usually specially constructed for the needs of storage, and therefore, previous mining activities are not determinant. Rock salt is particularly suitable for the development of storage caverns. It is used for the storage of oil, gas, or radioactive waste due to its properties, such as extremely low permeability and high ductility, which ensure the long-term stability of rock salt [36,39]. It has been used to store H₂ in the petrochemical and chemical industries for decades [40]. In [41] the potential storage of H₂ in salt formations in Poland was studied. However, the construction of caverns is time-consuming, as they are usually created using water solution mining. Debrining determines the success of the UGS project and its capacity [42]. Storage capacity is lower compared to porous formations, but higher injection withdrawal rates can be achieved [43] and the amount of cushion gas is much lower compared to other geological storage types [20]. It takes approx. 25–30% of the total volume [44]. Rock salt, in addition to its proven storage ability, serves as a natural geological barrier that prevents leakage [45]. The gas pressure must be set and maintained at a certain level to limit the creep of salt and avoid mechanical damage to the wall [46].

Caverns can also be created in carboniferous formations, e.g., unmineable coal seams, abandoned coal mines [47,48] or hard rock, such as granite in the Czech Republic [49], China [50,51], India [52] and Japan [53]. The results of a pilot CCS project in basalt, in Wallula, USA, were described in [19]. Basalt as reactive rock is used for CCS as CO₂ forms stable carbonates [54]. The process is known as mineral trapping. The CO₂-Vadose project focused on controlled injection of CO₂ into abandoned underground limestone quarries and analysis of diffusive migration within the system [55,56]. There were attempts to develop UGS in the declining uranium Rožná deposit (Czech Republic) [57]. CCS sites are also developed in closed underground coal mines [24,58] where CO₂, as a byproduct of combustion in power plants, is injected back [59]. Unmineable coal deposits are used for underground coal gasification (UCG) [60,61]. Storage caverns can be lined with concrete. Other cases include such constructions in hard rock [62,63], gneiss [64], or salt [65]. Cement is also used in depleted reservoirs to line the wellbore [27].

Porous rocks and aquifers can be used to store gases in cases where the porous sedimentary rock formation is overlaid with impermeable cap rock [43]. The natural barrier seals the storage and prevents gas from migrating. Underground storage sites in porous media are developed for various materials, most commonly natural gas, CO₂ [66], H₂ [67], but they are also tested for CAES [68,69]. Aquifers provide the largest storage volume among all geological formations, but a significant portion reaching 80% of the total capacity, is occupied by cushion gas [26]. Hydrogen in gaseous form requires large storage volume, which is available in subsurface porous media [70] and saline aquifers [71]. Saturated sandstone was tested for hydrogen storage in [70], drawing attention to the unwanted residual trapping that causes hydrogen losses, as it is impossible to recover it. The issue of low viscosity and greater H₂ mobility compared to NG or CO₂ in aquifers was raised in [72]. Problems relate to maintaining recovery efficiency. On the contrary, such a phenomenon is desired in the case of CCS, because it increases the storage capacity. Other cases described storage activities in saline aquifers [73,74].

Oxygen (O₂) found in UGS facilities can be an impurity [75], or an additive to biomethane [76]. However, it may solubilize in the formation water, leading to changes in the microbial communities and mineral oxidation of the surrounding rocks. Caverns have been used for compressed air energy storage [77,78,79]. Compressed ambient air is stored underground as a working fluid, which is then extended in turbines to restore electrical energy [77]. Geological storage also allows storage of radioactive waste [49,80] and storage of thermal energy [81]. Since the technology for handling radioactive and nuclear waste differs significantly, it is not included in the paper.

3.2. Environmental Impacts

The most common potential negative effects of UGS on the environment discussed in the literature are shown in Figure 4. These include leakage, ground surface subsidence and uplift, groundwater pollution, seismic, geochemical, and microbial activity. A summary of the observed impacts and the main characteristics of the geological storage types are given in

Table 1. Summary of the observed environmental impacts according to the type of geology and material stored.

Geological Type	Environmental Impacts	Stored Material *	Benefits and Limitations
Aquifer	Blowout	Hydrogen (H2), natural gas (NG)	+ Working gas capacity greater than in cavern storage. + Natural ability to hold gas. − Risk of leakage through improperly sealed abandoned wells. − Reduced storage efficiency due to gas trapping.
	Changes in the microbial community	Oxygen (O2) (in biomethane or carbon dioxide (CO2)), H2
	Mineral oxidation of surrounding rocks, chemical changes in rocks and gas, corrosion
	Pollution of groundwater, soil and vegetation disturbance	CO2, NG
	Seismic activity
Cavern	Blowout	H2, NG	+ Ability to work at a high injection-withdrawal rate. + Salt provides long-term stability and safety. + Salt as a natural barrier prevents leakage. + Storage in reactive (e.g., basalt) allows mineral trapping of CO2 − Requires construction. − Lower volume than in the other types of storage.
	Surface displacement—subsidence and cyclical movement
	Mineral oxidation of surrounding rocks, chemical changes in rocks and gas, corrosion
	Pollution of groundwater, soil and vegetation disturbance	CO2, NG
	Seismic activity
Depleted reservoir	Blowout	H2, NG	+ Greater volume compared to other types of storage. + Overlying impermeable rocks prevent leakage. − Risk of gas loss through pores. − Reduced storage efficiency due to gas trapping and escaping through pores.
	Surface displacement—subsidence and cyclical movement
	Mineral oxidation of surrounding rocks, chemical changes in rocks and gas, corrosion
	Pollution of groundwater, soil and vegetation disturbance	CO2, NG
	Seismic activity

* If no material is given, the impact is independent of the material. Several environmental impacts are specific to the material, while others can be observed in all geological storage regions.

.

The operation of geological storage is associated with cyclic changes in the gas pressure within the reservoir. It interacts with surrounding rocks and can trigger seismic activity, induce earthquakes [82] or reactivate faults [83]. Microseismic events can also occur as a result of a partial roof collapse, gas leakage, or irregular cavern operation [84] in fault zones and therefore create a potential route of gas escape [85].

Leakage is the most discussed risk associated with UGS. Faults and fractures allow material to migrate upward. On the other hand, faults can act as a hydraulic seal in aquifers in the case where the permeability of the filling material is similar to the permeability of the cap rock [86]. The basic classification distinguishes between fault, well, and caprock leakage [59]. Gas is also lost during injection and withdrawal in water-flooded sandstones, as gas is injected under high pressure and becomes trapped in water. A detailed study on gas loss in sandstones is given in [87]. The injection of materials under different pressures into underground magazines generates shear stress, which contributes to the opening of fractures in areas near wellbores [27]. Another cause of leakage is the failure of the cement sheath along the wellbore.

The literature reports NG leaks that lead to blowouts [88] and roof collapse in salt caverns [35,89]. UHS sites, due to the high flammability of H₂, are at particular risk of explosion [90,91]. The potential and risk of hydrogen blowout in salt caverns were discussed in [92]. During leakage, hazardous air pollutants are released into the atmosphere and oily residues settle on the ground [15,93,94,95]. Air pollution causes health problems such as migraines, diarrhea, nausea, and other gastrointestinal problems [88], but also more acute and chronic diseases [96]. Exposure to active oil and gas storage operations can also cause health issues and to mitigate risk, population allocation was proposed in 6 states in the USA [97].

