Longwall mining is a productive method for underground coal recovery, however leaving extensive and long-term dewatering impacts on the regional hydrology (Booth 1986, 2006; Luan et al. 2020; Sidle et al. 2000; Zhang et al. 2011) and surface features that are susceptible to groundwater drawdown (Hebblewhite 2009). In Australia, concerns over past and potential future impacts have been challenging longwall approvals for years due to environmental, social, and governance (ESG) considerations. Evaluating groundwater depressurisation and its further influence on the regional water system has been a component of Australian longwall practice.

Assessment of the groundwater impact involves a parameter often referred to as “height of fracturing” (HoF), which measures the upper limit of continuous fracturing with respect to the mining horizon and can be determined via Lugeon style packer testing coupled with borehole acoustic or optical scanning. A number of conceptual models derived from longwall practice and field measurements summarise the fracturing mechanics and groundwater hydraulics above a mined panel (Adhikary and Guo 2015; Adhikary et al. 2017; Coe and Stowe 1984; Galvin 2016; Guo et al. 2007; Hebblewhite 2020; Hebblewhite et al. 2008; Kendorski 1993, 2006; Kendorski et al. 1979; Minns et al. 1995; Seedsman 2019; Tammetta 2013, 2015, 2016). Empirical descriptions indicate a positive correlation between HoF and mining height, while the coefficient varies a lot, from 15 to 60, between mines. Rock units below the HoF horizon exhibit fracturing and caving into the mined-out area, with pronounced increases in porosity and hydraulic conductivity (Heritage et al. 2022a, b). Strata above the HoF horizon coherently sag and remain uncracked, acting as barriers separating upper aquifers from the drainage effect underneath (Booth 1999). The ground surface has temporary fractures in tensile areas lateral of the subsidence trough (Kendorski 1993, 2006), featuring water table decline due to vertical permeability enhancement (Minns et al. 1995; Tieman and Rauch 1986) and potential groundwater regime change in the long term (Seedsman 2019). These identifications have formed a fundamental understanding of longwall mining impacting the regional hydraulics.

To current knowledge, the groundwater depressurisation is already easy to understand, but still challenging to predict due to multiple influencing factors, complex fluid flow behaviour in fractured rock masses, and geological and hydrogeological variations between mines. In addition, field data are often unavailable to underpin an improved understanding of these problems. The groundwater impact prediction is no longer simply about investigating rock fractures in terms of its intensity and distribution above mined panels, but also about assessing the short- to long-term drainage effect of fracture networks on groundwater regime. In this context, this paper takes a step towards determining the factors and mechanism of mine dewatering, identifying knowledge gaps and methodological limitations, and establishing a research framework to facilitate achieving further understanding and assessment of groundwater depressurisation at mine sites.

A direct result of groundwater depressurisation is an increase in mine inflow, which has been recognised from a flooding event at Heaton Colliery, England, in the early 1800s (Zhou 2017). By the late nineteenth century, engineers at an Ohio mine were conscious that the water stored in sandstones and gravels could flow into roadways and lead to convergence, and that coal pillars were significant for stabilising overlying waters (Haseltine 1892).

Hebblewhite et al. reported that underground mining in the Southern Coalfield, NSW, Australia, caused water leakage from deeper consolidated bedrocks and shallow unconsolidated sediments, along with loss of surface flows recharging subsurface voids through stream bed cracks (Hebblewhite 2009; Hebblewhite et al. 2008). Groundwater table declines because the recharge from surface waters is insufficient to offset the rapid leakage as longwall panel retreats underneath. Subsurface water leakage was reported at some Illinois mines (Booth 2006, 2007; Booth et al. 1999, 1998) and coalfields in Pennsylvania (Wahler 1979), West Virginia (Sgambat et al. 1980), and Kentucky (Hutcheson et al. 2000; Minns et al. 1995).

The Australian experience suggests that the dewatering effect can extend to surface flows and ecology. Significant natural features including watercourses, upland swamps, and important flora and fauna in aquatic ecosystems can be disrupted (Darmody et al. 2014; Hebblewhite et al. 2008; Jankowski and Knights 2010; Kay et al. 2006; Mills 2007; Zhang et al. 2018a). Such surface dewatering was also reported in coalfields in China (Fan and Zhang 2015; Fan and Ma 2018; Sun et al. 2021; Zhang et al. 2011), US (Dixon and Rauch 1990; Gill 2000; Sidle et al. 2000), and India (Chabukdhara and Singh 2016; Dhakate et al. 2019; PanIgrahy et al. 2015), and more pronounced in arid and semi-arid mine sites with scarce rainfall recharge (Luo et al. 2018). For example, the Kuye River in northwest China used to be a perennial watercourse through Ordos Basin coalfields. 30 years of underground mining has significantly shrunk the river, and the reduced drainage area, number of tributaries, and annual streamflow are less likely to recover (Guo et al. 2016, 2019a; Li et al. 2016b; Luan et al. 2020). The water level decline can cover a far broader area than the longwall domain (Frimpter and Maevsky 1979), which gives rise to extensive ecological impacts since the dropped water table is no longer as serviceable as before. Case studies show that the topsoil moisture can be impacted (Bian et al. 2009; Yang et al. 2016), followed by degradations of wetland, agricultural, and forest ecosystems whose survival is dependent on high moisture contents (Bai et al. 2022; Hebblewhite 2009; Lechner et al. 2016; Vishwakarma et al. 2020).

These cases demonstrate that the dewatering effect of longwall mining initiates from the seam horizon and propagates to bedrock and regolith aquifers and then to surface aquatic systems. Such features can be considered indicators of mine dewatering occurring, although to some extent different between mines and coalfields due to variations in mining parameters and regional hydrogeology. As summarised in Table 1, the indicators are categorised into five levels according to how closely they relate to and to what extent they represent the dewatering.

This section identifies factors influencing the groundwater depressurisation at longwall mines. An HoF database is established sourcing extensive field data from 168 mines in tens of coalfields in China, as summarised in Fig. 1. The 422 HoF datasets are interpreted from both the mining activity and geology and hydrogeology aspects. Case analyses and comparisons are also conducted to consolidate the depressurisation contribution of some factors, or to serve as supplements for factor identification.

Summary of an HoF database

3.1 Mining activity factors

3.1.1 Panel geometry

Mining height and panel width are geometric parameters of a panel, widely accepted as dominant factors influencing fracture propagation and groundwater drainage. Here an arch model is introduced to provide an insight into the HoF correlation to panel geometry, Fig. 2. In the panel domain, continuous fracturing propagates upwards and terminates at a unit where fragmented blocks are hinged, above which the layers sag but remain uncracked. An equilibrium arch structure forms with the upper limit of continuous fracturing as the top, Fig. 2a. The arch is composed of the suspended end of each layer, bearing vertical loading and non-uniform horizontal compression.

Fracturing domain a Arching above a mined panel and b Its structural mechanics

The arch is symmetric along the vertical centreline. The left half is taken for structural mechanics analysis, Fig. 2b. The bending moment at the top vertex O obeys:

where \({h}_{f}\) denotes the arch rise, pyhsically the height of fracturing; \(w\) is the arch span, i.e., panel width; \({F}_{vi}\) and \({F}_{hi}\) are vertical and horizontal forces acting on the ith straum; \({F}_{sv}\) and \({F}_{sh}\) are vertical and horizontal reactions of the abutment. The bending moment at any point along the arch trajectory should be zero because of the static equilibirum:

Combining Eqs. (1) and (2) obtains:

Equation (3) suggests a positive correlation between HoF and panel width in the ideal context of rigid abutment. In realistic scenarios, abutment (pillar) compression and failure cause mining height and rock strength factors to be involved via the mechanical term \({F}_{sv}\), for example the Rezaei semi-empirical equation (Rezaei et al. 2015a):

where \(t\) is mining height; \(\gamma\) is the unit weight of overlying rocks; \(D\) is the depth of cover; \({E}_{p}\) and \({E}_{g}\) are elastic modulus of pillar and goaf materials. Substituting Eq. 4 into Eq. 3 indicates that the height of fracturing is positively correlated to mining height and panel width, as demonstrated by many case studies (Fan and Zhang 2015; Fan et al. 2020; Hou et al. 2020; Karacan et al. 2005; Mills and O'Grady 1998; Zhang et al. 2011) and summaries (Fan et al. 2019c; Khanal et al. 2019b; Mondal et al. 2020).

