Monitoring Surface Water Dynamics in Mining Areas Using Remote Sensing Indices: A Review and Cross-Case Analysis

Abstract

Mining affects groundwater and surface water both during an active mining operation and after its termination. Continuous monitoring and both quantitative and qualitative assessment of water dynamics are crucial for the sustainable management of the mining and post-mining environment. This paper provides an extensive overview of water in the mining industry and of remote sensing methods for surface water monitoring. Moreover, selected spectral water indices are compared to assess their performance and usefulness in surface water monitoring. The Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Modified Normalized Difference Water Index (MNDWI) are applied to different case study areas affected by mining-induced multitemporal surface water changes. All the selected indices were found useful as proxies for surface water identification; however, their effectiveness and accuracy varied in subsequent case studies.

1. Introduction

Monitoring water in mining and post-mining regions is a critical component of environmental management. This involves continuous assessment of surface water and groundwater dynamics, as well as their quantity and quality, both during and after mining operations. Proper water monitoring is essential in minimizing the degradation of the environment and the risk to the human population. Satellite remote sensing-based data enables cost-efficient large-scale water monitoring throughout the mining lifecycle, from operation to post-mining, including the ability to track changes in the water regime back through time [1].

Water resources and water use in mining are the subject of ongoing research. The quality, pollution but also the flow pathways are monitored using various techniques, including in situ measurements and remote sensing (RS) solutions. A schematic diagram of water flows between a mine and the surrounding environment is presented in Figure 1. The water resources can be categorized by origin into atmospheric, surface, groundwater (including both renewable and fossil resources) and sea water. The classification is used during all stages of mining, including exploitation, production, and reclamation. On the other hand, the interaction between mining and mineral processing operations and water resources can be categorized into: (i) withdrawals, (ii) internal use, (iii) consumption, and (iv) discharge [2]. Water use refers to the water input to the process, whereas consumption refers to the amount of water used and unavailable for reuse. Different views on the impact of underground mining (coal in particular) on surface water resources were discussed in [3], as many scholars discuss the direct impacts.

The importance of assessing water risks and water balance in mining was highlighted in [4]. A positive water balance occurs when water accumulates in dams, ponds, and mine voids during heavy rainfall and low evaporation periods. Conversely, mine operations are considered to have a negative water balance when evaporation prevents the accumulation of on-site water. A positive balance requires the on-site water storage facilities to be discharged, while a negative balance requires support from the surrounding water resources. Excessive usage of water can lead to alterations in river flow regimes, drawdown of the groundwater aquifer level, or the reversal of groundwater flow. During mineral resource exploitation, water can not only change its flow direction but also develop point sources of elevated flow. On the other hand, the groundwater level may drop due to mine dewatering, causing water courses to dry up [5] and contributing to the development of a groundwater depression cone [6]. Long-term dewatering of mines can be the cause of hydrological droughts [7]. Changes in groundwater flow, and thus alterations in river regimes, can also be caused by the discharge of excess water [8] and further aggravated by climate change and a prolonged dry season [9]. Furthermore, the excess water is discharged into nearby water bodies, polluting them and increasing their salinity [4]. The topic of acid mine drainage (AMD), water pollution and changes in its chemical composition according to the type of ore (coal and lignite, iron, gold, salt, copper, lead/zinc/silver, and uranium) was discussed in detail in [5]. According to Northey et al. [4], mines include several types of ponds and dams for storing raw water, recycled process water, seepage containment, evaporation ponds, and fire water. Cacciuttolo and Cano [10] reviewed various methods of efficiently managing and using water in tailings. They are used for the storage of residues from the mining-metallurgical process [10]. Monitoring tailing ponds is crucial, as seepage or potential malfunctions could lead to an environmental disaster [11]. The metallurgical process may involve evaporation ponds, as in the case of lithium production [12]. The brine stored in the evaporation ponds is highly saline and may contain toxic elements such as arsenic or boron [13].

Following mine closure, reclamation activities focus on restoring the entire ecosystem, including ground and surface water resources. Mining alters the environment and leads to the development of new landscapes. As reclamation is often conducted under different conditions, the restored area can vary significantly from the pre-mining stage [14]. Secondary ground movements observed after mine closure attributed to the convergence of post-mining voids are referred to as residual subsidence, whereas the rebound of groundwater and restoration of the aquifer layers can cause surface uplift by rewatering and decompressing the aquifers [15,16]. In [16,17,18], the observed upward surface movement was linked to the flooding of the underground workings. In such areas, where land subsidence coincides with the recovery of the groundwater table, the inundation of subsided basins by groundwater may lead to flooding events and the appearance of new water surfaces [19,20,21]. He et al. [22] studied ponding in mining areas, specifically the occurrence of subsidence water in China. As in [4], impacts continued to occur once the drainage stopped. Conversely, in [5], the development of lakes by filling surface depressions with water was presented. Controlled flooding of surface mines and pit lakes are common reclamation methods [23,24]. Pit lakes store excess water and improve surface water retention. This contributes to maintaining the water balance and mitigating negative environmental impacts.

The monitoring of water dynamics and water quality is typically carried out during active mining, whereas observations of abandoned mines at the post-closure stage are often overlooked. These observations are of particular interest, as the completion of mine dewatering leads to changes in the water regime. The restoration of pre-mining water levels in disturbed rock mass can create challenges related to ground stability, including underground voids and slopes of open pits, as well as the risk of flooding in subsided areas. In such cases, remotely sensed data can be utilized as a proxy for detecting and studying the dynamics of water surfaces.

In this paper, we aim to

• Provide an extensive overview of water in the mining industry, including monitoring methods for surface water using RS techniques.
• Assess and compare the performance and usefulness of selected spectral water indices in different mining areas and analyze the change dynamics of surface water in both the time and space domains.
• Propose a monitoring workflow based on case studies presenting various mining environments.

The paper is organized as follows. Section 2 provides a summary of recent research and monitoring methods for surface water features in mining and post-mining areas using RS. A comparative study of three different mining environments is provided in Section 3. The results of water monitoring using multispectral (MS) RS are presented in Section 4.

2. Remote Sensing for Surface Water Monitoring

In situ measurements form the basis for monitoring the water extent, quality and pollution; however, due to limitations in data acquisition, such as labor intensity, limited spatial coverage and safety hazards, they are supported with RS techniques [25]. RS data and processing techniques are widely used and considered essential for this purpose due to their numerous advantages, such as a wide spectral range, compatibility with Geographic Information System (GIS) platforms, quick acquisition over large regions, or their usefulness in hard-to-reach or unsafe areas. Moreover, the use of satellite imagery reduces the cost of data acquisition and enables consistent, long-term multitemporal monitoring thanks to regularly collected data and extensive archives of past observations [26,27]. The selection of methods depends, among other things, on the purpose of research, the size of the study area and its environmental conditions, spatial and temporal resolution needs and required accuracy, as well as time, computing and financial resources.

Therefore, we conducted a consolidated literature review of the RS techniques used for surface water monitoring, focusing specifically on applications in mining and post-mining areas. The review covers literature published until July 2025. The methods described are grouped by the source of the remotely sensed data.