The rocks surrounding the storage site undergo cooling and freezing because of the injection of gases at high pressures and low temperatures [38,62,98]. An ice ring forms around the cavern, exposing it to extreme low temperatures, which can cause fracturing of the rock and pipes [99]. The ice ring functions as a barrier after gas injection and cessation of water drainage [62]. Due to the low temperature of liquid–gas, UGS sites may undergo gradual cooling over time. The temperature drop can reach 1–2 °C over 10 years of active operation [100]. Frozen rock can further alter or block the natural flow of groundwater [101].

As a result of leakage, NG, CO₂, H₂ migrate towards the surface and affect groundwater and soil [102,103] by changing the chemical composition and groundwater level. Newmark et al. [59] pointed out groundwater pollution caused by brine displacement. Saline water migrates and enters potable aquifers [104]. Once CO₂ dissolves in groundwater, it can lead to the mobilization of toxic metals, chloride, and sulphate [105]. Leakage can also affect underground sources of drinking water [106]. Soil pollution with NG, CO₂, or H₂ can be detected by analyzing vegetation stress [107]. CO₂ as a plant nutrient is required for healthy growth. However, a rapid increase in soil causes stress in the roots, leading to plant asphyxiation or soil acidification. Various types of plants can respond to CO₂ leaks and thus have different levels of tolerance to CO₂ [108].

Caverns experience volume loss, due to the strain of surrounding rock, referred to as cavern convergence [109]. Due to the convergence, the rock mass migrates towards the cavern, causing gradual surface subsidence [110,111]. Surface movements are also accompanied by gravity anomalies [112]. Caverns in rock salt are of particular interest, as the mechanical creep properties of salt are crucial in terms of ensuring long-term stability [22,36,113,114,115]. If the pressure inside a cavern is lower than the lithostatic stress, the cavern shrinks [116]. Salt rock undergoes three typical creep stages, namely initial, steady, and accelerated [117]. The subsidence is overlapped with cyclic ground movements caused by the operation of storage sites [43,118,119,120]. Analysis based on UGS case sites located in the Czech Republic and Slovakia showed a strong correlation between ground movements and operation phases, highlighting the temporal shift in observed changes with respect to the UGS operation cycle [121]. Noteworthy, the authors also drew attention to areas of specific geology and fault activity of the UGS within the Vienna basin, where uplift and subsidence occur inversely. Ground movements may affect surface infrastructure located close to the facility, such as pipelines, houses, or roads [122]. Cyclic loading is also responsible for cement fatigue in lined caverns [65].

Hydrogen acts as an electron donor for microbial processes and thus can lead to potential energy losses and compositional changes in stored hydrogen [123,124]. As a result of microbial activity, dangerous substances, such as methane or hydrogen sulphide, are formed [125,126,127,128,129,130]. Hydrogen sulphide (H₂S) is highly corrosive and toxic, severely affecting gas quality and storage infrastructure in caverns or deep saline aquifers [131]. Another issue is the alteration of stored hydrogen [72]. Geochemical reactions between reservoir geology, formation fluid, and stored hydrogen lead to mineral precipitation and dissolution, further impacting the integrity of the reservoir and energy recovery efficiency. In some cases, this can lead to an increase in porosity that binds to the storage integrity. In lined caverns, chemical degradation of the cement is observed. It is caused by the presence of H₂S [131,132] or CO₂ dissolved in brine [133]. Steel-reinforced wells are highly susceptible to corrosion by CO₂ bearing fluids [134]. Gas under high pressure and retrieved carbonic acid cause corrosion of low-carbon steel at a rate of millimetres per year [133]. Operation at high frequency may contribute and accelerate infrastructure corrosion [69].

3.3. Monitoring Methods

Monitoring geological storage sites includes a collection of methods aimed at analyzing the storage infrastructure and its surroundings. Authorities in many countries impose the obligation of monitoring underground storage facilities, but the regulations and scope of monitoring vary [135]. The operators are obliged to perform, among other things, the following duties: observation of the subsidence (surface protection), protection of soil and surface waters from the contamination (monitoring the quality of water and soil air), groundwater protection, and air protection (monitoring of emissions from process equipment and greenhouse gas emissions) [136,137]. Surface monitoring is often limited to periodic ground movement measurements with various temporal frequencies, using well-established methods such as geodetic surveys (Section 3.3.1). Remote sensing methods, although not yet considered as authored sources of spatiotemporal information on the environmental impacts of UGS, can provide insightful results augmenting standard monitoring approaches. These abilities of RS techniques are discussed next (Section 3.3.2 and Section 3.3.3) with reference to the established methods. Figure 5 presents a schematic classification of surface monitoring methods in mining areas.

3.3.1. Geodetic Monitoring Methods

Geodetic levelling measurements, based on benchmark networks, are aimed at monitoring surface displacements [110,139]. Permanent Global Navigation Satellite System (GNSS) stations are installed above UGS sites to continuously collect position data and assess the stability of the area [47,120,135]. There are also documented cases of the application of kinematic GNSS measurements, namely real-time kinematic (RTK) [112,140]. Geodetic techniques are often combined to improve the overall quality of the survey. Further, because of the high quality and precision, they are used as benchmarks for validation and verification of novel approaches such as satellite interferometric synthetic aperture radar (InSAR). InSAR-based monitoring of surface deformations but requires verification with levelling [141,142], GNSS [112,118,120,143,144,145,146,147], or gravimetric surveys [112]. A complex study of UGS fields in the Netherlands using geodetic monitoring methods was presented in [148]. Ground deformations associated with the operation of UGS were observed using GNSS, optical levelling and InSAR data from multiple satellite missions. Geodetic methods are also used underground during and after construction works to monitor stability and for early detection systems. Total stations are used to measure deformations during cavern construction [149]. The measurement results can be combined and compared with the convergence-metres and the displacement-metres.

3.3.2. Active Remote Sensing

The principle of active RS is the registration of an electromagnetic (EM) wave previously emitted from the transmitter. Sensors in an active system measure the return time of the transmitted EM wave and its phase. Additionally, the polarization and strength of the received signal are measured. In synthetic aperture radar (SAR) sensors, the time series of the return waveform’s amplitude and the Doppler frequency shift are used. The principles of active RS and SAR techniques were given in [150].

The unmanned aerial vehicles (UAVs) as a sensor platform are commonly used in monitoring during all stages of mining activity. The applications were discussed in detail in [151], focusing on geological and geophysical surveys for mineral exploration, surveying in surface and underground mines, monitoring of soil and water pollution during the reclamation phase, ecological restoration, and surface subsidence. The light detection and ranging (LiDAR) as an active RS technique, is widely used. Laser pulses are emitted to measure the distance between the investigated objects or surface and the sensor. In addition to the geometry product, LiDAR allows the collection of information on the structure, colour, and other spectral properties of the material or surface. It allows the creation of digital elevation models (DEM) [152] or shaded relief models [153]. Various types of LiDAR, namely differential absorption LiDAR (DIAL), Fourier transform infrared (FTIR), tunable diode lasers (TDL), or Raman LiDAR [154] find application in monitoring leakage at CCS sites. However, the use of Raman LiDAR is limited, as the echo of CO₂ is relatively weak. To successfully detect CO₂ leakage, LiDAR must be placed as close to the surface as possible.