Based on this mechanical analysis, the paper compares the contribution of mining height and panel width to HoF by regression analysis on the 422 datasets. Hypothesis testing has been performed in advance to ensure the statistical comparability required for data regression. Table 2 lists the regression equation (Eq. 5) and associated information. The coefficients indicate that mining height has a greater contribution than panel width. The R-square of 0.66 suggests that the linear model captures a moderate to substantial portion of the underlying relationship between HoF and panel geometry factors. The residual 34% of the variability is attributed to other factors that the model does not account for, implying that HoF predictions only based on panel geometry factors can be inaccurate or randomly accurate.

Figure 3 compares predictions and measurements, including an HoF-t plane projection in Fig. 3b and HoF-w plane projection in Fig. 3c. Comparing Figs. 3b, c shows that the height of fracturing is more dependent on mining height because of a better consistency between predictions and measurements in the HoF-t plane. The HoF-t ratio can be almost the same between panels with similar longwall parameters at a mine, for example, approximately 18.3 for three panels (w = 300 m, D = 215 m) at Daping Mine, 20.9 for five panels (w = 198 m, D = 117 m) at Jinjitan Mine, and 27.1 for four panels (w = 193 m, D = 146 m) at Ningtiaota Mine, as shown in Fig. 3b.

Regression analysis on the correlation between HoF and panel geometry. a Exhibits the 422 HoF datasets and prediction bounds of a confidence level of 95% b and c Show projections of the 422 HoF data points and bounds onto HoF-t and HoF-w planes

3.1.2 Depth of cover

Continuous fracturing is more likely to approach the shallow aquifer and cause subsurface dewatering in shallow mining conditions. Reportedly, the height of fracturing over panel LW9 at Dendrobium Mine, Australia, was almost 100% of the depth of cover at topographically low points (Brown and Minchin 2018), partially because the panel extraction caused the low surface topography to converge towards the seam horizon. Such extensive fracturing was also captured in Ordos Basin coalfields, China, for example, panel LW22407 at 131 m depth at Halagou Mine (He et al. 2020) and LW1203 at 50–65 m depth at Daliuta Mine (Wang et al. 2018a). In these cases, all rock units above mined panels are included into caved and fractured zones, and the conventional four-zone or five-zone structure becomes no longer applicable (Meng et al. 2016; Zhao et al. 2019).

Another consideration of the depth factor is about its correlation to mine subsidence and continuous fracturing, which involves a concept termed “critical extraction”. This terminology evaluates whether full ground subsidence is allowed to occur in the longwall domain, measured using the ratio of extraction width to depth, i.e., \(w/D\). Subcritical extraction refers to where the \(w/D\) ratio is insufficient to allow the full potential subsidence, and contrast is supercritical extraction (Li et al. 2022). The Australian longwall experience suggests that,

• If individual panels have \(w/D\) ratios ≥ 1.4, the longwall domain is critical or supercritical, while,

• If individual panels have \(w/D\) ratios < 1.4, the longwall domain can be critical or supercritical when the total void width is greater than 2.5D.

Figure 4 compares several cases of incremental subsidence under sub- to super-critical conditions. The Hunter Coalfield case demonstrates a supercritical extraction, featuring minor panel interactions since the caving and fracturing within a panel is encapsulated by chain pillars and thus not significantly impacted by the extraction of subsequent panels. The subsidence profile gradually fattens and panel interaction enhances with \(w/D\) decreasing, as indicated by the Daliuta case (Xu et al. 2021) and East Ordos case. The 450 m deep longwall domain in the Southern Coalfield has a \(w/D\) ratio of 0.32, characterising a “bath-tub” concave profile with substantial chain pillar compression and panel interaction (Li et al. 2022). The impact of sub- or super-critical extraction on groundwater depressurisation can be complicated. Supercritical extraction leads to insignificant panel interactions. The dewatering is constrained within each panel, while fractures can be more interconnected in areas with high tilts and strains. In contrast, subcritical extraction leads to significant panel interactions and compression of chain pillars and surrounding strata. The groundwater depressurisation of sequential panels is mutually connected to cover the entire longwall domain, while the post-mining groundwater recovery can be more satisfactory, as indicated by the reported Jefferson County case (Booth 1999, 2002, 2006).

Incremental subsidence profiles with different \(w/D\) ratios (Li et al. 2022; Xu et al. 2021)

A data regression is conducted to evaluate the contribution of the depth of cover, obtaining the following equation:

Equation (6) suggests a positive HoF-D correlation, while the cover depth has a comparatively smaller contribution than mining height and panel width. Equation (6) has an R² of 0.77, meaning that the HoF prediction is improved after the depth of cover is taken into account. Figure 5 exhibits the prediction interval with a confidence level of 95% and 422 HoF data points. The model underestimates several top coal caving cases with large mining height (9–13.5 m) and moderate depth of cover (420–695 m). The number of cases beyond prediction bounds is decreased from the previous 134 in Eq. (5) to 87 in Eq. (6), demonstrating the influence of the depth of cover on HoF and the advantage of incorporating it as one of independent variables for HoF prediction.

Comparison between HoF predictions and measurements

3.1.3 Retreat rate

Retreat rate is considered a flexible factor that can be adjusted to change the perturbance induced by coal removal (Luo et al. 2001; Zhang et al. 2009a, 2011). A potential explanation is that increasing the retreat rate helps reduce the frequency of lower, advance, and set (LAS) cycles, mitigate the roof failure induced by excessive confinement from canopy, prolong periodic weighting intervals, and shorten the propagation of frontal abutment pressurisation (Cui et al. 2020). Another explanation involves the voussoir beam theory, where an increased retreat rate can enhance inter-block friction and decrease sliding displacement of key voussoir blocks (Wang et al. 2020b). Longwall experience in Australia, China, and the US suggests a negative correlation of dynamic deformation, hydraulic conductivity, and fracture propagation versus retreat rate (He et al. 2020; Karacan and Goodman 2009). Comparatively, some German experience holds a view that lower retreat rates can mitigate the subsidence impact on surface structures (Luo et al. 2001).

Zhang et al. investigated the retreat rate impact on groundwater drainage (Zhang et al. 2010), as shown in Fig. 6. The water table decline decreased from greater than 1.5 m at BH11 and BH17 to about 1 m as the retreat rate was gradually increased from 5 m/d near BH17 to more than 15 m/d after BH04, Figs. 6b, c, accompanied by mine inflow decreasing from 20 to 6.7 m³/h. The results indicate that an increased retreat rate helps mitigate the regional groundwater drawdown. Another case is about the direct impact of different retreat rates on pillar rib convexity, measured by the authors at a longwall panel with depth of cover of 660 to 750 m. Figure 7 compares the rib deformation at three monitoring points MP1 to MP3. The results show that MP2 where the face passes with the highest retreat rate features the most pronounced rib deformation rate. The comparison suggests that a rapid retreat causes rib convexity to be accelerated, while the total deformation reduces. Such impacts need further investigation via more field data, accompanied by numerical parametric studies to quantify the pillar compression, continuous fracturing, and incremental subsidence responses to varying retreat rates.