2.1. Airborne LiDAR and RGB

RGB (Red, Green, Blue composition) imagery captured by cameras mounted on airborne platforms delivers surface data with centimeter-level Ground Sampling Distance (GSD). These systems are available in both standard RGB and RGB + near-infrared (NIR) configurations. While RGB imagery can facilitate the identification of water bodies, its reliance on natural color composition can lead to misclassification, particularly when distinguishing water from vegetated surfaces such as forests or meadows. Consequently, multispectral cameras incorporating an NIR band are employed to enhance the accuracy and reliability of water detection by providing better spectral differentiation [28]. Unmanned Aerial Vehicle (UAV) imagery has been used in studies of surface water in mining environments, i.e., to map a flooded open pit mine, analyze surface water properties, vegetation overgrowth and pollution [29], or contamination dispersion and the AMD process [30]. The Green Red Water Index (GRWI), developed in [31], was designed for the detection of surface water in a former coal mine. Airborne HS and MS sensors were used to collect data for the analysis of the concentration of heavy metals in mine water treatment systems [32]. In [33], in situ data and high-resolution MS data were combined to assess the water quality in small water bodies located in former mining areas.

Airborne Laser Scanning (ALS) using Light Detection and Ranging (LiDAR) technology provides high-resolution 3D data in the form of a point cloud. The choice of sensor and laser wavelength varies, depending on the purpose of the measurement. Topographic LiDAR uses a NIR wavelength (1064 nm), which is absorbed by water; therefore, it is mainly used for 3D terrain mapping. This absorption results in a gap in the acquired data that corresponds to water bodies, which can be used to detect surface water [34]. Bathymetric LiDAR used in surveying underwater topography (i.e., the bottom of a shallow water body), also includes the green wavelength (532 nm), utilizing its ability to penetrate water [35,36]. ALS LiDAR can be applied for water monitoring in post-mining areas. Studies of water resources mainly focus on determining their volume [37]. Notably, Jawecki et al. [37] found LiDAR data useful for detecting and estimating the retained water in a former granite quarry. In [21], the bottom relief of anthropogenic reservoirs in a coal basin region was studied to estimate the water resources. Kerfoot et al. [38] used LiDAR to map and assess the impact of copper waste leaching into the coastal environment. In [39,40], laser scanning was reported as beneficial in the estimation of the volume of waste stored in tailing ponds, and LiDAR technology was also described as a notable tool for managing environmental recovery.

While UAV-based LiDAR and optical imagery data offer advantages such as very good spatial resolution, and for LiDAR measurements, the ability to penetrate water and vegetation, as well as versatile applications [41], their disadvantages must also be noted. The data have low availability (depending on the scheduled measurement) and acquisition frequency (data is not acquired systematically, which restricts multitemporal analyses). Furthermore, a single measurement campaign covers a relatively small area (compared to satellite data) and requires the surveyor to be present in the vicinity of the surveyed area.

2.2. Radar—SAR Satellites

Synthetic aperture radar (SAR) data are acquired using active sensors. The emitted radiation in the microwave spectrum provides information that is independent of weather and illumination conditions. However, the processing of SAR imagery is more complex compared to MS data. Various processing products can be used for the purpose of surface water studies. The principle behind the monitoring of water using SAR technology is analyzing the backscatter signal [42]. Smooth water features act as specular reflectors, whereas rough water surfaces cause diffuse scattering [43]. The identification of water pixels is conducted using various thresholding algorithms [43,44,45,46]. In [44], the Otsu algorithm was used to separate water and non-water pixels based on the backscatter intensity in the vertical-horizontal (VH) polarization, since different polarizations have lower backscatter in flooded vegetated areas. On the other hand, in [47], the vertical-vertical (VV) polarization was found to be more suitable for water detection. In [43], an automated thresholding algorithm was compared with the spectral water index, the Normalized Difference Water Index (NDWI) and the Otsu algorithm. In [48], researchers discussed the C-band backscatter in relation to the vegetation type and water depth in flooded depression cones. Conversely, X-band sensors provide data of higher quality, but with limited penetration, as the signal scatterers from vegetation canopies, which limits the application of X-band sensors in studying water underneath vegetation [49].

SAR imagery is typically combined with spectral data [42,47]. In [50], Sentinel-2 imagery was combined with Sentinel-1 single-look complex (SLC) data to extract water in the Yangtze River Basin, China. A false image composite was created using Sentinel-1 dual-polarized water index (SDWI), Automated Water Extraction Index no shadow (AWEInsh), and Automated Water Extraction Index shadow (AWEIsh) as RGB. Water features were extracted using object segmentation. The SDWI, as a tool to extract water bodies using Sentinel-1 data, was designed to enhance water features, while minimizing the effects of soil and vegetation. In [51], the backscattering coefficient was used alongside a set of eight spectral indices, Sentinel-2 bands, and in situ data to estimate the variation in soil moisture caused by the subsidence and disturbance of underground water table in a coal mining region. In [43], a time series of Radarsat-2 was compiled with precipitation data. The Small Baseline Subset (SBAS) Interferometric SAR (InSAR) was used in [52] to extract surface settlements, data which were then used to form the Modified Remote Sensing Ecological Index (MRSEI) for the environmental assessment of a mining region in China. Recent contributions have utilized machine learning and deep learning algorithms, such as deep neural networks [53], for the detection of surface water in SAR images. A detailed review of the development of algorithms for water segmentation using SAR data can be found in [54].

2.3. Multispectral and Hyperspectral Satellite Data

While LiDAR and SAR are active RS methods that rely on emitted electromagnetic signals, multispectral and hyperspectral (HS) imaging are passive techniques that detect reflected sunlight. Satellite-based MS and HS sensors differ mainly in spectral resolution, i.e., in the quantity and width of spectral bands, but also in spatial and temporal resolution, data availability, and processing complexity.

A sensor can be considered HS if it includes at least 10 bands with a bandwidth range of 1–15 nm [55]. However, most HS satellite systems have sensors with a few hundred narrow bands. Spatial resolution varies depending on the system, e.g., the GSD for the EO-1, EnMAP, and PRISMA satellites is 30 m [56]. The HS data is not easily available, as some satellite systems do not have a strict orbital cycle, so the data are only available on demand. Moreover, acquisitions can only cover specific, limited areas [57]. The quantity and narrowness of the spectral bands enable continuous acquisition of spectral characteristics and more detailed analyses. Hyperspectral analysis was used for the RS of water in mining or post-mining areas, e.g., to map the abundance of polluted mine water in surface water [58], to quickly map acid water ponds at both high and low spatial resolution [59], to monitor river pollution from acid water in sulfide mines [60], and to examine acid mine drainage [61,62,63]. Studies using HS data processing are time- and cost-intensive, as they require advanced analysis, specialist knowledge and software, and significant computing power [53]. Given the demanding nature of the data and processing requirements, the choice of hyperspectral RS should depend on the level of spectral detail needed. If the main aim is to differentiate between land cover types or to detect changes, rather than conducting complex water quality analyses, MS data are more suitable due to their faster processing times, cost-effectiveness, spatial and temporal resolution, and data availability [64,65].