Radar interferometry (InSAR) processing methods were developed for SAR images [155]. InSAR is used primarily to investigate ground movements, deformations of anthropogenic structures, glaciers, peat, and geohazards such as earthquakes or volcano eruptions, but also geological storage sites [156]. The concept is based on data acquired during consecutive satellite flights above the same region at different times. The EM wave phase difference is computed, which is used to determine the relative difference between the topographic heights. The change in distance between the satellite sensor and a point on the surface is measured in one dimension along the satellite line-of-sight (LoS). The measured displacements are processed so that the LoS displacements are translated into vertical and horizontal components of the observed ground movement. Satellites move along orbits and capture data in ascending and descending paths. Processing of images from both tracks improves the accuracy of displacement determination. Horizontal surface movements are determined only in the East–West direction, as due to the near-polar satellite orbit configuration, the sensors are insensitive to movements in the North–South direction [118]. Multi-temporal InSAR (MT-InSAR) refers to all InSAR techniques developed for time series monitoring by compensating for the inherent deficiencies such as atmospheric artefacts or decorrelation noise of differential InSAR (D-InSAR). It is used to analyze surface deformations [33,122,157]. In the D-InSAR method, interferograms are calculated from 2 SAR images captured on different dates and processed to extract the temporal evolution of the interferometric phase, which corresponds to surface displacements [145]. MT-InSAR can be classified into two main techniques, persistent scatterer InSAR (PS-InSAR) [158] and small baseline subset (SBAS) [159], respectively. The LoS displacement time series can be used to create velocity maps and 3D deformation time series with the corresponding gas volume changes [109]. In [160] a combined approach employing PSInSAR and SBAS was used to investigate ground movements in a UGS in depleted gas field, in China. Other studies report the application of SqueeSAR to derive a time series of measurement points in UGS fields in Italy [161] and Germany [162].

Table 2. Monitoring of ground movements in geological storage areas classified according to the type of geological storage and the applied InSAR processing approach including satellite sensors and geodetic reference data.

Study Site	Geological Type	Stored Material	Methods	Satellite Mission	Geodetic Reference	Ref.
Sulcis Coal Basin, Sardinia, Italy	Coal deposit	Carbon dioxide (CO2)	Persistent scatter interferometric synthetic aperture radar (PS-InSAR)	ERS-1/2	Global Positioning System (GPS)—12 permanent stations	[47]
Tvrdonice, Czech Republic	Depleted reservoir	Natural gas (NG)	PS-InSAR	Sentinel-1	GPS—39 permanent stations	[118]
Underground gas storage (UGS) sites in Czech Republic (7), UGS sites in Slovakia (3)	Depleted reservoir (9), aquifer (1)	NG	Stanford Method for Persistent Scatterers (StaMPS)	Sentinel-1	-	[121]
Po Plain, Italy	Depleted gas reservoir	NG	PS-InSAR	Radarsat-1/2, Sentinel-1	Global Navigation Satellite System (GNSS) station	[120]
			PS-InSAR	Radarsat-1	-	[163]
			Small Baseline Subset InSAR (SBAS)	ERS-1/2, ENVISAT, Sentinel-1	GNSS permanent station and seismic monitoring	[135]
			SqueeSAR	Sentinel-1	GNSS	[161]
Ketzin, Germany	Aquifer	CO2	PS-InSAR, SBAS	TerraSAR-X	-	[157]
Krechba, Salah, Algieria	Depleted gas reservoir	CO2	PS-InSAR	Envisat ASAR	-	[164,165,166]
			Differential InSAR (D-InSAR)	Envisat ASAR	-	[167]
Pendleton, Oregon, USA	Aquifer	CO2	SBAS	Radarsat-2	GNSS and gravity measurements	[112]
Hutubi, China	Depleted gas reservoir	NG	SBAS, Point Target Analysis (IPTA-InSAR)	ALOS-1, Envisat ASAR, TerraSAR-X, TanDEM-X, Sentinel-1	GNSS permanent stations	[146]
Shizhuang, Shanxi Province, China	Coal deposit	CO2	SBAS	Sentinel-1	GNSS Real Time Kinematic measurements, unmanned aerial vehicle (UAV) 3D surface model	[142]
Groningen gas field, Norg UGS, the Netherlands	Depleted gas reservoir	NG	PS-InSAR	Radarsat-3, TerraSAR-X, Sentinel-1	GNSS, optical levelling	[148]
XiangGuoSi, China	Depleted gas reservoir	NG	PS-InSAR, SBAS	Sentinel-1	-	[160]
Lower Saxony, Germany	Depleted gas reservoir (1), aquifer (1), salt cavern (1)	NG	PS-InSAR from European Ground Motion Service (EGMS)	Sentinel-1	-	[168]
	Salt cavern (2)	NG	SqueeSAR	Sentinel-1		[162]
Epe, Germany and the Netherlands	Salt cavern	NG	StaMPS	Sentinel-1	-	[169]
Hatfield Moors, the Netherlands	Depleted gas reservoir	CO2	PS-InSAR from EGMS	Sentinel-1	-	[170]

lists notable InSAR applications in studies of geological storage sites.

SAR interferometry is used not only for the analysis of surface displacements but also for further investigation of the underlying processes and geology. It is used to derive general geomechanical parameters of a reservoir [171], or to estimate changes in the volume of the reservoir, and thus detect anomalies [172]. MT-InSAR has been shown to be useful in investigating topsoil moisture [33]. Knowledge of the topsoil helps to fully understand the processes that occur at CCS sites. Changes in topsoil moisture can indicate areas of CO₂ flux. The backscattering radar signal is influenced by the dielectric constant of soil [173] or the salinity of soil, which are used to evaluate the moisture. SAR polarimetry is also used to investigate changes in land cover correlated with seasonal vegetation changes, surface moisture, and variations in water level [152].

3.3.3. Passive Remote Sensing

In contrast to active RS, passive sensors register radiation reflected from the surface. Radiation in the visible, infrared, and thermal regions of the electromagnetic spectrum is registered. Spectral resolution of the data refers to the spectral detection range, number of spectral bands and their width [174]. Depending on the number of acquired spectral bands, sensors are classified as multispectral (several bands) and hyperspectral (tens and hundreds of bands).

The instruments can be installed on the surface, on board UAVs, airplanes, and satellites. Photogrammetric cameras installed on UAVs are commonly used to create DEMs, digital surface models (DSM), digital terrain models (DTM), digital orthophoto mosaics (DOM), and 3D models based on acquired images [140].

Satellite-based sensors are used for Earth observation (EO), as they allow monitoring of the environment through the assessment of selected features. Each surface or material has a unique spectral signature that describes the relationship between the radiation emitted and the reflected radiation by an object. Green vegetation reflects green, and near-infrared (NIR) channels and absorbs red, blue, and short-wave infrared (SWIR) radiation. On the other hand, water has moderate reflection in the visible part of the EM spectrum. The reflectance in selected bands is used as a proxy of vegetation health, enabling analysis of various events. Gas leakage or water shortage cause stress in plants visible as changes in spectral reflectance over time [175,176]. Additionally, mineral alterations can be observed in CO₂ leakage zones. Several contributions that used terrestrial [107,175,177,178,179], airborne, and satellite sensors [179,180,181] are summarized in

Table 3. Contributions on gas leakage detection using passive remote sensing data.