Groundwater drawdown responses to retreat rates at Bulianta Mine (Zhang et al. 2010). a Exhibits the distribution of monitoring boreholes b Summaries the water table variation at 11 boreholes c Compares the average water table decline before and after the face acceleration

Pillar rib convexity with face retreating at different rates. The MP1 position is taken x-axis 0, and MP2 and MP3 are respectively x = 48 and 88. The 40 m long zone centred around each monitoring point is considered the critical period of pillar pressurisation and compression due to face pass, labelled in different colours for ease of comparison

3.1.4 Chain pillar and panel interaction

Chain pillar is a key component of a longwall system. It has been identified that the pillar geometry influences rock fracturing and piezometric responses (Booth 1986; Tammetta 2016), and that increasing the aspect ratio helps enhance the pillar strength and thus stabilise overlying waters (Galvin et al. 1999; Unlu et al. 2013). The pillar contribution can be understood from the arch structure in Fig. 2. Assuming the pillar rib has a yielding depth of \({w}_{py}\), the extraction width (\(w\)) should sum the primary mined-out span (\({w}_{m}\)) and rib yielding depth (\({w}_{py}\)):

Substituting Eq. (7) into Eqs. (3) and (4) obtains that the height of fracturing increases with the yielding depth of pillar rib. Such impact is pronounced in narrow pillar configurations where the yielding depth increases due to low pillar strength and hence panel interaction intensifies in sequential panel extractions (Xie et al. 2020). For example, the Kemira Colliery experience indicates significant pillar failure and ground subsidence enhancement by more than 1 m above three sequential panels with chain pillars about 20 m (Mills 2022). Comparatively, as indicated by the Bulianta Mine experience, Fig. 6, the 50 m wide pillar constrained the horizontal expansion of groundwater drainage to BH18, to some extent contributing to the mitigation of aquifer dewatering and mine inflow (Zhang et al. 2010).

The HoF database collects several cases regarding panel interaction and HoF increment due to narrow pillars between panels. Figure 8a exhibits HoF increments with sequential panel extractions at four mines. Comparing the two histograms of each mine suggests that the second panel has a greater HoF than the first. Xiagou Mine adopted the smallest pillar width under a large mining height, featuring the most pronounced panel interaction and HoF increment. Xinglongzhuang Mine shows a slight HoF increment due to the largest pillar aspect ratio \({w}_{p}/{h}_{p}\) among the four cases. The panel interaction degree is negatively correlated to \({w}_{p}/{h}_{p}\). The comparison consolidates that panel interaction can lead to fracture network expansion and dewatering intensification (Booth 1986).

HoF increment due to panel interaction a Within a single seam and b Between adjacent seams. t, w, and w_p are mining height, panel width, and pillar width, respectively. w_p/h_p refers to the pillar aspect ratio, i.e., pillar width divided by pillar height (taking the mining height herein)

Vertical interaction can occur in multi-seam mining where the upper goaf is reactivated due to lacking competent interlayers to buffer the upward depressurisation induced by the panel extraction underneath (Yang et al. 2019). Figure 8b compares two scenarios of multi-seam mining and induced continuous fracturing above two seams. At Hongliulin Mine, the lower panel extraction causes the new fracture network to connect to the upper one because the distance between two seams is smaller than the HoF of the upper panel. The two networks are hydraulically connected and superposed onto the lower seam, resulting in an extreme HoF measured by packer testing for the lower panel. In contrast, the Kuangou case has no vertical interactions since the interlayers are thick enough to prevent the continuous fracturing from reaching the upper goaf. In longwall practice, a stacked panel layout tends to result in more concentrated depressurisation, while a staggered layout increases the draw angle and depressurisation area (Ghabraie et al. 2017). Therefore, optimising the extension distance of the lower panel may help control the interaction impact on overlying rock units and groundwater hydraulics when vertical panel interaction is inevitable in multi-seam mining.

3.1.5 Mining methods

Longwall mining usually ultises single pass, longwall top coal caving (LTCC), or slice mining method according to the coal seam thickness and mine geology. The three methods have different impacts on groundwater depressurisation, as compared in Fig. 9. Figure 9a shows that the LTCC method features the greatest HoF/t ratio with respect to the single pass and slice mining methods. The Lu’an Coalfield datasets capture some details of the HoF responses to slice mining and LTCC mining, Fig. 9b, showing that, for the 6 m thick seam, slice mining has lower HoF/t ratios (11.8–15.8) than LTCC (18.8–20.2). The difference between single pass and LTCC methods is also significant as indicated by the Zhangji Mine and Kuangou Mine datasets. Du and Gao investigated such difference between LTCC and slice mining using physical and numerical simulation techniques (Du and Gao 2017), Fig. 9c. They attributed the smaller impact of slice mining to an increase in the bulking factors of caved materials. Slice mining causes a secondary depressurisation environment where caved rocks further crack and hence refill the mined-out area to a greater extent, such that the overburden bending and fracturing can be mitigated.

Comparison of single pass, slice, and LTCC mining methods in terms of induced continuous fracturing. a Compares the mining methods at three mines b Details the Lu’an Coalfield data c Compares the height of fracturing in LTCC and slice mining scenarios (Du and Gao 2017)

Slice mining can be employed as the primary method and combined with the single pass or top coal caving method within each slice to mine ultra-thick coal seams greater than 20 m (Fan et al. 2011; Wang et al. 2019). For example, Yaojie No. 3 Mine combines slice mining and sublevel caving for a 34–85 m thick seam (Dai et al. 2022). Such combination is also used at some mines in Junggar coalfields, where the average seam thickness is 90 m (Huang et al. 2018; Qin et al. 2019; Wang et al. 2019). Underground mining of such ultra-thick coal seams inevitably causes severe groundwater drainage and flow system dysfunction (Ren and Wang 2020; Zhang et al. 2018b). Therefore, further understanding the impact of panel geometry within each slice and interaction between adjacent slices is significant for drainage mitigation and post-mining groundwater recovery.

3.2 Geological and hydrogeological factors

3.2.1 In situ stress

The stress environment in coal measure formations can be very different between rock units. In longwall practice, the horizontal stress is measured by maximum (\({\sigma }_{H}\)) and minimum (\({\sigma }_{h}\)) horizontal principal stresses. Figure 10 shows 58 stress measurements at 49 mines in China, from which the stress variation with depth is obtained as follows:

Stress conditions at 49 longwall mines, updated from (Kang et al. 2010; Yang et al. 2021). a Presents all data points and linear trendlines b Compares the variation of \({\sigma }_{H}/{\sigma }_{v}\) and \({\sigma }_{h}/{\sigma }_{v}\) with depth

Figure 10a and Eq. (8) reflect that the vertical stress increases by approximately 25 kPa per metre with depth, while the increase in \({\sigma }_{H}\) and \({\sigma }_{h}\) is comparatively irregular. Figure 10b compares \({\sigma }_{H}/{\sigma }_{v}\) and \({\sigma }_{h}/{\sigma }_{v}\), indicating that (1) Horizontal stresses are greater than the vertical in shallow depth (D < 200 m) conditions, and (2) Horizontal stresses increase with depth but show a decreasing trend relative to the vertical stress. The increase in \({\sigma }_{v}\) and decrease in \({\sigma }_{H,h}/{\sigma }_{v}\) ratios may contribute to the positive correlation between HoF and depth of cover because the horizontal confinement on overburden is equivalently reduced. In longwall systems, underground workings are generally orientated parallel to \({\sigma }_{H}\) to minimise potential rock failures (Galvin 2016). In this context, the fracture propagation, viewed from the panel width section, is mainly influenced by the \({\sigma }_{h}\) magnitude, with a negative correlation (Vakili and Hebblewhite 2010). Shear stress also exists in coal measure formations, enhancing bedding plane shearing and parting as parallel to formations, while contributing to high-angle fracturing as perpendicular to formations. Quantification of the impact of horizontal stress on continuous fracturing and dewatering remains unanswered at present. Further research based on field data coupled with numerical parametric studies is therefore recommended.