MS sensors usually acquire data in 4–15 spectral bands [66], with a broader bandwidth range. The most commonly used Earth observation (EO) missions, i.e., Sentinel-2 and Landsat, provide free-of-charge recent and archival data registered every 5–10 days (Sentinel-2) or 8–16 days (Landsat), with a spatial resolution of 10–60 m (Sentinel-2) or 15–120 m (Landsat), depending on the spectral band [67,68]. Higher-resolution data (from satellite systems such as RapidEye, PlanetScope, and WorldView) are also available, but only upon request for purchase or for scientific purposes at no cost [69,70,71]. Therefore, the number of studies of monitoring surface water in mining based solely on high-resolution commercial data is sparse. Nevertheless, such data are used in the study of surface water bodies outside mining environments as support to Sentinel imagery, in order to improve the quality and accuracy of water detection [46,72]. Due to the parameters, frequency, and consistency in data acquisition, MS data are prominently used in studies of mining grounds, often through spectral indices, which are the result of mathematical equations involving reflectance in different spectral bands, created to highlight the specific properties of a surface. Water indices most often utilize the contrast between the absorption in the NIR or mid-infrared range and reflectivity in the visible spectrum [73]. In optical RS, the blue (0.45–0.5 µm), green (0.5–0.57 µm), red (0.61–0.7 µm), NIR (0.7–1.3 µm), and Short-Wave Infrared (SWIR) (1.5–3 µm) bands have been widely used in water-related applications [74].

When applying MS-derived indices, it is crucial to consider factors such as atmospheric conditions, sensor characteristics, and land cover types. For instance, the effectiveness of spectral water indices depends on various environmental factors, e.g., water turbidity, vegetation cover, type of bedding or the presence of artificial built-up features. For example, the NDWI is used for general water detection, while the Modified Normalized Difference Water Index (MNDWI) reduces built-up land noise [75]. Many water indices have been developed, but not all serve the same purpose. Beyond the detection of water bodies [73,76], MS water indices are also used for the assessment of the water content in vegetation [77], or the analysis of soil moisture [78]. While some indices were not originally designed for water-related studies, they can be used for this purpose (e.g., the Normalized Difference Vegetation Index (NDVI), which is focused on vegetation; however, its negative values indicate water). A list of selected reviewed spectral indices, along with their formulae, benefits, and limitations, is provided in

Table A1. Summary of selected spectral indices used in studies of surface water dynamics.

Index and Formula	Benefits	Limitations
Normalized Difference Water Index $NDWI= {\displaystyle \frac{Green-NIR}{Green+NIR}}$ [73]	Easy to calculate [73] Requires only green and NIR bands, available in most multispectral data [73] Effective elimination of soil and vegetation backgrounds [123,124]	Confuses buildings with bare soil [123] Water features often mixed with built-up areas [115] Low sensitivity to small and narrow water bodies [124,125,126] Poor detection of turbid water [93]
Modified Normalized Difference Water Index $MNDWI= {\displaystyle \frac{Green-SWIR}{Green+SWIR}}$ [115]	Improved differentiation between water and shadow [127] Effectively reduces or removes built-up land noise [115] Accurate for small water bodies detection [123] Good enhancement of open water features [93]	Low boundary precision [125] Fails to exclude snow cover [128] Poor performance in distinguishing mountain shadows from water [124]
Water Index ${WI}_{2015}= 1.7204+{171}_{Green}+3_{Red}-{70}_{NIR}-{45}_{SWIR1}-{71}_{SWIR2}$ [129]	Relatively good overall performance [93] Well-suited for identifying rivers in flat areas [126] Developed for Landsat data, but also effective for Sentinel data [126,130]	Performance declines in urban and mountainous areas [126] Tends to overestimate hydrothermal water [126] Complicated calculation [129]
Enhanced Water Index $EWI= {\displaystyle \frac{MNDWI-\left(PC1+PC2\right)}{NDWI+\left(PC1+PC2\right)}}$ Where PC1—Principal Component no 1 PC2—Principal Component no 2 [131]	Effective for water extraction in mountainous urban areas [131] Minimizes background noise and cloud influence [132] Effective monitoring of surface water across seasons [132]	Requires manual thresholding for each case study site [131] Designed for sub-pixel surface water proportion mapping in inland rivers [133]
Water Ratio Index $WRI= {\displaystyle \frac{Green+Red}{NIR+SWIR}}$ [134]	Easy to calculate [135] Suitable for areas with dense vegetation and widespread shadows [136] Can indicate water bodies, as well as soil and vegetation water content [135]	Serves to examine soil or vegetation water content [135] Can lead to false positives, detecting features such as shadows, wet soil, and vegetation as water [137] Misclassifies urban areas, artificial surfaces, mountains, snow, shadows, colored lakes, and high chlorophyll content [136]
Augmented Normalized Difference Water Index $ANDWI= {\displaystyle \frac{Blue+Green+Red-NIR-{SWIR}_1-{SWIR}_2}{Blue+Green+Red+NIR+{SWIR}_1+{SWIR}_2}}$ [93]	Reduces errors from anthropological land cover and natural events [93] Good performance for muddy and hydrothermal waters [93] Successfully distinguishes dark vegetation from water [93]	Threshold adjustment required for discriminating dark vegetation inside water [93] Inaccurate for small water body detection [125] Low boundary precision [125] Bright surfaces (e.g., salt, synthetic materials) may cause omission of brighter water pixels [138]
Automated Water Extraction Index (shadow and no shadow) ${AWEI}_{sh}= Blue+2.5 \ast Green-1.5 \ast \left(NIR+{SWIR}_1\right)-0.25 \ast {SWIR}_2$ ${AWEI}_{nsh}= 4 \ast \left(Green-{SWIR}_1\right)-(0.25 \ast NIR+2.75 \ast {SWIR}_2)$ [76]	Stable and optimal threshold value [76] Good at water classification in mountainous and urban areas—effectively removes errors caused by the terrain and rooftop shadows [76,124] Good shadow and water discrimination [76] Designed for Landsat data, but also applicable to Sentinel-2 imagery [126]	Misclassifies dark and bright rooftops, dark vegetation and low-albedo objects as water [93,136] Unable to detect small and narrow waterbodies (AWEInsh) [126,136] Misclassifies snow as water [128,136] Common misclassifications in turbid water [93,136]
Sentinel-2 Water Index $SWI= {\displaystyle \frac{{Vegetation Red Edge}_1-{SWIR}_2}{{Vegetation Red Edge}_1+{SWIR}_2}}$ [124]	Effectively detects different types of water bodies [124] Accurate extraction of large water bodies and wide river channels in urban areas [124]	Inaccurate for small water bodies and narrow rivers (low spatial resolution (20 m)) [124] Prone to misclassification near vegetated and urban areas [126]
Simple Water Index $SWI= {\displaystyle \frac{1}{\sqrt{Blue-{SWIR}_1}}}$ [139]	Can distinguish water bodies from shadows and urban areas [139] Good for detecting both small and large water bodies [139] Applicable in various climate zones [139]	Less suitable for water bodies with dynamic boundaries [140] Developed for dry/semi-dry African regions [141] Rarely applied beyond the original study [141]
Normalized Difference Moisture Index $NDMI= {\displaystyle \frac{NIR-SWIR}{NIR+SWIR}}$ [77]	Estimates soil moisture—can be a proxy for wetlands [142]	Primarily designed to determine vegetation water content [82]
Normalized Difference Vegetation Index $NDVI= {\displaystyle \frac{NIR-RED}{NIR+RED}}$ [114]	Can distinguish water (values below 0) [143] Good for seasonal wetland identification [144] Can work as proxy for flooded vegetation [99]	Designed for vegetation monitoring [145] Results impacted by the surrounding vegetation [144] Will show poor vegetation instead of water on flooded vegetation [144] Value thresholds dependent on the season and land cover [144]

in Appendix A.