Study Site	Stored Material	Platform	Instrument	Spectral Resolution	Method	Year Ref.
Zero Emissions Research and Technology (ZERT) field experiment, Bozeman, USA	Carbon dioxide (CO2)	Terrestrial	Multispectral imager MS3100 (Geospatial Systems Inc., West Henrietta, NY, USA)	Green (500–580 nm), red (630–710 nm), near-infrared (NIR) (735–865 nm)	Normalized Difference Vegetation Index (NDVI)	2010 [107]
			Multispectral imager, PixeLink PL-B741U camera with CMOS sensor and Thorlabs FW102B filter (NAVITAR, Rochester, NY, USA)	Red (630–670 nm), NIR (780–820 nm)	NDVI, spectral reflectance in red and NIR, regression analysis	2012 [175]
			Hyperspectral imager (Resonon Inc., Bozeman, MT, USA)	160 spectral bands with 3.21 nm channel width, visible—NIR (400–900 nm)	Spectral reflectance in red edge, random forest regression	2009 [181]
			Multispectral imager, PixeLink PL-B741U camera with CMOS sensor and Thorlabs FW102B Filter (NAVITAR, Rochester, NY, USA)	Red (630–670 nm), NIR (780–820 nm)	NDVI, linear regression analysis	2014 [177]
			FLIR photon 320 LWIR camera (Teledyne FLIR LLC, Wilsonville, OR, USA)	Long-wave infrared (LWIR)	Thermal brightness temperature, linear regression analysis
			ASD Field Spec Pro 350 (Malvern Panalytical, Almelo, the Netherlands; Malvern, UK)	1512 spectral bands with sampling interval: 1.4 nm (350–1000 nm), 2 nm (1000–2500 nm)	Classification tree analysis	2014 [182]
		Aerial	Pika II hyperspectral imager (Resonon Inc., Bozeman, MT, USA)	80 spectral bands with 6.3 nm channel width, visible—NIR (424–929 nm)	Red Edge Index (REI), unsupervised classification	2013 [178]
Big Sky Carbon Sequestration Partnership (BSCSP), Montana, USA	CO2	Aerial	Pika II hyperspectral imager (Resonon Inc., Bozeman, MT, USA)	80 spectral bands with 6.3 nm channel width, visible—NIR (424–929 nm)	Unsupervised classification of spectral data, Median Absolute Deviation (MAD)	2017 [183]
		Aerial	Pika II hyperspectral imager (Resonon Inc., Bozeman, MT, USA)	80 spectral bands with 6.3 nm channel width, visible—NIR (424–929 nm)	Stress indicator threshold values, classification of pixels based on stress indicators	2017 [184]
		Satellite	Landsat 8 Operational Land Imager (OLI) (Ball Aerospace & Technologies Corporation, Boulder, CO, USA)	11 spectral bands (433–12,500 nm)
			RapidEye Earth-imaging System (REIS) (Jena-Optronik GmbH, Jena, Germany)	Blue (440–510 nm), green (510–590 nm), red (630–730 nm), red edge (690–730 nm), NIR (760–850 nm)
CCS natural analogue site, Latera, Italy	CO2	Aerial	Daedalus 1268 Airborne Thematic Mapper (ATM) (Daedalus Enterprises, Ann Arbor, MI, USA)	11 spectral bands, visible, NIR, SWIR and TIR, spatial resolution 2.5 m	Ratio NIR/red, EVI, atmospheric resistant vegetation index (ARVI), red edge normalized difference, Vogelmann red edge index, red edge position index, Anthocyan reflectance index, NDVI, spectral signature analysis	2008 [153]
			CASI 2 (Itres Research Limited, Calgary, Alberta, Canada)	15 spectral bands, visible and NIR, spatial resolution 2 m
			AISA Eagle 1K hyperspectral push broom scanning system (Specim, Oulu, Finland)	63 spectral bands, visible—NIR (402.35–989.09 nm)
			Rollei 6008 db45 digital camera (Rollei, Braunschweig, Germany)	-	RGB orthoimage
		Aerial	AISA Eagle 1K hyperspectral push broom scanning system (Specim, Oulu, Finland)	63 spectral bands, visible—NIR (402.35–989.09 nm)	Spectral reflectance in red and NIR, geostatistical and probabilistic analysis, ICA	2011 [185]
		Aerial	AISA Eagle 1K hyperspectral pushbroom scanning system (Specim, Oulu, Finland)	63 spectral bands, visible—NIR (402.35–989.09 nm)	Spectral reflectance in red, NIR and SWIR, geostatistical and probabilistic analysis, fuzzy clustering	2011 [180]
		Satellite	Terra ASTER multispectral Instrument (Ministry of Economy, Trade and Industry (METI), Japan)	14 spectral bands, visible, NIR, SWIR, thermal infrared (TIR) (520–11,650 nm)
Field experiment, Daxing District, Beijing, China	Natural gas (NG)	Terrestrial	SVC HR-1024i spectrometer (Spectra Vista Corporation, Poughkeepsie, NY, USA)	1024 spectral bands, channel width: 1.5 nm (350–1000 nm), 3.8 nm (1000–1890 nm), 2.5 nm (1890–2500 nm)	Spectral reflectance investigation using analysis of variance (ANOVA), Natural Gas Stress Index (NGSI)	2020 [186]
		Terrestrial (platform 5 m above ground)	SOC710-VP spectrometer (Surface Optics Corporation, San Diego, CA, USA)	128 spectral channels, channel width: 4.69 nm (370–1045 nm)	Baseline Slope Index (BLSI) to identity stress, Otsu thresholding	2025 [187]
Sutton Bonington Campus test field, Nottingham University, UK	CO2	Terrestrial	ASD Fieldspec FR Spectroradiometer (Malvern Panalytical, Almelo, the Netherlands; Malvern, UK)	Channel width: 3 nm (350–1050 nm) with 1.4 nm sampling interval, 10–12 nm (1050–2500 nm) with 2 nm sampling interval	Statistical processing and analysis of spectral bands	2016 [179]
					First derivative of reflectance data	2014 [188]
12 UGS sites, California, USA	Methane (CO4)	Aerial	AVIRIS-C (Jet Propulsion Laboratory, La Cañada Flintridge, CA, USA)	224 spectral bands with 10 nm channel width (400–2500 nm)	Analysis of spectral reflectance in the range 2100–2500 nm	2020 [95]
			AVIRIS-NG (Jet Propulsion Laboratory, La Cañada Flintridge, CA, USA)	425 spectral bands with 5 nm channel width (380–2510 nm)
CO2- Enhanced Oil Recovery experimental area, Shandong Province, China	CO2	Satellite	Pleiades High Resolution Imager (HiRI) (Thales Alenia Space (TAS-F), Cannes, France)	PAN (480–820 nm), blue (450–530 nm), green (510–590 nm), red (620–700 nm), NIR (775–915 nm)	Modified and adjusted NDWI	2016 [189]

.