3.2.2 Competent strata and soft infillings

Overburden of longwall panels is composed of varying stratum units, among which the competent ones provide mechanical support and soft ones help narrow fracture openings due to intrinsic dilatability (Ren and Wang 2020; Zhang and Peng 2005). It has been proposed that the HoF/t ratio has a positive correlation to the number of bedding planes and average porosity of overlying strata (Palchik 2002, 2003). A mechanical understanding involves the rotation and bulking behaviour of caved rock blocks. Competent and massive rocks such as sandstone and siltstone tend to cave at a gentler angle and bulk greater than weak laminated rocks such as shale and mudstone (Galvin 2016), thus refilling the goaf and mitigating fracture propagation to a greater extent (Rezaei et al. 2015b).

Mining experience of eastern coalfields in China suggests that the height of fracturing can be smaller in formations characterised by soft and weathered soft mudstones and claystones, exhibited in Fig. 11a. A potential explanation derives from the low permeability nature and limited permeability enhancement of soft geomaterials. Figure 11b compares several coal measure rocks in terms of their permeability variation in response to external loading. The bar chart indicates that the mudstone features (1) A lower initial permeability, approximately two to three orders of magnitude, than brittle rocks, and (2) Minor permeability changes from the elastic to failure stages. Scanning electron microscope images in Fig. 11b indicate that the mudstone is rich in clay minerals on the surface of voids and fissures. Such soft infillings can swell and narrow flow pathways (Gale 2008; Seedsman 2019), potentially contributing to the halt of aquifer dewatering and hence post-mining groundwater recovery (Fan et al. 2018; Sgambat et al. 1980). Further investigations are needed to better understand the mechanism underlying the disintegration and swelling behaviour of soft infillings in the fluid flow context.

Potential contribution of soft geomaterials to mitigating rock fracturing. a Presents HoF increase with mining height in different rock strength and lithology conditions, sourcing the mining experience of eastern coalfields in China b Compares the permeability of typical coal measure rocks and their microstructures (Fan et al. 2018; Zhang et al. 2013b)

3.2.3 Geological structures

Fault is a frequent geological structure at coal mines. Such large discontinuities in the vicinity of longwall panels can be activated due to mining perturbance, causing loose geomaterials to be more fragmented and natural fissures to be extensively connected as excellent fluid flow pathways. Massive groundwater migration can occur along faults that extend from panels to upper aquifers (Zhang and Peng 2005). A recent study details hydraulic features of mine faults via investigations on the Elouera Fault, a geological structure connecting the Wongawilli Seam at Dendrobium Mine and upper Avon Reservoir in NSW, Australia (Walsh et al. 2022). Borehole loggings indicate that (1) The mine fault is characterised by fault cores comprising centimetre-sized clay or pulverised gouges and breccias, (2) Fault surfaces can be slippery due to frequent slickensides, and (3) Fault cores are surrounded by damaged zones where the fracture density is 3 to 6 per metre, greater than the 1.5 fractures per metre in host rocks. Packer testings indicate that (1) Damaged zones are more permeable than 90% of host rocks, (2) High permeable zones can be locally discontinued and thus do not form continuous water flow channels, and (3) The fault shows a decrease in transverse permeability of 0.3–0.5 orders of magnitude, which depends on the juxtaposition of less permeable strata, cataclasis and mineralisation of the cores, and clay and shale smear (Fossen 2016).

Another typical structure is outcrop. Weathered geomaterials, especially coals, near outcrops often exhibit high permeabilities due to ubiquitous cleats and joints, enabling enhanced fluid flow pathways as preferential zones for rapid groundwater recharge (Brown and Parizek 1971; Zhang and Peng 2005). Karst-pillar intrusion is a less frequent defect at coal mines, potentially carrying a large volume of water and leading to water inrush into mine workings when mining or tunnelling passes the intrusion structure (Ma et al. 2015, 2018a, 2022). The two structures are more about mine safety and localised hydrology.

3.2.4 Irregular topography

Valley is an irregular topography at mine sites, featuring steeply incised terrains where surface water accumulates and mine subsidence mechanics becomes complicated. The valley floor may contain natural fractures through which the hydraulic exchange between surface and subsurface waters becomes frequent (Booth 1986). Valley floor upsidence and valleyside closure can occur due to longwall mining and reshape the valley geomorphology, further altering the localised flow and circulation patterns, as illustrated in Fig. 12.

Surface and subsurface cracking in irregular topography. a Describes basics of upsidence and valley closure (Mills 2007, 2022) b and c Show strata failure on the valley floor (Hebblewhite et al. 2008; Zhang 2014) d and e Show topsoil cracking along hillside

The upsidence and closure are identified as nonconventional subsidence behaviours and have been investigated for many years in Australia (Hebblewhite 2009; Mills 2007; Zhang et al. 2016). As shown in Fig. 12a, the incised valley structure hinders the continuity of horizontal stress and redirects it from hills into the valley floor. After panel extraction, the floor becomes overstressed and experiences near-surface shearing and buckling to release accumulated energy, accompanied by upsidence behaviour at the base of the valley. Figures 12b, c demonstrate near-surface strata buckling and bedding slab shearing at Waratah Rivulet in NSW, Australia. Valley closure is defined as an inward convexity of valley flanks (Zhang 2014; Zhang et al. 2016), forming due to horizontal compression at topographically low points. In these cases, hillsides move towards the valley centreline, often accompanied by tensile and shear cracking on both high sides (Hebblewhite et al. 2008). Nonconventional subsidence is also captured in some valley terrains in northwestern coalfields, China, as exhibited in Figs. 12d, e.

Previous research has studied the mechanical responses of irregular topography to mine subsidence (Barbato et al. 2016a, b; Fan et al. 2017; Hebblewhite 2009; Zhang et al. 2012, 2013a). Further investigations are expected to clarify the longwall impact on the valley hydrology, especially about the natural water occurrence in valley floor and enhanced hydraulic interaction between surface and subsurface waters due to mining.

3.2.5 Mine hydrogeology

The longwall impact on groundwater varies with hydrological settings regarding water distribution, abundance, and conductivity (Hebblewhite et al. 2008; Mills 2022; Walsh et al. 2022). At mine sites, water carriers can be surface watercourses, unconsolidated regolith gravels, and porous bedrocks. Such three carriers form seven types of water occurrence, further up to 14 types by taking into account the connectedness between water and panels, as summarised in Table 3. The single surface water occurrence is a hydrologic type that the water above longwall panels primarily exists in surface containers. Rivers, lakes, and reservoirs are large-scale water bodies posing the greatest threat to longwall mines due to abundant water supplies (Brown and Minchin 2018; Hebblewhite et al. 2008). When no impermeable unit intercepts surface and subsurface interactions or induced fractures propagate to the shallow surface, mine water inrush and surface water loss are highly likely to occur. Other types of containers are less risky because of comparatively small volume and seasonal recharge.