The use of MS satellite data to detect the impact of mining activities on both land and water has been found useful in the long-term monitoring of rehabilitated post-mining areas [27]. Environmental assessments in mining regions have been conducted using various vegetation and water indices, including the NDVI, NDWI, and the Normalized Differential Built-Up and Bare Soil Index (NDBBI) [79]. Land cover classification based on NDVI values has also been used to detect changes, such as reductions in surface water bodies and vegetation degradation, which are often associated with deteriorating water quality and habitat loss [80]. In [81], Google Earth Engine (GEE) algorithms were applied to Landsat satellite imagery to develop a method for detecting and mapping dynamic changes in surface water and land areas in a reclaimed coalfield area. A similar approach was used in [22], where Landsat imagery and the LandTrendr algorithm in GEE were used to analyze subsidence water in mining areas. The authors selected specific thresholds for the calculated indices to detect water. In [82], the flooding of subsidence basins over underground mines was observed using a combination of in situ data, RS imagery, the Knothe time function, and the principles of water balance. The MNDWI was used in [83] to discriminate surface water features from different land cover forms, while [50] used segmentation algorithms to extract surface water data based on satellite spectral and radar data. In [84], various spectral and topographic water indices were applied to remotely sensed data in order to develop the hydrological profile of an open-pit lignite mining area. The profile included the hydrological characteristics of water bodies, streams, vegetated and high moisture content areas. Moreover, the authors developed maps of flood-prone areas to support their claim of the usefulness of RS data in mine closure planning with regard to water management and flood prevention. Mine-related flooding was also studied in [85]. Using RapidEye data, the authors identified flooded areas and discriminated them from artificial water reservoirs based on the different spectral response and low albedo characteristic for water bodies. Floods and droughts, being opposite phenomena, can provide information on water distribution and hydrogeological characteristics of an area. In [7], hydrological drought caused by underground mine dewatering and the disappearance of rivers due to the discontinuation of mine drainage was analyzed. The authors applied a set of spectral indices, including the NDVI, Normalized Difference Moisture Index (NDMI), the NDWI, the Moisture Stress Index (MSI), and the Normalized Multiband Drought Index (NMDI) derived from Sentinel-2 imagery. A negative water balance in mining regions can lead to water stress in plants and changes in soil moisture. While the assessment of vegetation and soil is beyond the scope of this paper, it is worth noting that vegetation condition and soil moisture indices are used in studies on surface water. Another hazardous impact of mining that can be monitored with satellite spectral data is tailing ponds. The study described in [86] used Landsat 8 imagery for the detection of tailing ponds. Spectral and textural characteristics extracted from MS data allowed discriminating the tailings from iron ore stopes. Both Landsat-8/9 and Sentinel-2 data were used in [87], where applied machine learning (ML) methods and convolutional neural networks (CNN) were used for the bathymetry estimation of tailing storage facilities. Failures of tailing ponds dams, although rare, are also important, due to their impact on the surrounding environment, and are repeatedly investigated. Breaches in tailing pond dams can have a detrimental effect on the environment, especially on water quality. The study of the temporal behavior of wet tailings and supernatant water highlighted the necessity of monitoring water dynamics in tailing ponds to enable more effective near-real time management and provide an early warning system [10]. The research involved applying spectral indices such as the MNDWI, the Enhanced Vegetation Index (EVI), and the NDVI to Sentinel-2 imagery, as well as using the GIS and GEE. The authors confirmed that the use of RS technologies helps in making conscious and responsible decisions in tailings management. An algorithm using the Red and NIR bands of Landsat and MODIS-Aqua imagery was employed to study the impact of tailing dam failure on rivers and coastal waters. RS data made it possible to create a turbidity map, and to distinguish between natural turbidity and turbidity caused by dam failure [88]. Crioni et al. [89] examined increased turbidity after a dam failure, utilizing Sentinel-2 imagery as the basis for an empirical single-band model to estimate water turbidity and to perform a quantitative analysis of seasonal variations. In [90], the water quality in a mining area was assessed by comparing data from various satellite spectral sensors with water samples. Water quality is also often affected by AMD, making it a frequently studied phenomenon. The possible use of Sentinel-2 imagery and artificial neural networks for mapping AMD in water was investigated in [83], whereas [91] applied the Acid Mine Water Index (AMWI) to Sentinel-2 imagery to monitor surface water contamination and estimate its inter-annual variations. Bijeesh and Narasimhamurthy [92] reviewed the established RS methods for surface water studies and grouped them into four classes: single band-based methods, spectral index-based methods, machine learning-based methods and spectral unmixing-based methods. They noted the abundance of RS water survey methods and pointed out that they were developed based on specific cases. They found that spectral indices are the most often used method (35% of all methods), due to their ease of use and understanding. In [93], the authors agreed that, due to these qualities, as well as the low computational cost and versatility between platforms, spectral water indices are a favored tool for analyzing surface water.

2.4. Summary

Remote sensing offers multiple complementary approaches for detecting surface water and monitoring its quality. Considering the various advantages and disadvantages, the choice of the data source and RS methods for monitoring water in mining and post-mining areas depends on the purpose of the study. Although airborne LiDAR and optical imagery offer very high spatial resolutions, their limited temporal availability restricts multitemporal studies. SAR data enables analysis regardless of atmospheric conditions, but its interpretation can be challenging. HS satellite data offer detailed spectral information that is beneficial for water quality assessment. However, the processing is complex, and the data have medium spatial resolution and limited accessibility, which makes water identification difficult. MS satellite systems strike a balance between the aforementioned sources by providing sufficient spectral and spatial resolution imagery, with systematic and high temporal coverage, and simple data processing. A comparison and summary of the selected RD data sources used in water detection is provided in

Table A2. Comparison of various remote sensing data sources used in water detection.

	Light Detection and Ranging and Optic Imagery	Multispectral	Hyperspectral	Synthetic Aperture Radar
Product, data geometry	Point cloud, 3D (x, y, z)	Raster, 2D (x, y)	Raster, 2D (x, y)	Point/raster, 2D (x, y)
Spatial resolution, precision	High Centimeter level	Moderate/high Depending on the sensor, the GSD varies from submeter level to tens of meters	Moderate/high Depending on the sensor, the GSD varies from submeter level to tens of meters	Depending on the sensor, the GSD varies from submeter level to tens of meters
Temporal coverage, Acquisition frequency	Varying, depends on scheduled measurement campaign	Satellite imagery: high (every few days) Airborne: Varying, depends on scheduled measurement campaign	Satellite imagery: high (every few days) Airborne: Varying, depends on scheduled measurement campaign	Satellite: high (every few days)
Data availability	Not global or regularly updated, on demand, limited (shared by regional portals)	Satellite: Globally available recent and archival data Airborne: depends on scheduled measurement campaign	Satellite: globally available recent and archival data Airborne: depends on scheduled measurement campaign	Satellite: Globally available recent and archival data
Cost	Depends on custom acquisition Open access in some countries (e.g., Poland)	Satellite: commercial or open access (depending on the satellite mission) Airborne: depends on custom acquisition	Satellite: commercial or open access (depending on the satellite mission) Airborne: depends on custom acquisition	Paid access or available free of charge (from particular satellites)
Processing requirements	Software capable of processing point cloud data	Software or cloud platforms capable of processing imagery data	Software or cloud platforms capable of processing imagery data	Software capable of processing imagery data
Field of use, usability in surface water monitoring	Surface water extent	Surface water detection, extent, quantitative analyses, flood mapping, water indices, multitemporal analysis	Surface water detection, extent, quantitative analyses, flood mapping, water indices, multitemporal analysis, complex qualitative analysis	Surface water detection, water extent time series analysis, flood mapping, wetland mapping, reservoir extent

, Appendix A.