The analyzed literature reports applications of terrestrial sensors at test sites [107,175,177], where the reflectance in red, NIR, and long-wave infrared (LWIR) was measured to detect CO₂ leakage. To provide verification, multispectral and hyperspectral data are analyzed together with in situ measurements of soil CO₂ concentration [179]. In [186] hyperspectral data and ground chlorophyll measurements were used to study plants with stress symptoms after NG leakage. The Natural Gas Stress Index (NGSI) is a vegetation index designed to identify plants exposed to NG [186]. Another index was proposed in [187] to detect microleakage of NG. Airborne and satellite imagery were also analyzed for their suitability in the detection of CO₂ and NG leaks [95,180]. The spectral reflectance in the range of 2100–2500 nm allows retrieving and quantifying methane concentration [95]. Spectrometers (Raman and FTIR sensors) are installed in piezometers to continuously monitor dissolved gasses [190]. In [191] the application of optical fibre sensors installed inside the wellbore for leakage detection by monitoring temperature and strain, was described. The chemical composition of gas can be analyzed by mass spectrometry (MS) [130]. The in situ measurements provide concentrations of CO₂, O₂, H₂, nitrogen (N₂), and methane (CH₄), and their temporal variability are crucial for leakage detection. Multivariate isotope analysis of soil gas composition allows determining anomalies and therefore providing information on gas origin [192]. Passive RS data can augment mapping mining subsidence by analyzing soil moisture variations. It can be derived using the soil moisture monitoring index (SMMI) based on spectral reflectance in NIR and SWIR [193].

3.4. Modelling and Prediction Methods

Modelling of environmental impacts of the UGS operation aims to solve two problem types. Regression is used to study the relationships between observed phenomena and possible causal factors. It is used for analysis of relations and patterns in time and space domains. Regression analysis carried out on archive observations allows for development of prediction models. The second type of modelling problem involves classification. Data are processed in terms of their likeness with the aim of assigning the correct class to a data point based on specific criteria. The process is carried out using either a supervised or unsupervised approach, with ML algorithms increasingly used for this purpose. It is common to adapt methods developed for mining due to several analogies with UGS. Figure 6 shows a general classification of the methods used to determine the ground deformation caused by mining activities. Most methods were adapted to monitoring and modelling UGS sites [111,116,139,194,195]. Notable studies on the subject are described in Section 3.4.1. Modelling methods of other environmental impacts are not included in the classification.

3.4.1. Empirical Models and Influence Functions

Subsidence can be described using the influence functions of mining activity [196,197,198,199,200,201], where the vertical displacement in a given location depends on the distance from the cavern axis at a given time and the so-called angle of influence (Figure 7). The functions were adapted to model the subsidence above the UGS facilities [111,116,139] and to predict the convergence of salt caverns [110]. In [139] a function of subsidence above salt caverns in Etzel UGS, Germany, was proposed. It was a modification of functions proposed by Knothe and Sroka. Later, in [110] a cavern convergence prediction model based on land subsidence data was developed for the same area. The authors used the Gauss-Markov algorithm and the function of Sroka and Schober to estimate the convergence. In [109] the relationship between cavern internal pressure and surface subsidence was studied for both spherical and cylindrical caverns. A dynamic model of subsidence throughout the life cycle based on the Gaussian curve was proposed in [202]. However, the model assumed a vertical cylinder shape of a cavern and pure salt rock. In [111] the subsidence above horizontal salt caverns were studied, as the cavern type causes a different range of observed changes. In [203] a Mogi-based model was proposed to predict surface subsidence above a UGS site. The Mogi model is widely used in volcano seismology, but there are some analogies of UGS in salt caverns to volcanoes. A magma chamber undergoes expansion or contraction reflecting internal pressure changes, which are then transmitted to the surface and cause surface uplift and subsidence. Another application of the influence function and the Mogi model that considers the asymmetric subsidence above a cluster of caverns was shown in [194]. The model was also applied to model surface deformations above a UGS field in Germany [169].

3.4.2. Deterministic Models

Deterministic models are used mainly in the studies of UGS operation and its impact on the rock mass. Geomechanical models describe changes in stress and strain in the storage reservoir [172]. The literature reports applications of finite element method (FEM) in the analysis of deformation of salt rock in potash mines [204], rock mass behaviour during cavern leaching [205], or during the operation of CCS sites [206], different methods of cavern construction [149], and the impacts of injection and withdrawal fluid pressure on fault weak planes, wellbore wall and cement sheath [27]. FEM was used to model the hydromechanical behaviour of various storage cavern configurations [51] or to model the flow field of an underground water sealed storage cavern [206]. In [207] numerical simulations in FEM software were developed to analyze surface subsidence above UGS in salt caverns. While FEM is applied to continuous problems, the discrete element method (DEM) is suitable to solve discontinuous problems. A combination of these methods, the continuous-discontinuous element method (CDEM), allows modelling the evolution of rocks from continuous to discontinuous state [208]. The approach, based on the Mohr-Coulomb law or the maximum tensile strength criterion, is used to model the damage evolution, failure and the stability of an underground cavern storage project.

The modelling of salt creep behaviour is subject to many studies, which can be classified into empirical models, component models, and constitutive models based on the creep mechanism [22,36,114]. The fractional derivative creep damage (FDCD) constitutive models allow for the analysis of salt cavern deformation and shrinkage [36]. The viscoelastic and elastic models of rock salt are used to predict the surface subsidence above UGS sites [209,210]. Studies report that the greater the stiffness of the overlying strata, the smaller the surface subsidence. Creep behaviour modelling has been used to determine deformations above a crude oil storage site in a closed anhydrite mine [211].

3.4.3. Data-Driven Models

Data-based approaches are increasingly being applied in fields that use satellite RS and image processing. These include geostatistical tools for exploratory analysis, interpolation, classification, and regression. Geostatistical analysis is based on statistical methods and concepts applied to data with spatial reference, in which location is an important element of the analysis. Spatial statistics allow us to describe the analyzed dataset using statistical measures. It can be used to study the distribution of a single variable (univariate analysis) or the relationships between more variables (bivariate and multivariate analysis) [212]. Analysis of variance analysis (ANOVA) allows us to identify statistically significant variables. In monitoring UGS operation, it can be used to distinguish spectral bands sensitive to NG or CO₂ leakage [186]. All spectral bands acquired by a sensor are analyzed to identify a significantly different band. Based on the selected bands, a spectral index can be proposed to emphasize vegetation stress.

Exploratory spatial data analysis (ESDA) is used to describe, visualize, and interpret spatial data with the aim of identifying patterns, trends, and relationships [213]. The variables represent selected elements of the environment or factors, e.g., the NDVI, surface subsidence, precipitation. They are analyzed in terms of variability in spatial and temporal domains. The time series approach allows us to study temporal trends and patterns [214].

The spatial distribution of a phenomena is the subject of interpolation tasks aimed at estimating the value of a random variable in an unknown location based on a subset of known measurements. The methods can be classified into deterministic based on exactly predetermined spatial contexts such as nearest neighbour (NN), inverse distance weighted (IDW), and stochastic, which consider random functions, including the spatial dependence between points such as kriging and cokriging. The latter is based on the semivariogram, which reflects the spatial autocorrelation of a variable, and thus quantifies the spatial autocorrelation among measured points and accounts for the spatial configuration of the sample points around the prediction location. In [33] IDW was applied to estimate the spatial distribution of CO₂ concentration from discrete samples, while in [43] it was used to interpolate ground displacements above an UGS site from InSAR measurements. IDW was used to produce continuous ground movement maps above UGS fields in Germany based on EGMS data [168] or measurement points from the SqueeSAR algorithm [162]. Kriging is also reported in the interpolation of InSAR data [163] or the soil gas content from in situ measurements [192]. The spatial autocorrelation expresses the relationship between the value of a variable at one location in space and the neighbouring values of the same variables. It allows for the identification of clusters of similar values. The variables analyzed can be classified into groups depending on their similarity, which is the aim of unsupervised and supervised classification [215].