Subsurface waters are stored in unconsolidated sand and gravel clusters. Hydrologic surveys suggest that (i) upper and middle regolith water bodies have greater volumes and smoother discharge-recharge conditions than bottom ones, and (ii) Tertiary clays have better water-resisting capability than Quaternary clays because of longer sedimentation and thus lower porosity. As measured, three unconsolidated aquifers in the 360–484 m thick regolith of Pansan Mine, China, have a decreasing water yield from the shallower 1.41 to deeper 0.56 L/(m·s); five unconsolidated aquifers in the 260 thick regolith of Xingtai Mine, China, also show a decreasing yield from the shallower 8.54 to deeper 0.01 L/(m·s), with conductivity decreasing from 73.97 to 0.02 m/d. In deeper area the bedrock water is stored in pores of sandstone and limestone units, sometimes in fissures of geological discontinuities. Pore water mainly behaves percolation under micro- to super-capillary effects, influenced by pore diameters and porosity. Bedrock water should be drained before mining if contacting the coal seam or potentially within the scope of continuous fracturing.

Compound occurrences are more common at mine sites, as depicted in the last four rows of Table 3. The hydrological impact of longwall mining is most pronounced in cases where surface and subsurface waters are hydraulically connected, as surface flows can extensively recharge the groundwater. For instance, the maximum inflow reaches 3000 m³/h at Gaojiabao Mine, China, where longwall panels are merely 84 m below an ultra-thick bedrock aquifer (400 m in thickness and 7.4 MPa in water pressure) recharged by the Weihe River. The paper further classifies groundwater bodies into four structures in terms of the aquifer-aquiclude system and its position relative to the coal seam, as depicted in Fig. 13. The four structures include:

1. (i)

   The simple structure has only one aquiclude of a single layer in the overburden, Fig. 13a. The aquiclude shows weaker anti-cracking performance than a multi-layer combination. Aquifer dewatering occurs when fractures penetrate the aquiclude.

2. (ii)

   The interlayered structure has more than one aquifer-aquiclude structure in the overburden, Fig. 13b. Aquifers are hydraulically separated by impermeable materials in the vertical direction, and thus the groundwater mainly behaves horizontal flow.

3. (iii)

   The enclosed and semi-enclosed structure represents that water bodies are entirely or partially encapsulated by impermeable materials, Fig. 13c. This structure characterises a limited source of recharge compared to others, and therefore the induced dewatering can be localised.

4. (iv)

   The unconformable structure forms when the aquifer-aquiclude structure has an unconformable contact with the coal seam, Fig. 13d. This structure may contain one of the above three structures in localised areas. Water barriers should be installed along coal seam boundaries to prevent water inrush.

Typical hydrological structures at mine sites

Besides the above representative structures, mines may have hydrological anomalies associated with faults and Karst intrusions. Identifying the mine groundwater hydraulics should be fundamental for assessing the potential of aquifer dewatering due to sequential panel extractions and hence the likelihood and duration of groundwater recovery.

3.2.6 Time-dependency of groundwater recovery

Groundwater recovery is a component of the hydrological impact of longwall mining, featuring a time dependency as it occurs with fracture closure after mining. Longwall experience of Yangquan coalfields suggests a time course of overburden responses, including the first 13 days for depressurisation development, 24 days for caved rock compaction, and 42 days for goaf reconsolidation stabilisation. Therefore, a groundwater recovery can be 5–11 weeks after the face passes the monitoring point. The Bulianta experience, Fig. 6, represents that the declined water started to recover about 7 weeks after mining.

A complete recovery needs a longer period. Figure 14 presents the water table variation at two mines in Illinois coalfields, US (Booth 1999; Booth et al. 2000). In the Jefferson County case, Well P350 captured a 10 m recovery 6 months after the maximum drawdown, followed by complete recovery in about one year. The adjacent Saline County featured a delayed recovery, where Well P54B recorded a further drawdown even two years after mining, and a complete recovery therein can be very challenging or may not occur if the site has poor transmissivity and inadequate recharge pathways. Besides, the long-term monitoring at Jefferson County captured that the recovered water table was higher than the pre-mining level. This is because the panel extraction resulted in a closed-end “bath-tub” zone in which the aquifer was elevated and well yields were increased (Booth et al. 1998). The difference between the two cases relates to the regional hydrology (Booth et al. 1998), including surface runoff and precipitation supply, aquifer storativity and rechargeability, and regolith water-holding capacity. This case comparison demonstrates the significance of mine hydrogeology identification and long-term piezometric monitoring for longwall impact assessment.

Water drawdown and post-mining recovery at Jefferson County and Saline County, US (Booth 1999). The monitored mining was from December 1988 to April 1989 at Jefferson, and from April 1992 to April 1993 at Saline

Researchers have adopted field measurement, lab test, and numerical modelling methods to understand the mechanism of longwall mining impacting groundwater stability, revealing that longwall mining leads to significant increases in rock porosity and permeability. This section analyses the dewatering mechanism from permeability enhancement and fracture propagation perspectives, considering two different roles they play in groundwater drainage.

4.1 Rock permeability enhancement

In groundwater hydraulics, permeability is often used to measure the ability of rock media to allow fluid to pass through them. The paper uses permeability to describe the changes in fluid flow condition. There have been extensive laboratorial studies about rock permeability variation under external loadings (Wang et al. 1997; Zhang et al. 2018c; Zhu and Wong 1997), which includes a slight decrease due to pore convergence in the elastic stage, significant increase with progressive failure in the yield stage, and fluctuation in the residual stage (Fan et al. 2018; Heiland 2003; Heiland and Raab 2001; Zhao et al. 2017). The overall permeability enhancement features a positive correlation to the specimen fracturing degree (Fan et al. 2018; Wu et al. 2005). The lab-scale fractured rock hydraulics implies that, within a panel domain, the caved rock mass should have the highest permeability due to extreme fracturing, while the overlying intact strata have minor hydraulic changes. Such field-scale understanding has been represented by a number of conceptual models listed in Table 4.

Hydraulic characteristics of a panel domain can be drawn from varying models in Table 4. Horizontally, the permeability enhancement is symmetrical along the panel centreline and more pronounced near trough edges due to high-angle tensile fracturing, indicated by Models III, V, and IX. Vertically, rock permeability generally increases with depth and to the maximum in the caved zone. The fractured zone features significant permeability enhancement by 1000 (Mills 2022) to 2000 times (Guo et al. 2007) and water head decline due to smooth drainage, ranging from 15 to 60 times of the mining height. The area labelled aquiclude zone (Models I and II), constrained zone (Models III, IV, VI, and VIII), or additional zone (Model V) has minor permeability enhancement but comparatively greater increases in transmissivity and storativity due to bedding plane shearing and parting (Mills 2022), potentially preventing the intrusion of overlying waters into mine voids underneath. Near ground surface the tensile areas lateral of the subsidence trough show an increased permeability, through which rainwater or other surface water sources can infiltrate the ground. Overall, the hydraulic differential forms a gradient along which the upper water tends to drain to the goaf under gravitation and flow momentum effects.

4.2 Connected fractures as flow pathways

An indispensable condition for mine dewatering is that induced fractures should form a connected and conductive network to allow significant volumes of groundwater flow (Gale 2008), among which the isolated ones have rather limited contribution to smooth drainage (Kowalski et al. 2021). It is therefore necessary to assess the distribution of connected fractures and their connectivity. Lab experiments coupled with interpolated estimations from in situ packer testings have identified influencing factors of fracture connectivity, as summarised in Table 5, where internal factors are about attributes of fractures within a network, and external factors involve the rock environment in which fractures propagate.