Given the above, we recognize the importance of analyzing the currently available spectral indices, testing them in different research areas and comparing their performance in order to identify their respective advantages and disadvantages and enable the correct, informed selection of an index for further surface water analyses.

3. Application in Mining and Post-Mining Case Studies

Based on the literature review, we have selected three spectral indices, namely the NDVI, NDWI, and MNDWI, to analyze their performance and applicability in monitoring the dynamics of the surface water area in mining areas. Changes in water dynamics are an important aspect of restoring balance in post-mining areas; therefore, the study focuses on water extent changes, and water quality aspects are outside of the scope. Post-mining areas vary significantly, as their characteristics depend on a number of factors, including the natural pre-mining conditions, the ore type and extraction method, the duration of operation and the manner of termination of the mining activity, as well as the time and means of reclamation. This study was conducted at three different mining and post-mining sites in Poland. These sites were selected to test the indices in areas of diverse land cover characteristics. Another aim of this study was to assess the spatiotemporal dynamics of surface water resulting from waterlogging and flooding in the specified areas.

With this purpose, we propose an approach based on open MS data to monitor surface water dynamics in mining environments. The location of the case study sites is presented in Figure 2, and their basic characteristics are given in

Table 1. General characteristics of the study areas (based on [94,95,96,97,98]).

	Hutki	Babina	Kosakowo
Area [km2]	3.5	2.2	3.9—mining concession area 26.2—study area
Status of the area	Post-mining (closed 5 years ago)	Post-mining (closed 50 years ago)	Active
Type of mining	Underground and open pit	Underground and open pit	Underground gas storage (UGS)
Ore and deposit geology	Underground zinc and lead ore, open pit sand mines (70–190 m below ground level/b.g.l.)	Underground brown coal (5–100 m b.g.l.)	Mechelinki salt deposit Depth 970 m b.g.l. (top) Deposit thickness 170–200 m
Mining system	Room and pillar	System of shallow underground workings	Leaching (development of storage caverns)
Topography and land cover type	Gently undulating terrain, mostly forested. Mean elevation of 312.05 m above sea level (a.s.l.)	Mostly forest, surface waters, and hiking trails. Anthropogenic lakes (elevation 133 m–155 m a.s.l.), mining-related ground deformations, and other artificial landforms.	Plain; mean elevation of 11.4 m a.s.l.; upland reaching over 70 m a.s.l. south to the UGS permit area; rural areas with agricultural land.

.

3.1. Olkusz—Hutki—Region Impacted by Underground Mine Closure

The zinc and lead ore mines in the Olkusz region required extensive dewatering due to the inflow of groundwater from within the Triassic aquifers into the underground excavations (70–190 m below ground level/b.g.l.) [99]. Prolonged dewatering of the mine and the discharge of water into nearby rivers caused hydrological drought in the area [7]. The cone of depression was estimated to cover 700 km² [100]. The last mine in the region was closed on 31 December 2020, with the water pumps being shut down between December 2021 and January 2022. The completion of mine dewatering initiated the groundwater table recovery, leading to the flooding of the underground workings. It is expected that the original, pre-mining hydrodynamic state will be restored, causing flooding or waterlogging of the basins from former open-pit sand mines and subsidence above underground workings [101]. This process started sooner than predicted, with surface water emerging in several parts of the Bolesław municipality. Therefore, the area of interest (Figure 2) was selected for tracking this process over time, using three different spectral indices. The selected Area Of Interest (AOI) covers an area of 3.5 km², extending between longitudes 19°29′51.91″ E and 19°31′33.65” E and latitudes 50°18′54.97″ N and 50°17′57.85″. It is located in southern Poland, in the Olkusz district and the Bolesław municipality, near the Hutki village. The selected AOI covers the inactive open-pit backfilling sand mine “Hutki II” and historical underground zinc mines.

The area was chosen as a case study of RS of surface water in post-mining areas due to the rapid emergence of water on the surface resulting from the termination of mining activity. Most of the AOI’s land cover consists of coniferous tree canopies [102]. The presence of tall, evergreen green trees poses a challenge in detecting waterlogging with satellite RS techniques.

3.2. Babina Post-Mining Area

The second case study area is located in western Poland, in the south-western part of a glaciotectonic structure known as the Muskau Arch. This terminal moraine was formed during the South Polish glaciation and subsequently remodeled during successive glacial periods [103,104]. These processes have left the area rich in shallow deposits of sands, gravels, kaolin clays, silts, loams, and predominantly brown coal seams [105,106,107]. Brown coal mining began on a small scale in the mid-19th century, and on an industrial scale in the 1920s [108]. The shallow lying (approx. up to 100 m b.g.l.) and inclined coal seams were initially mined using the underground method, but over time, open-pit mining became dominant. Intensive underground and open-pit mining activities have caused severe anthropogenic changes in the region, including the development of elongated subsidence basins over the underground mining fields, as well as open pits and waste heaps. Due to economic reasons, mining activities ceased in 1973 [95]. The selected area of interest (Figure 2) is located in the “Pustków” field of the former “Przyjaźń Narodów—Szyb Babina” (also known as the “Babina”) mine and approximately covers 2.2 km². With the completion of mine dewatering, the subsided and excavated areas were waterlogged and subsequently flooded. There are 10 water bodies in the analyzed underground part of the post-mining area. The lakes vary in shape and size. According to the Polish National Database of Topographical Objects, the smallest lake covers 1079 m² and the largest one covers 61,312 m². They formed in large subsidence basins that developed over old underground mining fields where several overlying brown coal deposits were mined with roof support [109]. The narrow, irregularly elongated shape of the post-mining lakes is a consequence of the shallow underground mining of brown coal deposits.

The post-mining lakes were suspected to undergo water level changes, as well as processes associated with vegetation change in the shoreline zones. However, no previous research in this domain had been conducted there. Thus, this area was selected as the AOI for the detection and analysis of surface water area changes.