The analysis of relationships between several variables and their changes in time can be carried out using global and local regression [216]. A model or a set of models is used to describe spatially varying relationships. The purpose of linear regression is to estimate linear relationships in the data. It was applied to spectral reflectance in red and NIR, NDVI, and thermal brightness temperature to identify hot spots associated with CO₂ leaks [177]. Filtration of a spectral image to extract anomalies using an intrinsic random function (IRF) was used to analyze vegetation under gas leakage stress [180,185]. To reduce the dimensionality of spectral datasets, principal component analysis (PCA) can be used [217]. In [148] the independent component analysis was applied to decompose InSAR time series to extract the different sources of ground movements above a gas field and UGS facility. The subsidence due to gas extraction, seasonal movements caused by pressure changes within the UGS, and thermal effects on infrastructures (power lines) were identified as the most significant components.

Machine learning (ML) and its subset of deep learning (DL) are becoming increasingly important in the analysis and predictive modelling of environmental changes in UGS sites. These data-driven approaches are suitable for multifactor analysis of diverse and large RS datasets. The algorithms are used for classification and prediction tasks. The Selection of the ML algorithms depend strictly on the nature of the data use and the size of the input dataset. Image-based data, such as multi- and hyperspectral images or SAR interferograms, are analyzed using, e.g., convolutional neural networks (CNN). Such feature extraction algorithms allow for identification of spatial hierarchies and local patterns, e.g., deformation areas [218]. Another type of neural networks, namely generative adversarial networks (GAN) are used in cases of insufficient data amount, as they are capable of generating and simulating data samples [219,220], which may be particularly useful in cases of uneven data distribution. On the other hand, time series of InSAR displacements or reservoir pressure are processed using algorithms like the long short-term memory network (LSTM), which allow for analysis of temporal dependencies and long-term correlations. Random Forest (RF) and XGBoost are both based on decision trees, but they differ in the architecture. While RF is a bagging model that trains multiple trees in parallel, XGBoost creates a sequence of tree models, which work in a way to improve the final output. As robust method, they are particularly suitable for analysis of structured, tabular datasets [221,222,223].

In [224] unsupervised classification was performed on decomposed InSAR vertical displacement time series to identify ground movements induced by UGS operation. Random forest (RF) regression was used to analyze plant spectra and to identify vegetation exposed to elevated CO₂ levels [181]. The XGBoost algorithm based on regression was used in [225] to deal with uneven temporal sample distribution and interpolate missing pressure values. Non-linear regression methods such as the generalized regression neural network (GRNN), error backpropagation neural networks (BPNN), or (LSTM) are used for prediction of impacts based on the identified relationships. The latter was developed to improve the performance of recurrent neural networks (RNN). GRNN is basically a radial basis function neural network suitable for mapping of non-linear variables. The BPNN algorithm continuously adjusts weights and thresholds through backpropagation. The LSTM is used to create prediction models of, e.g., seepage pressure [226], transient pressure [225] or groundwater level in time series [227] or mining induced seismicity [228] and surface deformations [141]. Wamriew et al. [229] used a CNN to distinguish between seismic events and noise samples, and thus detect microseismic events with an error of locating seismic activity not exceeding 2.2%. Various ML algorithms are tested and compared to each other to select the optimal solution [141,225,227,230].

3.4.4. Hybrid Models

Literature review shows that integration of data-driven approaches and traditional models to form hybrid predictive models, is one of the growing fields in monitoring impacts of geological storage. The latter, including theoretical and empirical models, are built upon physical laws, conservation principles, or statistical relationships derived from past observations. These offer robust and knowledge-based solutions. Conversely, data-driven approaches take the advantage of learning from data and the capability to model complex, unknown relationships. Thus, such models are more flexible compared to traditional that rely solely on physical equations, but on the other hand, they are subject to overfitting [220].

Combined models allow to overcome and limitations of standalone models, as empirical models utilize physical laws, while data-driven models aim at identifying complex, non-linear relationships in observational data. Additionally, prior knowledge of the theoretical and empirical models can reduce the requirement of large training datasets. Hybrid models can, therefore be used in dynamic analysis and predictive modelling of the impacts or operation parameters, which is crucial in managing UGS facilities [231]. However, the integration of diverse models is complex and computing expensive [232]. Conceptual and numerical compatibility of the models poses another challenge.

4. Critical Analysis and Discussion

Geological storage sites are the subject of many research and development projects, widely reported in the scientific literature. These concern the feasibility of facilities, construction, technology, operation, and monitoring. The latter can be divided into 2 primary groups according to the spheres of influence: underground and surface. As geological energy storage is located below the surface, most of the influence is not directly visible on the ground. The critical assessment of monitoring and modelling methods discussed in previous sections is based on SWOT (Strengths, Weaknesses, Opportunities and Threats) analysis [233]. The approach helps to organize the analyzed methods considering their benefits and limitations and thus identify the emerging opportunities and accompanying threats in the field.

4.1. Strengths and Weaknesses

The in situ monitoring of the technical aspects of UGS site, including reservoir integrity, leakage detection, fluid flow, gas concentrations, pressure, and behaviour of host rocks, is not the subject of this paper. Nonetheless, they are referred to in critical assessment, as they can serve as a benchmark for other methods.

Table 4. Side by side comparison of surface monitoring methods.

	Geodesy			Remote Sensing
Measurement Method	Precise Levelling	GNSS, Total Station	UAV LiDAR & Photogrammetry		InSAR	Multispectral and Hyperspectral Imaging
Observation	Vertical displacements	Horizontal and vertical displacements	Horizontal and vertical displacements		Horizontal and vertical displacements	Land surface changes
Precision	Approx. ±1 mm	Approx. ±1.5 to 2 mm	Approx. ±20 mm vertically, ±40 mm horizontally		Approx. ±1 to 2 mm vertically and horizontally in the East–West direction	Depending on the sensor, ground pixel size varies from centimetres to tens of metres
Verification	Not required	Not required	Not required		Comparison to geodetic measurements	In situ measurement, ground truth validation
Measurement frequency	Low (usually on an annual basis)	Low (survey campaigns, on an annual basis) or high (permanent monitoring stations)	High (as requested)		High (every few days)	High (every few days)
Data geometry	Point	Point	Point cloud/pixel		Pixel/point	Pixel/point
Information	Absolute vertical displacements of benchmark points (height difference)	Absolute horizontal and vertical displacements (coordinate difference)	Digital surface/elevation/terrain model		Relative surface LoS displacements or translated to vertical and horizontal (W-E) components	Change in pixel value (reflectance in spectral bands)
Cost	High	High	Medium		Low	Low
Strengths	- Highest accuracy and precision - Reliable - Relatively easy processing that does not require large computing power - Information on absolute surface displacements	- Highest accuracy and precision - Reliable - Relatively easy processing that does not require large computing power - Information on absolute surface displacements	- Detailed information continuous in the space domain - Possibility to carry out measurements with high temporal frequency		- High accuracy in determining vertical displacements - High temporal coverage of data - Wide area coverage - Semi-continuous data coverage in the space domain - Independent of weather and illumination (day/night)	- Information on selected land surface characteristics - Versatile environmental applications
Weaknesses	- Time-consuming - Limited to benchmark locations - Dependent on weather - Unstable reference points as source of errors	- Limited to survey points locations - Unstable reference points as source of errors	- Dependent on weather and illumination - Limited time of acquisition/single flight - Limited area coverage due to limited operational time - Processing of point clouds require large computing power and storage		- Processing requires large computing power and storage - Inability to derive horizontal displacements in the North–South direction - Reduced data coverage due to shadows in varied topography regions - Limited use in areas of vegetation, which is a course of low coherence and measurement quality	- Dependent on weather - Processing requires large computing power and storage

provides a summary of surface impact monitoring methods with their characteristics.