Internal factors indicate that the interconnectivity of mine fracture network should be greater in the caved zone and minimum in the constrained zone where rock layers sag but remain intact (Hebblewhite 2020). Areas lateral of the subsidence trough have an increased connectivity due to local high-angle fractures (Heritage et al. 2022a). In shallow mining scenarios, the connectivity may show a monotonic decrease with elevation because of through-going fracturing (Brown and Minchin 2018; Brown and Walsh 2022; Meng et al. 2016; Zhao et al. 2019). Influence of external factors implies that (i) stiff and brittle rock masses have greater fracture connectivity than coaly and tuffaceous ones (Galvin 2016), (ii) the high connectivity of caved and fractured zones decreases with goaf reconsolidation over time, accounting for the post-mining recovery of in situ stress and groundwater, and (iii) coal measure rocks in context of narrow pillar configuration or multi seam mining have a higher degree of fragmentation and fracture connectivity due to intensified panel interaction or repeated pressurisation and depressurisation.

The increasing connectivity with depth forms a smooth fracture flow network through which overlying waters drain to the mined-out area underneath. Fracture flow dominates the drainage behaviour, as connected fractures provide steeper pressure drop gradients and smoother fast-flow pathways than rock matrices (Booth and Spande 1992; Hill and Price 1983; Jeanpert et al. 2019). A fundamental model for fracture flow description is the cubic law:

where \(Q\) is flow rate, in \({\text{m}}^{3}/\text{s}\); \(\Delta p/L\) measures the pressure gradient along flow path, in \(\text{pa}/\text{m}\); \(\mu\) is dynamic viscosity, in \(\text{Pa}\bullet \text{s}\); \(l\) is the size of fracture orthogonal to flow, in \(\text{m}\); \(a\) is fracture aperture, in \(\text{m}\). The cubic law assumes laminar flow through a smooth tabular chamber, ideally two parallel plates with distance (aperture) much smaller than length along the fluid flow direction. Snow investigated multiple pathways in jointed rocks and proposed a model for hydraulic conductivity (Snow 1965, 1969):

where \(\rho\) is fluid density, in \(\text{kg}/{\text{m}}^{3}\); \(g\) is gravitational acceleration; \({n}_{i}\) is the unit normal vector of the conduit and \({n}_{j}\) is direction cosines of the unit vector, together determining the orientation of the conduit; \({s}_{i}\) is an average spacing of the joint set along the sampling line; \({\delta }_{ij}\) is the Kronecker delta. The summation is taken over all members of the joint set. Zoorabadi et al. innovated the Snow model to describe the hydraulic conductivity of jointed rock mass (Zoorabadi et al. 2012, 2022):

where \({V}_{R}\) is the volume of rock mass before excavation and \({V}_{r}\) is volume of excavation; \({m}_{k}\) is the number of fractures of the kth set; \({c}_{k}\) is interconnectivity coefficient, a positive number not greater than 1 and calculated via \(({\sum }_{k=1}^{n}\frac{{l}_{k}}{{s}_{j}}\text{sin}{\gamma }_{kj})/{I}_{max}\) (\(k\ne j\), \({l}_{k}\) is the mean trace length of set \(k\), \({s}_{j}\) is the mean spacing of set \(j\), \({\gamma }_{kj}\) is the average angle between fractures of set \(k\) and set \(j\), and \({I}_{max}\) is the total interconnectivity index of the fracture network); \({d}_{k}\) and \({a}_{k}\) are respectively the diameter and aperture of set \(k\); \({n}_{ik}\) and \({n}_{jk}\) are direction cosines of the unit vector normal to each fracture. The Zoorabadi model considers fracture frequency, orientation, size, and aperture, obtaining a more accurate result than the Snow model when applied to estimate the hydraulic conductivity of rock mass (containing five joint sets with micron-scale apertures, with a mean conductivity of \(1.8\times {10}^{-8} \text{m}/\text{s}\)) surrounding a tunnel.

Compared to existing fracture flow descriptions, flow pathways in longwall domain have a broader distribution (15–60 times of mining height), higher frequency (5–10 on the total water loss horizon and 10–15 near the goaf centre Heritage et al. 2022a, b), and larger apertures (up to 100 mm (Price et al. 2022)). Extensive intersections further complicate the pressure drop of fracture flow, and the drainage behaviour in fractured and caved zones may involve high-Reynolds flow and non-Darcian turbulence (Tammetta 2015). There have not been analytical or semi-empirical models applicable to describe fluid flow through extensive interconnected pathways. Addressing the knowledge gap should be based on fracture network identification, which, however, is limited by field data acquisition, especially for bedrock fractures beyond the detection range of acoustic and seismic techniques. In this context, mine fracture observation should be improved, accompanied by continuous fracturing representation in numerical models, to help quantify fracture connectivity and hence establish a description of the fracture flow behaviour.

Empirical equation, field measurement, and numerical modelling are frequently used to evaluate continuous fracturing and the potential of groundwater drainage. This section discusses current applications of the three methods to groundwater drainage assessment and identifies their limitations that should be concerned in future research.

5.1 Empirical equations for mine dewatering prediction

Empirical prediction of mine dewatering utilises HoF prediction equations. Australian longwall practice often adopts two empirical equations, one established by Ditton et al. using HoF datasets from Australian coalfields (Ditton and Merrick 2014), and one by Tammetta using data collected from tens of mines in eight countries (Tammetta 2013). Chinese mines often use equations proposed by China Coal Research Institute (CCRI) according to the longwall experience of eastern coalfields in China (CCRI 1981). Table 6 lists the three equations and their applicability.

Figure 15 compares the three equations along with Eq. 6 in terms of their performance in estimating the height of fracturing at 20 panels in NSW and Bowen Basin coalfields in Australia (Ditton 2014) and 20 panels in Ordos Basin coalfields in China (He et al. 2020). The Ditton Equation has the best performance for Australian cases, potentially because it considers more influencing factors than others. Equation 6 has a better performance for Chinese cases, while less accurate for Australian cases since the equation is established from HoF datasets collected from Chinese coalfields. The CCRI Equation underestimates all the Chinese cases with significant errors, meaning that such an equation based on the eastern longwall experience is no longer applicable to Ordos Basin coalfields where longwall panels usually have a greater geometry and shallower depth.

Comparison of four empirical equations for HoF estimation. Notes The first 20 data points shaded in green are Australian cases, and the following 20 points shaded in red are Chinese cases. The CCRI Equation uses Eq. 11.1 for Australian cases considering a competent rock environment (Ditton 2014), and Eq. 11.2 for Chinese cases, where the overburden is medium competent (He et al. 2020; Ren and Wang 2020). The effective thickness of competent stratum is 20 m for Chinese cases (Fan and Zhang 2015; Zhang et al. 2010, 2011)

The comparison indicates that (i) empirical prediction is often constrained by the background information of the employed equation due to potentially significant differences in mining parameters and geological conditions between mines, and (ii) considering more contributing factors can help improve the accuracy of empirical prediction. Therefore, continuous HoF data collection and high-quality statistical interpretation should be conducted, which are significant for establishing a reliable model for longwall impact estimation.

5.2 Mine data collection and management

Continuous measurement and data collection have been a critical part of Australian longwall practice, helping evaluate mine risks, update longwall experience, and provide parameters and criteria for numerical model construction and verification. For example, at Elouera Mine, massive data collected from field mapping, downhole video inspection, geophysical logging, packer testing, and overcore measurements helped the mine control water inflow from the Avon Reservoir to underground workings (Walsh et al. 2022). The Standardised Subsidence Information Management System (SSIMS) established in NSW, Australia, allows users to interrogate the information stored in the Subsidence Database for purposes of coal mining and risk regulation in regard to subsidence mechanics (Li et al. 2022).