3.3. Kosakowo—Underground Gas Storage Site

Underground gas storage (UGS), as a specific form of mining activity, also interacts with the environment in a number of ways. Storage facilities are developed in favorable geological conditions, enabling the safe and leak-proof storage of hydrocarbons, compressed air, and, increasingly, hydrogen [110]. They are also used for the long-term storage of carbon dioxide (CO₂), in a process known as carbon capture and storage (CCS). The most common types of UGS include depleted oil and gas reservoirs, aquifers and porous media, rock caverns, and salt caverns in particular. The Kosakowo UGS facility is located in northern Poland at 18°27′17″ E and 54°36′23″ N (Figure 2). The facility is located close to the city of Gdynia and the Baltic Sea. The average elevation of the area is approximately 3 m above sea level (a.s.l.), with upland areas reaching over 70 m a.s.l. in the south. The area to the south of the facility is mainly built up, with industrial, urban and residential zones. To the north of the UGS, the area is covered with pastures and meadows. Due to the geology, which includes shallow peat, river and marine accumulations [111], the area is prone to waterlogging. Additionally, the Polish national hydrogeological services have identified this region as being at risk of groundwater flooding. The UGS facility comprises two clusters (clusters A and B) of five caverns each. The total capacity is 293.4 million m³. The leaching of the caverns began in 2009. Cluster A was commissioned in 2014, and cluster B in 2021. The deposit dates to the Permian period. The caverns were leached within the Mechelinki salt deposit, which lies at a depth of 970 m below the surface (top), with an average thickness of 170–200 m [112]. Due to the plastic properties of rock salt, the surface impacts of salt mining and UGS in salt occur over a long period of time.

As the UGS facility is relatively young, the observed impacts are yet minor as they began to occur only recently. It is, therefore, crucial to observe the facility before and at the beginning stage of operation, in order to establish a baseline for future monitoring.

3.4. Materials and Methods

A unified methodology was adopted for all the case studies to ensure consistent and comparable conditions for evaluating the performance of the indices. Due to the unique nature of the changes in each case study, the dates of the satellite images used were selected individually, while the processing of the selected indices and spatial statistics was performed uniformly. Figure 3 shows the schematic workflow.

3.4.1. Input—Multispectral Satellite Imagery

Taking into account the characteristics of the data sources compiled in Section 2, the purpose of this study (water detection with no detailed analysis of water quality), and the multi-temporal nature of the study, the Sentinel-2 mission was chosen as the data source. The data were downloaded from the Copernicus Data Space Ecosystem [113] Collection 1 as a Level-2A product, which is orthorectified and atmospherically corrected; therefore, it does not require any further preprocessing.

Table A3. Dates of satellite data used in selected case studies.

Case Study	Dates of Used Satellite Data			Comment
Hutki	2022/04/14 2022/10/06 2023/04/22 2023/10/14 2024/03/29 2024/09/18 2025/04/16			The timeframe of the study was chosen with regard to the season and the date of the first appearance of water on the surface in the selected area [101]. The first acquired image shows the study area before the appearance of water.
Babina	2015/08/10 2018/08/07 2021/08/13 2024/08/07			Water bodies in the region formed in subsidence basins over 50 years ago. The dates were chosen to monitor changes in water masses in 3-year periods for a period of approximately 10 years.
Kosakowo	2015/08/20 2015/09/19 2016/05/06 2016/06/05 2016/06/15 2016/06/25 2016/09/13 2017/05/01 2017/05/06 2017/05/11 2017/05/16 2017/05/21 2017/05/26 2017/07/30 2017/08/09 2017/09/28 2018/05/06 2018/05/16 2018/05/21 2018/05/31 2018/06/05 2018/06/10 2018/06/20 2018/07/20 2018/09/18	2019/06/05 2019/06/20 2019/06/30 2019/07/15 2019/07/20 2019/08/24 2019/08/29 2019/09/28 2020/06/04 2020/06/09 2020/06/14 2020/06/19 2020/07/19 2020/08/13 2020/08/18 2020/09/12 2021/05/10 2021/05/30 2021/06/09 2021/06/14 2021/06/29 2021/07/14 2021/09/27 2022/06/04 2022/06/24	2022/07/19 2022/07/29 2022/08/03 2022/08/13 2022/08/28 2022/09/07 2023/05/05 2023/05/10 2023/05/30 2023/06/09 2023/06/29 2023/09/02 2023/09/07 2023/09/12 2023/09/17 2023/09/22 2023/09/27 2024/05/09 2024/05/14 2024/05/19 2024/08/07 2024/09/06 2024/09/16 2024/09/21	The collection includes imagery acquired since the launch of the Sentinel-2 mission. Images with cloud coverage of less than 30%, covering the vegetation period from May to September, were filtered. Images without clouds or shadows above the area of interest were selected.

in Appendix B presents the dates of the satellite images used for each case study.

The images were clipped to the extent of the respective AOIs. Additionally, in the case of the Kosakowo UGS, a mask was applied to exclude existing surface water reservoirs and built-up features from the analysis. The mask was developed based on the national database of topographic objects (BDOT10k).

3.4.2. Preprocessing and Calculating Spectral Indices

Considering the research objectives and the characteristics of the indices described in the literature review section, as well as the characteristics of the selected study areas, the Normalized Difference Vegetation Index [114], Normalized Difference Water Index [73] and Modified Normalized Difference Water Index [115] were selected. The indices were calculated using the formulae presented in Table A1 (Appendix A).

3.4.3. Surface Water Detection and Statistical Analysis

To distinguish between surface water and other land cover types, threshold values for the selected indices were defined based on the literature, empirical validation (tested on a set of randomly selected pixels known to be water), and accounting for the characteristics of the AOIs. The empirical validation involved creating histograms of the index values to highlight the natural thresholds in each study region. While the literature states that water in the MNDWI is represented by values above 0 [115,116], the threshold value is often adjusted according to the examined area [75]. In [83], a threshold of the MNDWI greater than 0.4 was selected to delineate surface water bodies in mining environments. On the other hand, in [52], a threshold of 0.25 for the MNDWI was applied. The adopted thresholds for each case study (compiled in

Table 2. Summary of the indices threshold values applied in study sites.

Index	Pixel Classification	Case Study Area
		Hutki	Babina	Kosakowo
Normalized Difference Vegetation Index (NDVI)	Water	[−1.0; 0.0)	[−1.0; 0.09)	[−1.0; 0.2)
	Non-water	[0.0; 1.0]	[0.09; 1.0]	[0.2; 1.0]
Normalized Difference Water Index (NDWI)	Water	[0.0; 1.0]	[0.0; 1.0]	[−0.25; 1.0]
	Non-water	[−1.0; 0.0)	[−1.0; 0.0)	[−1.0; −0.25)
Modified Normalized Difference Water Index (MNDWI)	Water	[0.0; 1.0]	[0.0; 1.0]	[−0.1; 1.0]
	Non-water	[−1.0; 0.0)	[−1.0; 0.0)	[−1.0; −0.1)

) were used for the binary reclassification of the images, with water pixels represented by 1, and non-water features by 0.

The binary rasters were subjected to statistical analysis to determine the area of surface water in the image series taken at consecutive dates. The area occupied by water was calculated by multiplying the pixel count by the GSD squared.

The next step included the analysis of dynamics, referring to either the rates at which the water appears or disappears on the surface, or to the temporal behavior of reoccurring water clusters. In the case of the Hutki area, the development of lakes was observed, and thus the dynamics were presented by calculating the surface water area on a given date. Due to the longer history of the Babina area, the changes are not expected to be related to groundwater level restoration. In the Kosakowo area, the dynamics refer to the analysis of the temporal behavior and changes in waterlogged areas, as shown by the time series analysis. The performance of the selected indices was then compared in different types of mining environments.