RS data are characterized by spectral, temporal, and spatial resolution. The latter refers to the area covered by a single pixel in the image. Open access satellite missions [154] deliver imagery, both radar and spectral, with ground pixel size ranging from several to tens of metres, making it difficult to investigate small scale local events. The ground sampling distance is crucial in monitoring UGS, as the facilities are relatively small and can be included within a few pixels, significantly decreasing the quality and accuracy of the analysis. In this case, a single pixel represents various surface features causing spectral mixing. As an alternative, commercial projects offer images with submeter (satellite) and centimetre (airborne) resolution. RS data points are distributed evenly in space covering large areas, e.g., a single Sentinel-2 scene is 110 × 110 km. Spectral imagery offers complete coverage of the AOI, while in InSAR the spatial distribution of data points varies depending on the land cover. The persistent scatterers are easily identified in urban areas as there are many surfaces with constant scattering, e.g., concrete. In natural areas, such as mountains, forests, and agriculture, the use of this method is limited [157]. Even though the UGS sites can hold vast volumes of gas, their surface infrastructure is small. Most of the area above the storage is nonurban and covered with various types of vegetation, limiting the utility of InSAR. It is possible to improve the spatial density of the samples in areas without stable points, but at the expense of quality [33]. Installation of corner reflectors within the areas of interest also improves the InSAR monitoring. Currently, multifunctional stations incorporating various monitoring technologies such as corner reflectors, GNSS receivers, levelling benchmarks and ground sensors, i.e., piezometers, are installed [234,235].

Furthermore, spatial resolution can be understood in two ways, as the size of a pixel, and as the distance between neighbouring data points. Geodetic surveys are carried out in lines or networks of established reference points and even though the spatial extent of the network is large, the density of the data samples is moderate. As an alternative, satellite SAR sensors offer coverage of large areas and dense, quasi-continuous distribution of data points. However, depending on the InSAR product used, the ground movement information can be resampled to a lower spatial density, as it is performed in the case of EGMS. The service provides analysis-ready data resampled to a grid of 100 m. In both cases, based on either ready products or independent processing, However the location of a single point cannot be precisely determined. It can be improved by setting up ground control points, namely corner reflectors [236].

Another strength of RS-based monitoring is the high temporal resolution compared to geodetic surveys. Satellite missions such as ESA Sentinel-1 (6 or 12 days) or TerraSAR-X (11 days), provide images every few days, enabling detection and analysis of short-term changes, while levelling campaigns are carried out every few months or years due to the cost and labour and time-demanding, time-consuming nature. Thus, levelling campaigns can provide information related to the general ground movement trend (subsidence) above UGS, and InSAR-based monitoring allows us to capture the short-term movement related to the operation of the facility. However, the nature of ground movement above UGS sites is usually three-dimensional and non-linear, as the subsidence (due to cavern convergence) is superimposed with simultaneous uplift, surface tilt and horizontal translation [169,237]. This poses a major challenge as InSAR techniques are insensitive to horizontal deformations in the North–South direction caused by the near polar orbit of the sensors [238]. Although multitemporal InSAR techniques, such as Multiple SBAS (MSBAS), offer advanced time series analysis, the displacement is limited to one dimension [239]. To retrieve three-dimensional deformations, a combination of ascending and descending tracks is necessary, which poses another challenge due to limited spatial overlap and insufficient coherence [240].

Additionally, satellite imagery is usually acquired on a regular basis and stored in archives, allowing for processing of historical data, analysis of trends, and development of prediction models. The same applies to satellite spectral sensors, as airborne data are mainly acquired on demand in planned missions, so there are no such extensive data archives.

Archive imagery is crucial in modelling the observed phenomena in time. Historical data are used to identify trends, patterns, and relationships, which then serve as a basis in predictive modelling. As the datasets require large computing power and storage capacity, ML algorithms are used for processing. There are various libraries dedicated to image processing and machine learning that can be applied in Earth sciences. The main benefit of applying ML is the optimization and automation of the process. An alternative to processing InSAR data is to use ready-to-use data published online offered by services like The European Ground Motion Service (EGMS) (https://egms.land.copernicus.eu/) that provides data from the ESA Copernicus Sentinel-1 mission [43,241].

Geodetic measurement methods are no longer the subject of research due to their established quality and reliability. They are treated as ground truth and used as a reference in other techniques such as InSAR. To ensure the reliability of RS-based monitoring, it must be verified and validated with ground truth data. Which seems to be the major benefit of RS, is also its weakness. Data are required remotely skipping the time- and cost-intensive survey campaigns. However, the information obtained this way is limited and represents only a part of reality. The analysis allows us to see the change but there is no information on the underlying cause. Environmental studies aimed at classification and feature identification based on multispectral images usually incorporate field samples. This is why most case studies based on RS data, use also other data. Various methods and data can complement each other, contributing to the improvement of the overall quality and reliability of the study. The information is of different kinds and accuracy, e.g., GNSS has a lower quality in determining the vertical component of the displacements while providing high accuracy for horizontal components. On the other hand, the performance of InSAR is better in the vertical, than in the horizontal direction [118].

4.2. Opportunities and Threats

The importance of UGS worldwide is growing, as the available storage capacity increases. It has quadrupled over the last 50 years (as of 2022). According to CEDIGAZ reports, the number of UGS sites increased from 667 in 2022 to 681 in 2023 [242]. By 2050, NG is expected to supply 28% of global primary energy [20,243], while it was reported to provide 23% of primary energy in 2019. Trends in energy production indicate the need to create and further develop storage capacity. In [244] the perspectives of different energy systems based on electro fuels, namely hydrogen, methane and ammonia were discussed, underlining the importance of storage caverns, as they are the most economical. The increasing number of scientific publications on UHS in particular [30,245,246,247], follows the global trend of the green energy transition (Figure 8). With progress in the research and development of alternative energy sources, existing UGS facilities are transformed to UHS sites. However, effective storage and limitation of energy losses during the recovery process remain the main concern [248]. The adaptation of existing and operational hydrogen storage sites, and the challenges involved, was reviewed in [3].

Geological storage projects also interact with society by forming and creating policies, public perception, and a sense of security [249]. The concept of geological storage fits into the objectives of the SDGs mentioned above, contributing to the development of sustainable, safe energy and the combat against climate change. Social acceptance and understanding of such projects are necessary and can be achieved by closer studies of clean energy storage as a holistic process. Currently, there is lack of common international standards that would regulate the surface monitoring of the environmental impacts caused by geological storage activities. Factors like the spatial and temporal frequency of observations, standardized measurement procedures in the industry, and land cover change monitoring could be addressed in the work of international societies such as the International Society for Mine Surveying (ISM) and Solution Mining Research Institute (SMRI). Work on this topic has already been initiated by national bodies such as the Deutscher Merkscheider-Verein e.V. in Germany [250].