Measured data should be managed with explicit classifications that enable efficient data retrieval and interpretation, detection of missing components, and identification of geological difference between panels or mines. The paper designs a structure for a future potential mine groundwater database, as exhibited in Fig. 16. In the database, every panel should contain at least one dataset of ground subsidence, height of fracturing, rock mass conductivity, and water table as key data. Panel geometry, stratigraphic profile, in situ stress, and rock strength are included as basic data, determined from mine drawings, borehole logs, borehole breakout records, and core test reports, respectively. The database should have a wide breadth, requiring that the datasets of sequential panels within a coal seam should be extended to deeper seams, adjacent mines, and even neighbouring coalfields. Another requirement is the coverage in depth, involving regular borehole geophysical logging and packer testing to determine the variation in HoF and hydraulic conductivity with longwall progress and time, accompanied by long-term monitoring of groundwater drawdown and recovery. Many mines in Australian coalfields, such as Dendrobium Mine (Brown and Walsh 2022), United Colliery (Heritage et al. 2022a), and Springvale Mine (Corbett 2022) are examples of extensive field measurement and long-term monitoring.

Conceptual structure of a mine groundwater database

5.3 Field-scale numerical modelling

Numerical modelling is often used to assess rock mass failure and fracture propagation. The approach relies on a methodical workflow including mine data analysis, rock strength determination, in situ stress transformation, coding for model construction, model calibration, and model verification. Compared to empirical predictions, field-scale modelling is time-consuming but potentially more informative by configuring site-specific parameters about panel geometry and geological condition. Many numerical techniques have been developed to assess fluid flow or percolation due to underground excavations, as summarised in Table 7.

Representation of rock fracturing and fluid flow can be different between numerical techniques. As compared in Fig. 17, discontinuum model is composed of discretised rigid particle or deformable blocky entities, and in-between physical contacts constituting joints along which adjacent entities can separate and shear to form rock fractures, Fig. 17c. Particle diameters and joints are fabricated in model construction, usually referencing the frequency and orientation of discontinuities captured in lab tests, Fig. 17a, or field measurements, Fig. 17b. Continuum models do not exhibit fractures, instead indicating permeable areas via user-defined criteria such as plastic states (Whittles et al. 2006; Zhang et al. 2019), pore pressure enhancement (Guo et al. 2008; Smith et al. 2019), and damage evolution (Li et al. 2011).

Rock fractures and their discontinuum and continuum representation in numerical modelling. a Shows a fractured mudstone specimen after UCS test b Exhibits fractured cores collected from a borehole at Springvale Mine, Australia (Corbett 2022) c and d Indicate the difference between discontinuum and continuum models in representing rock fracturing and fluid flow

On a panel scale, the continuum method has been applied to investigate overlying rock failure and potential flow pathways. Figure 18a shows three FLAC2D models established in a parametric study (Gale 2008). The modelling obtains that (i) the fracture network is characterised by bedding plane shear and tensile fracturing or mobilisation of existing joints in tensile zones, (ii) flow pathways are direct in the caved zone and gradually tortuous with elevation, and (iii) the height of fracturing, fracture connectivity, and hydraulic conductivity (assuming Darcian flow) increase as greater subsidence occurs. Discrete Element Method (DEM) modelling has increasing applications due to its intrinsic feature in exhibiting rock fracturing. As shown in Fig. 18b, Adhikary et al. simulated a panel extraction within Lidsdale/Lithgow Seam at Springvale Mine and induced continuous fracturing by means of the PFC2D software (Adhikary et al. 2017, 2020). The model incorporated the Discrete Fracture Network (DFN) technology to configure high-angle interfaces along the immediate roof for caving initiation. It was identified that the height of connected fractures was 260 m above the goaf centre and 290 m above the abutment, suggesting no complete groundwater drainage zone forming above the mined panel (Adhikary et al. 2017).

Continuum and discontinuum modellings of panel extraction and induced rock failure and fracture propagation. a Exhibits three FLAC2D models for longwall mining within Whybrow Seam at Hunter Valley, Australia b Exhibits a PFC2D model of panel LW411 (3.2 m in mining height, 315 m in panel width, and 350 m in depth of cover) at Springvale Mine, Australia. The DEM model is 457 by 450 m and contains approximately 503 thousand rigid particles, adopting the built-in ‘flat-joint’ model for physical contacts

Given these present applications, the paper proposes three elements that should be concerned in future panel- to mine-scale modelling.

1. (1)

   Size effect in rock parameter determination

Lab test on intact rocks is a major source of strength parameter acquisition, typically using standardised cylinders or cubes with edge lengths less than 200 mm. The intact rock strength should be upscaled with respect to the geometry of rock matrices in panel- to mine-scale models considering the size effect; otherwise, the rock deformation and failure are underestimated (Baghbanan 2008; Castelli et al. 2003; Follington and Isaac 1990; Kong et al. 2021). There have been several equations for rock strength upscaling. Hoek and Brown established an equation \({\sigma }_{m}={\sigma }_{r}{\left(50/{d}_{m}\right)}^{0.18}\) to determine rock mass UCS (\({\sigma }_{m}\)) via cylindrical specimen UCS (\({\sigma }_{r}\)) and rock mass geometry (\({d}_{m}\)) (Hoek and Brown 1980). Yoshinaka et al. proposed \({\sigma }_{m}={\sigma }_{r}{\left(58.1/{d}_{m}\right)}^{k}\), \(k\) from 0.1 to 0.3 for homogenous hard rocks and 0.3 to 0.9 for weathered or microflawed rocks (Yoshinaka et al. 2008). The paper updates a new lab-to-field relationship in rock UCS (Eq. (12)) and elastic modulus (Eq. (13)), by combining coal measure rock (mass) strength collected from labs and mines with some reported data points (Bieniawski 1992; Gale 1992; Pratt et al. 1972; Yoshinaka et al. 2008), as exhibited in Fig. 19.

where \({\sigma }_{m}\) and \({E}_{m}\) are the UCS and elastic modulus of rock matrix, and \({\sigma }_{r}\) and \({E}_{r}\) the UCS and elastic modulus of rock specimen. \({d}_{m}\) is the size of rock matrix, taking the length of the edge normal to the major loading. \({d}_{r}\) is the diameter of cylindrical specimens, or edge length of cubes, using a typical value of 50 mm if no test data available. Equation (12) enlarges the upper limit of UCS upscaling to about 8 m by incorporating the strength of intact core of chain pillars. The size effect also influences the stiffness and strength of physical contacts in DEM models. Such rock joint parameters can be calibrated by simulating uniaxial and triaxial compression tests and fitting the model results to measured values. There have been some workflows for the calibration (Christianson et al. 2006; Gao and Stead 2014).

UCS and elastic modulus of rocks with different size

1. (2)

   Rock fracture representation via DFN technology

The DFN technology can be used to fabricate rock joints and identify connected fractures, having two major types of application in fracture propagation and fracture flow assessments:

1. (i)

   Pure DFN configuration without mechanical calculation. This method constructs a fracture network by interpreting fracture intensity, orientation, length, and aperture data. Fracture flow is calculated by setting hydraulic boundaries, sometimes accompanied by rock matrix percolation. The pore pressure effect on fracture aperture is neglected. This method is used for understanding fluid flow through jointed rock masses (Liu et al. 2016b; Min et al. 2004; Zoorabadi 2014)

2. (ii)

   DFN regeneration after mechanical calculation. In model construction, a zero-aperture DFN is built according to fracture statistics to fabricate jointed rock masses. Aperture generates with rock matrices displacing and deforming in mechanical calculations. A new DFN is generated by mapping open contacts, hence forming flow pathways with apertures inherited from contacts. This method has been used in roadway- to panel-scale models (Adhikary et al. 2017, 2020; Khanal et al. 2019a; Poulsen et al. 2018; Wang et al. 2020a; Wei et al. 2022).