4. Results

The performance of the indices, as well as the surface water dynamics, was assessed qualitatively and quantitatively using visual, statistical and descriptive analyses. For each case study, the results derived from the applied indices are displayed over a Color Infrared (CIR) background to facilitate visual interpretation. The CIR composition is used to improve the visual perception of water and vegetation, making the land cover types more distinguishable (water is represented by dark pixels, vegetation in shades of red) [89].

4.1. Hutki

The indices enabled the retrospective assessment of surface water dynamics in the Hutki area. Figure 4 shows the maps with surface water detected by the selected indices in the 2022–2025 period. A comparative analysis shows varying levels of correspondence among the spectral water indices. Before October 2023, no surface water was detected. In October 2023, all indices detected a small area of surface water (southern part of the AOI), with the NDVI identifying more pixels than the NDWI and the MNDWI. The next image (taken on 29 March 2024) showed increased water presence, and all indices performed similarly, except for a narrow water body in the north. The NDVI detected its wider upper part, while the NDWI and the MNDWI missed it entirely. As the water area expanded in September 2024 and April 2025, the performance of the NDVI deteriorated. The MNDWI generally detected water more accurately than the NDWI, but it missed a small round water body near the western AOI border and misclassified urban areas in the south-west as water. None of the indices detected the water at the edges of the floodplains.

The dynamics of surface water change are presented in Figure 5. The surface water area values were calculated for each measurement date, based on selected indices. The quantitative analysis shows that the NDWI and the MNDWI detected similar surface water area values, while the NDVI consistently showed a smaller area. The largest discrepancy occurred in September 2024, with a difference of 225,700 m² between the MNDWI and the NDVI. In the spring of 2025, the values for the NDWI and the MNDWI remained consistent, but those for the NDVI continued to diverge. No significant changes were observed during the first two years following the pump shutdown. The low values recorded prior to October 2023 may reflect index errors or the presence of isolated water patches. After October 2023, however, the surface water dynamics increased noticeably. By April 2025, the total area of detected surface water within the AOI exceeded 600,000 m², though the water was primarily concentrated in three distinct, lake-like reservoirs.

Figure 6a presents the extent of the lakes with their boundaries approximated from the surface water extent in April 2025, whereas Figure 6b–d show graphs of surface water calculated based on the selected indices in each lake over time. The area of the subsequent water bodies varies—the largest water area was detected in lake 3 (more than 300,000 m²), while the smallest water area was found in the area of lake 1 (no more than 100,000 m²). The most rapid increase in the water surface area was observed for Lake 2. The performance of the selected indices in identifying the presence of water within the boundaries of the three reservoirs varied. For Lake 2, all indices performed similarly. The largest discrepancies between the NDVI and the water indices (the NDWI and the MNDWI) occurred for Lake 1, where the NDVI indicated the water surface area had decreased over time, and for Lake 3, where the NDVI detected significantly less surface water. The NDWI and MNDWI performed similarly, with the exception of water identification in September 2024 for Lake 3, when the NDWI identified significantly less water.

For the purpose of identifying the surface water in the Hutki AOI, the NDVI demonstrated the lowest accuracy among the evaluated indices. While both the NDWI and the MNDWI produced comparable and reliable results, the MNDWI was found to be the most accurate for the selected study area based on qualitative, quantitative, and visual analyses.

4.2. Babina

The general trend of the surface water area, determined using the three spectral indices in the analyzed period between 2015 and 2024, shows a gradual decrease in the surface water area (Figure 7). The area identified as water using the NDWI decreased from 175,900 m² in August 2015 to 130,800 m² in August 2024. The area identified as water with the MNDWI decreased from 128,700 m² to 117,100 m² over the same period. The difference in the surface water area identified with the NDWI and MNDWI ranges from 47,200 m² (2% of the total area) in 2015 to 13,700 m² (0.6% of the total area) in 2024. The trend exhibited by the NDVI also shows a general decrease in the area identified as water, with the lowest value identified in August 2021. This value is considered an outlier in this analysis.

Maps of surface water area on the four analyzed dates are presented in Figure 8a for the NDWI, Figure 8b for the MNDWI, and Figure 8c for the NDVI. The corresponding water surfaces of these lakes in the analyzed periods are shown in Figure 9. During the analyzed period, lakes 3, 8, and 9 experienced the greatest decrease in their surface water area, as indicated consistently by all the three indices (Figure 9). Lakes 6 and 7 registered the least or no change in the surface water area. In the case of the four smallest lakes, 1, 2, 4, and 10, the results showing a decrease in surface water area should be treated as indicative, due to the spatial resolution of the data and the potential for the spectral mixing of vegetation and water at the sub-pixel level. All three spectral indices (the NDVI, the NDWI and the MNDWI) revealed a consistent trend, although the actual surface area determined with their use differs due to the inherent properties of the respective indices.

Figure 10 shows the surface water area for the first (2015) and the last (2024) measurement of lake 8, one of the three water features that lost the most surface according to the spectral proxies for water. Despite the expected differences in performance between the spectral indices, they proved to serve as useful tools in tracing historical water level changes in the absence of in situ observations.

We identified the following potential causes of the decrease in surface water area at the post-mining study site. Firstly, the hydrological drought observed in Poland since 2015 had intensified as a result of lower or sparser, but more intense precipitation events, and higher evaporation rates due to the increasing air and surface temperatures [117,118,119]. Secondly, we observed the gradual encroachment of water vegetation in parts of the analyzed lakes, especially near shorelines [120]. Figure 11 presents a south-western part of lake 8 viewed toward the north-east in 2021 and 2024. The photos were taken during field investigations of land cover in the post-mining area, and confirm the decline in water level, as evidenced by the exposed shoreline and emergence of dead vegetation. To limit the potential influence of seasonality, we used images taken at the same time of the year.

4.3. Kosakowo

The analysis period spans the years 2015–2024 and covers the vegetation period with the aim to capture short-term variation in the indices’ values. The general trend of the spectral indices follows the seasonal changes (Figure 12). The NDVI values are low at the beginning of the vegetation season in May, increasing with the vegetation growth in June and July (Figure 12c). The indices’ values then drop at the start of the harvesting season (turn of July/August). The water indices, namely the NDWI and the MNDWI, exhibit inverse temporal patterns, as the vegetation takes negative values (Figure 12a,b). However, several peaks in the water indices were observed, which may be attributed to temporal variations in the occurrence of surface water.

The water pixels identified in each image were aggregated, and the resulting time series of the surface water area is showed in Figure 13. The sizes of the water area derived from the NDWI and the NDVI align well, following the same pattern and order of magnitude, with a slight advantage for the NDWI. However, the area derived using the MNDWI is notably smaller.

The spatial distribution of water pixels for each index is presented in Figure 14a–c. Water pixels were mainly observed in the agricultural land to the North and North-West direction of the facility. The water area follows the spatial pattern of the crop fields. Waterlogging is observed in the spring. The maps show how the indices change on dates when the most and least water pixels are detected. As with the previous maps, the MNDWI (Figure 14a) displayed different results to those created using the NDWI (Figure 14b) and NDVI (Figure 14c). In all cases, surface water was detected in the north-west part of the study area. However, in the case of the MNDWI, the area is smaller and covers crop fields. The background CIR composite image highlights water features that are not covered by water pixels. The NDVI- and NDWI-based maps showing the maximum coverage represent the same period in May 2017. Both maps show surface water in the agricultural region, but also in the eastern part. The area is located directly on the Baltic Sea shoreline, within the zone of risk of flooding from the groundwater. The main land cover types include wetlands and rushes.