RS solutions provide transparent monitoring of environmental impacts, which promotes acceptance within society. The increasing number of satellite missions, both open and commercial, operating on a wide part of the spectrum, contributes to the development of complex monitoring approaches incorporating various data sources and acquisition techniques. The satellite RS sector is expanding, as more open-access and commercial EO projects are being launched. Space agencies carry out surveys on the current state of civil Earth observation missions. All completed, current and future satellite missions and instruments are presented in the Committee on Earth Observation Satellites (CEOS) database by the European Space Agency [251]. The Earth Observing Satellites Online Compendium is run by the Joint Agency Commercial Imagery Evaluation (JACIE) and U.S. Geological Survey [252]. The compendium provides a comprehensive record of EO missions with respect to sensor types. The total number of EO satellites varies between the databases but oscillates around 740. A summary of current and future EO satellites for environmental monitoring is given in [253]. The temporal development of EO satellite missions according to CEOS, is shown in Figure 9.

The intensive development of EO projects contributes to the improvement of data quality and resolution, which will further enable more detailed and accurate spatiotemporal analysis of land, oceans, and atmosphere.

Machine and deep learning tools can handle large EO datasets, providing means to investigate the relationships between different factors. The classification and prediction models can be trained on various data structures, such as satellite optical data, InSAR deformations, land surface temperature, precipitation, but also factors and indicators used in conventional approaches, e.g., indicators used to describe mining subsidence. The main drawback of ML is the requirement of large datasets for training. To limit the dimensionality of the data and optimize the process, approaches like the PCA are applied. Reduction in dimensionality is particularly important in the case of big data, where identifying significant variables is difficult. Therefore, cloud solutions such as Google Earth Engine [254] offer processing and analysis that overcome processing requirements. Open repositories offer ready-to-use products or scripts that can be reused. The use of artificial intelligence (AI) algorithms in studies on the impacts of UGS is increasing. As a comprehensive overview of AI in geoenergy is provided in [255], it is not part of this review.

Passive RS offers data and tools for the analysis of various environmental factors in time and space. Optical satellite imagery and spectral indices can be successfully used in the detection and analysis of changes in land cover, vegetation disturbance, or surface water changes. Changes in vegetation vigour can be the result of leaking or disturbance of groundwater conditions. The latter is a common issue in mining areas. Groundwater can sink or come to the surface and fill the subsidence basin. Images can be processed to identify and analyze spatiotemporal changes [193]. Multispectral and hyperspectral images can be used for the detection of gas leaks, but mainly at CO₂ sequestration sites [175,177,181], while few studies considered NG storage sites [186,187]. Methods developed for sites of natural CO₂ emission, such as a volcanic caldera in Latera, Italy [180], could be transferred to the CCS and UGS areas. Multi- and hyperspectral imaging sensors are used also in quantifying soil geochemical properties and identifying mineral alternations, which can be caused by microleakage [256,257]. They support ground-based monitoring by offering high-resolution maps of various soil attributes. However, the challenges of the complexity of atmospheric correction, spectral mixing, or calibration with ground truth data, still need to be addressed [258,259]. The application of passive remote sensing in gas leak detection has been a subject of research in previous decades. The contributions come from the period 2008–1017 and 2020 and mainly describe experiments in test fields using terrestrial sensors. There are no recent publications on the subject or real use cases.

One of the main challenges in investigating UGS facilities is to adequately identify the underlying causes of observed environmental impacts. Surface movements, as well as seismic activity, may be of natural origin [135]. Moreover, it is necessary to consider the seasonality in phenology, water circulation, and meteorology, as it may be the dominant causative factor for observed changes. In [160], the observed seasonal movements were not correlated to the operation of the UGS, but were attributed to the natural ground movement (areas at landslide risk). Precipitation and soil water content can also affect observed surface displacements. In [170] the nature of surface movements above an UGS site in the UK was studied, highlighting the contribution of overlying peat and changing water content to the observed movements, which dominated the influence of UGS. In the Netherlands, where most of the country’s area is covered by grasslands and croplands with shallow groundwater, the soil surface level changes are observed throughout the year [260]. Peat drainage activities and peat oxidation are directly related to soil subsidence [261]. The sensitivity of plants to CO₂ stress differs throughout their growth cycle. Further, groundwater level and waterlogging can contribute to vegetation disturbance and alter the underground migration of CO₂ [178]. However, the resolution of the optical RS data may be too coarse for the precise requirements of small-scale leakage monitoring [154].

The most recent works on RS-based monitoring and data-driven modelling methods indicate their growing importance in the field. However, the data, especially publicly available, still lack desired spatial and spectral resolution and require further development especially in the field of multisource data integration. Monitoring schemes based on multisensory data are the subject of current research [135,170]. The information provided to the public must be explicitly interpreted. These aspects are within the scope of the bilateral German Polish research and development project CLEAR aimed at providing a web-based smart communication platform to manage impacts of UGS based on available public databases supported by sensor networks, A based analytics and optimization algorithms [234]. These methods can be used as the basis of a multidisciplinary approach, where the problem is treated in a complex way. It is a key to understanding and evaluating the results and conclusions obtained. The construction of monitoring networks and systems in an adequate way is extremely important to ensure that there are no misinterpretations or errors.

5. Conclusions

This paper provided an overview of the environmental impacts of geological storage, along with a critical assessment of recent trends in monitoring and prediction methods with a particular focus on remote sensing techniques. It has been concluded that the research on the subsurface aspects of UGS operation is widely reported, while the monitoring of surface impacts is limited to ground movements. Operators are obliged by regulatory frameworks to carry out monitoring within the facility limits with the use of well-established geodetic techniques. There are, however, satellite-based technologies and machine learning algorithms that could support the baseline measurements, but due to their novelty they have not yet been approved by the administration.

Thus, our objective was to review the state of the art in monitoring surface impacts of the environment in the regions of geological energy storage, including CCS sites, with a particular focus on remote sensing. The article identified research contributions in the field, analyzed the recent developments, and pinpointed the future research directions. The aim was to provide easy access to systemized information.

Studies on the applications of RS in monitoring UGS are sparse while the existing approaches combine RS data with in situ measurements and ground truth data whenever available, to provide verification of the results obtained. RS-based monitoring is still of limited use in areas where alternative measurement methods are not available. Geodetic surveys provide the highest accuracy, but due to the limited spatial coverage and low frequency, they are more and more often supported with satellite InSAR observations [148]. Active satellite data are collected every few days and cover large areas. Satellite data also offer valuable insight into the past, as archive images are available for processing and analysis. Combined with in situ measurements, e.g., with information on stored gas volume [168], they can create a comprehensive source of information on the environmental impacts of geological storage projects.

Geological storage contributes to the SDGs by providing clean and safe energy, conserving natural resources, and reducing carbon emissions through CCS. Responsible and conscious investment with environmental, social and governance (ESG) considerations requires continuous observation of the respective impacts of the project. A multidisciplinary approach, combining various monitoring and modelling techniques, is essential for a comprehensive understanding and evaluation of UGS impacts. The need for further development in multisource data integration and improved data resolution, especially in publicly available datasets, is also highlighted. Development of a scalable methodology will foster the operators and administration in promoting UGS as an integral part of the green energy transition.