The applications indicate that, in model construction, the DFN configuration is controlled by fracture statistics determined from measured fracture attributes. Geological mapping helps detect the trace length, frequency, and orientation, for example the LiDAR system for fractures along outcrops or walls of underground excavations (Vlachopoulos et al. 2020), and ground penetrating radar (GPR) for subsurface fractures (Hawkins et al. 2017; Molron et al. 2020). Borehole geophysical logging uses acoustic and optical scanners to obtain the frequency, orientation, and position of bedrock fractures that intersect the borehole (Brown and Walsh 2022; Corbett 2022; Heritage et al. 2022a; Walsh et al. 2022).

Measured data can be statistically interpreted to determine the probability distribution function (PDF) of each attribute. Lognormal, exponential, gamma, and power-law functions are frequently used for rock fracture statistics (Bonnet et al. 2001). Figure 20 shows a workflow by which the fracture length measured by authors in Ordos Basin coalfields is identified to prefer a lognormal distribution. A primary assessment is first conducted according to the skewness (\({S}_{k}\)) and shape of the probability curve. \({S}_{k}\) measures the asymmetry of the probability distribution of variables (\(X\)) with respect to their mean value (\(\overline{X }\)), expressed as:

where \(n\) is the total number of variables, and \(s\) the standard deviation. \({S}_{k}\) is 0 in normal and 2 in exponential but not a fixed value in lognormal, gamma, and power law distributions. The distribution function should feature a long tail (the Pareto principle), a single mode of a bell-shaped curve, and a bulge left of the peak in power law, lognormal, and gamma distributions, respectively. The primary assessment suggests higher likelihoods of lognormal and gamma distributions. A further data fitting identifies the best-fit distribution to be lognormal with a mean (\(\mu\)) of 0.91 and standard deviation (\(\sigma\)) of 0.44. The lognormal function is validated via K-S test. The workflow is also available for interpreting other fracture attributes.

A workflow for identifying fracture length statistics

1. (3)

   Model verification and hybrid modelling

Panel- to mine-scale models should have minor errors, usually no more than 10% (Bahrani and Hadjigeorgiou 2018; Gao et al. 2015; Zhang et al. 2020a, 2021), with respect to measurements in chain pillar compression, height of fracturing, and incremental subsidence. This ensures an accurate representation of the mechanical behaviour of rocks from the seam horizon to ground surface. Hydrogeological models should verify groundwater hydraulics when fluid flow is incorporated, for example piezometric variation (Adhikary and Guo 2015; Adhikary et al. 2017; Booth 2002; David 2018; Khanal et al. 2019b; Rapantová et al. 2007, 2012) and mine inflow (Adhikary et al. 2017; Guo et al. 2008; Khanal et al. 2019a).

Different techniques can be combined to form a hybrid modelling method. FDM-DEM coupling has been applied to investigate rock mechanics (Adhikary et al. 2017; Cai et al. 2007; Mahabadi et al. 2010) and field-scale rock mass depressurisation (Wang et al. 2015; Zhao 2020), reducing the long computation time in single DEM simulation. PFC2D and COSFLOW were combined to assess the fluid flow environment above mined panels (Adhikary et al. 2017, 2020; Khanal et al. 2019a), where COSFLOW served for complementing and benchmarking the strata permeability enhancement obtained from PFC2D models. Such hybrid modelling helps address the limitation of geomechanical models in representing mine- to coalfield-scale piezometric variation.

The paper identifies knowledge gaps that should be addressed for obtaining more accurate mine dewatering assessment. Quantification of fracture network interconnectivity and description of fracture flow should be developed at first as they determine whether the hydraulic gradient and mine inflow rate can be accurately captured. In addition, identification of the regional hydrology needs an increased attention, especially for valley terrains, because the induced drawdown and post-mining recovery are influenced by the hydrology itself in regard to (i) aquifer occurrence and water abundance, (ii) groundwater circulation pattern, and (iii) interaction between surface and subsurface waters. The effect of varying retreat rates and stress conditions on continuous fracturing and dewatering can be investigated later, accompanied by the contribution of mineral infillings to fracture closure and long-term water recovery.

Figure 21 presents a framework and lab- to field-scale topics for future research. Methodologically, continuous data collection and field-scale numerical modelling should be enhanced as key technical support, requiring that

• A mine groundwater database with HoF and piezometric drawdown as key data should be established, effectively managed, and continuously updated from long-term monitoring at sequential panels.

• Empirical predictions should take into account at least panel geometry, depth of cover, and rock strength factors as predictors, keeping a constant revision with the database renewal.

• Numerical models should be able to indicate continuous fracturing and hydraulic head decline due to mining, with geological to hydrological verifications to improve the model reliability for parametric studies.

A research framework for improving mine groundwater drainage assessment

In the first stage, measured fractures are statistically interpreted and reproduced by the DFN technology into field-scale numerical models. Induced fractures that are mutually interconnected and directly or indirectly connected to the goaf should be identified with an appropriate logic and its algorithmic implementation. Typical morphology of intersected fractures within the network can be captured to provide reference for fabricating flow chambers for fracture flow tests in lab. The correlation between hydraulic gradient (J) and flow rate (Q) can be obtained from lab tests and used to benchmark CFD models to investigate the contribution of aperture, frequency, and intersection to head loss and flow nonlinearity via sensitivity analysis. A semi-empirical model should be established to describe fluid flow through large-aperture and interconnected fractures.

In the second stage, hydro-mechanical model should be established by incorporating the fracture flow description and verified against measured water table decline and rock mass conductivity. Two series of parametric studies are needed, including one based on packer testing simulation to investigate the correlation of rock mass conductivity and head drop to fracturing degree, and another based on mine-scale models to quantify the response of continuous fracturing and dewatering to different longwall parameters. HoF data recorded in the database and the correlation between HoF and longwall parameters are combined to develop a new empirical prediction equation. The results of parametric studies and empirical interpretations constitute a methodology to evaluate mine hydrology and groundwater hydraulics variation due to longwall mining. Cases updated into the database can be used to revise the methodology and keep it reliable for mine groundwater depressurisation assessments.

The paper investigates the factors and mechanism of mine dewatering and identifies knowledge gaps to be addressed in future research to gain better understanding and assessment of groundwater depressurisation due to longwall mining.

It is identified that the piezometric drawdown as a direct result of sequential panel extractions can further affect subsurface and surface hydrologic features at mine sites. Statistical analyses of 422 HoF datasets collected from 168 mines indicate that mining height has a greater contribution to the groundwater impact than panel width and depth of cover. The dewatering impact is more pronounced in cases where (i) the seam is extracted using LTCC mining method and (ii) panel interaction is intensified due to narrow pillar configuration within a seam or thin and incompetent interlayers between adjacent seams. Geological and hydrogeological factors such as in situ stress, rock strength and clay mineral infillings, fault structures, valley terrains, and surface and subsurface water interactions also contribute to the longwall impact on groundwater and determine the likelihood and duration of water recovery in the long term.

Interconnected fracture network formed by continuous fracturing dominate the dewatering mechanism, providing steep hydraulic gradients and smooth flow pathways for overlying waters draining to the mined-out area underneath. The fluid flow environment may involve high-Reynolds flow and non-Darcian turbulence due to high fracture frequency and large fracture aperture, complicated than those that have been considered in conventional fracture flow models. In this context, fracture connectivity quantification and fracture flow description should be developed in future research. Continuous field data collection, data management and interpretation, and panel- to mine-scale numerical modelling should be enhanced as technical support. The paper designs a research framework that contains lab- to field-scale topics to narrow the knowledge gaps and achieve improved understanding and assessment of groundwater depressurisation due to longwall mining.