A cumulative map of the detected surface water is given in Figure 15. Following the binary reclassification, a temporal aggregation of pixel value was carried out to evaluate the temporal coverage of water in the area. As can be seen from previous maps, the region to the north-east of the UGS permit area exhibits the highest temporal occurrence of water. Water pixels were identified for up to 60% of the analyzed acquisition dates. A different temporal behavior is shown on the map derived from the MNDWI, confirming the limited application of the index in areas experiencing waterlogging.

5. Discussion

A cross-case analysis revealed that changes in surface water over time vary depending on the characteristics of the area in question. Although the selected case studies share some similarities, they also have differences. Each of the selected case study areas is impacted by mining operations, which further affected the surface water conditions of the region. In the newly formed post-mining area of Hutki, surface water is steadily increasing due to the groundwater table restoration after the end of the mine dewatering. In Babina, surface water fluctuations and a general decrease in area have been observed for many years after the restoration of the groundwater table. Conversely, in the region of active mining involving underground gas storage in Kosakowo, seasonal waterlogging in the agricultural areas is noted. The different conditions of the areas studied, the behavior of the surface water, as well as the duration of the mining operations and the date of their closure determine the temporal resolution of the analysis. The AOIs also differ in land cover type and water surface characteristics, which can impair the results of water identification. While all the selected spectral water indices were found useful for surface water identification, the analysis showed varying effectiveness in the subsequent case studies.

In the Hutki area, the NDWI and the MNDWI showed coinciding values for the detected surface water area, and better performance than the NDVI. However, the NDVI worked better for small and narrow water bodies. A comparative visual analysis indicated that the MNDWI identified water more accurately than the NDWI. While the MNDWI and NDWI outperformed the NDVI in detecting surface water in large lakes, both indices still underestimated the extent of water in specific locations, where partially flooded coniferous trees were present. The tree canopy remaining above the water level was detected by the indices (both correctly and incorrectly) as non-water. This highlights the limitations of spectral indices in accurately delineating water bodies in particularly complex or mixed-pixel environments. None of the selected indices correctly identified water near the lake’s shoreline, which may be related to the depth and turbidity of the water, or to the pixel-level spectral mixture, which may also be associated with the spatial resolution of the image.

In the Babina site, all of the spectral indices showed a gradual decrease in the total surface water area and the same temporal trend. Although the values in the 2021 NDVI results followed the general decreasing trend in water mass in lakes, the surface of the water areas detected was only half that of other years. Overall, these results show that the NDVI is the most effective index for detecting lakes in the Babina region. The second-best water detection index was the NDWI. The MNDWI produced the least reliable results for water detection in the region. While the index results for large lakes with low plant density were similar, the indices showed different results for large lakes with increasing plant density. The functionality of the indices in the detection of small lakes is limited. However, the NDVI has produced the best results out of the three indices. In addition, the spectral resolution of satellite images is 10 × 10 m, which should also be considered as a factor that makes it difficult to detect small lakes. The study area is characterized by post-mining lakes of different sizes and shapes, including small and narrow water bodies. Thus, the effectiveness of water detection in our results corresponds to the findings, especially for small and narrow waterbodies in the Hutki site.

Although there are no permanent water reservoirs within the mining permit area of the Kosakowo UGS, the analysis revealed areas with periodical occurrence of surface water. The water appears as a result of waterlogging in the agricultural areas to the north and north-east. The results derived using the selected indices vary, but the general trend is similar. The NDWI and NDVI exhibit good alignment and consistency throughout the time and space domains. The MNDWI is significantly less sensitive to such forms of surface water, which was reflected in the poor detection of water and the low count of water pixels.

The proposed approach can be effectively applied to surface water dynamics analysis in all the selected case study areas. However, the accuracy of the results must be further critically addressed. The underestimation of the water extent near the lake shoreline and in the mixed-pixel regions of submerged vegetation in Hutki and Babina case studies, as well as the misclassification of pixels presenting bare soil as water in the Kosakowo AOI, indicate that the use of the NDVI, NDWI, and MNDWI on Sentinel-2 imagery can only provide an approximation of the temporal change and dynamics of the surface water area. The applicability of the selected indices is impacted by the conditions of the natural environment of the studied area. Satellite spectral index-based water detection is highly sensitive to adopted threshold values, which may vary by season, terrain, area, or vegetation cover, and which may require local calibration or more advanced approaches, such as machine learning-based classification, either supervised (with labeled samples) or unsupervised (clustering). Moreover, the use of Sentinel-2 imagery with 10 m spatial resolution contributes to the spectral mixing issue, which lowers the accuracy of water detection in small or narrow water bodies, especially those surrounded by vegetation or with flooded vegetation. Other studies have also noted this issue. Adjovu et al. [121] points out that sparse spatial resolution leads to inaccurate single-pixel spectral signatures. Similarly, Huang et al. [75] report that coarse resolution leads to high generalization of information and thus, it decreases accuracy, whereas Kathirvelu et al. [122] propose a spectral unmixing process leading to more accurate detection of surface water changes.

There is a need for further research with the use of high-resolution data to assess the accuracy of the estimation of the said surface water area, to inspect the performance of indices for narrow lakes or lake shorelines, and to develop a methodology for the automatic detection of water hidden under tree canopies or within crop fields. Despite machine learning methods being a promising direction [54,110], they require large datasets, which are unavailable for the analyzed cases. Nevertheless, the use of RS techniques on Sentinel-2 imagery is beneficial in multitemporal surface water analysis in mining-impacted environments, mainly due to the frequent revisit time, rich archives of open access data, and the availability of spectral bands that make it possible to calculate spectral water indices as proxies for surface water.

6. Conclusions

This study provides an overview and a comparative assessment of remote sensing techniques for surface water monitoring in mining areas. Additionally, the workflow for analyzing surface water change dynamics was proposed and applied in three different case studies to assess the effectiveness of selected spectral indices (the NDVI, NDWI and MNDWI) in detecting surface water across environments with different mining stages and land types.

The results indicate that the performance of the selected spectral indices for water detection and change dynamics analysis is highly dependent on other factors, such as land cover type, the nature of the water-related phenomena, the surface water extent and temporal trends. No single most accurate index for water studies in mining areas was selected, as different indices performed best in different case studies. The NDVI produced the best results for Babina, and the worst for Hutki, while in Kosakowo, the NDVI performed slightly less accurately than the NDWI. For the Hutki and Babina sites, the NDWI provided second-best results. The MNDWI was found to be the most accurate for large water bodies at the Hutki site, and the least accurate for Babina and Kosakowo.

Although the selected study areas differ from one another, the proposed workflow is applicable, and it enables the assessment of the surface water change dynamics in each study area. Nevertheless, given the variability in water detection accuracy among the selected indices, they should be considered only as proxies for surface water detection. Inaccurate results arise not only from the factor dependency of spectral indices, but also from the spatial resolution of the selected input data, which affects the spectral mixing issue. This highlights the need for an area-based data source, index and threshold selection, as well as multi-sensor or high-resolution data integration to improve accuracy in small and narrow water bodies detection. Our study advances the practical application of remote sensing for monitoring surface water in mining environments, offering a robust approach to studying water dynamics in both active mining and post-mining sites.