1. Introduction
Environmental protection is becoming an integrated part of mining policies and technologies. Protecting the surface, its natural, economic, and other values from the effects of mining is an important part of ecological endeavors [
1,
2]. In addition to general degradation, mining production is increasingly moving to urban areas, where much more important infrastructure is at risk [
3,
4]. The mining organization is therefore obliged to implement direct protective measures on the surface or underground infrastructure and water regime within the exploitation area [
5]. The measured displacements and deformation of the surface are legal proof of the incurred mining damage. Obtaining and analyzing these data is the starting point for evaluating the prognostic theories and their rational application. Mining damage is reflected on the surface in the form of subsidence troughs which are that part of the earth’s surface that has been transformed under the influence of mining [
6,
7,
8]. A change in shape in the vertical direction (subsidence) is observed between the initial and final state of the surface when equilibrium in the undermined rock mass is reached. Intermediated states before equilibrium are defined as dynamic troughs [
9]. The undermining process takes place in space and time, from the collapse of equilibrium in the rock mass at the time of excavation, during excavation, and even after the excavation has ended, until the asymptotic cessation of subsidence when equilibrium is restored [
6,
10,
11,
12].
Approaches with most practical value for subsidence prediction can be generalized to profile functions [
13,
14,
15,
16], influence functions [
17,
18,
19,
20], stochastic models [
21,
22], and numerical models [
23,
24,
25]. In recent years, various modern approaches to the treatment of surface subsidence have been introduced, among which the cellular automata (CA) [
26] and smoothing spline [
27] theories should be mentioned. By means of profile functions the final shape of the surface subsidence trough is predicted in different profiles through the excavation area [
5]. In order to determine the profile, it is necessary to know the values of the maximum subsidence, which is calculated using the technological parameters of the excavation, and some of the parameters specific to the shape of the trough [
16]. In general, these methods are used in such a way that the unknown parameters in the equations are adjusted until the calculated function represented the best fitting curve compared to the real profile of subsidence trough observed in nature. A commonly known prognosis method for determining the effect of an elemental or differentially small underground excavation on the formation of a surface subsidence trough are influence functions. The combined impact of these elemental excavations represents the impact of the entire excavation field and enables the calculation of the shape of the final subsidence trough and its intermediate states [
17]. The principle of such approaches is the functional relationship between surface subsidence and the height of the excavation or the displacements (convergence) of the rock mass directly above the excavation. Elemental subsidence increases to a total maximum value at the critical size of the excavation field (subsidence does not increase as the excavation continues to increase) [
18]. In addition, the effect of full excavation within the impact cone of the critical surface is dependent on the depth of excavation. Intermediate or partial dynamic subsidence troughs transit into a definite static state or final trough [
12]. Another approach are stochastic models of discontinuous media used to mimic what is happening during undermining, where the collapsing process of an undermined rock mass is considered a random process [
21]. A stochastic medium is an infinite set of elements that move through mutual contacts under the influence of a field of (gravitational) forces. Solving such a system of elements with multiple degrees of freedom is very difficult using classical methods from mechanics, but results can be obtained with other methods. In practical terms, the stochastic model is also limited to the behavior of the undermined rock mass or soil bound to the observation area and in some cases requires verification with physical laboratory models [
28,
29].
The demand for local and time-varying mechanical properties of the undermined rock and coal mass has guided researchers to develop much more successful numerical models of continuum mechanics. With numerical analysis, using the finite element method in 2D or 3D space, results can achieve very good comparison to actual rock mass deformations around the longwall excavation [
23,
24] or to actual surface subsidence above it [
25]. Though numerical modeling offers great potential of predicting subsidence with very good accuracy, their practicality is restricted by acquisition of necessary rock and coal mass mechanic characteristics. An alternative are rule-based models, such as cellular automata (CA), which advantages are shown in simple formulation and fast computations of complicated subsidence processes, allowing numerous simulations in a short time period. According to Sikora [
26], application of cellular automata leads to results that are consistent with influence functions and stochastic models, making CA a practical approach with great expansion probabilities. A less extensive but efficient method was proposed by Orwat and Mielimaka [
27], who used smoothing splines. Here, in addition to the measured data, a reliable prognosis is also needed to estimate future subsidence. On the basis of the averaged measured values and the predicted subsidence, an approximation of the surface subsidence and terrain inclination courses can be made.
The focus of our research was the development and demonstration of a nonexponential mathematical model, and practical verification based on field measurements and observations of various technological exploitation conditions. Some fundamental principles were taken into account: (i) Surface subsidence value is asymptotically ceasing to its final and maximum value; (ii) the shape of the predicted subsidence trough can be easily analyzed with profiles throughout the observation area; (iii) the behavior of surface subsidence process can be interpreted by stochastic analysis; (iv) the final subsidence trough form can be obtained by summing individual impacts of the excavation.
In this paper, we present a method for surface subsidence prediction based on statistical analysis of the measured surface subsidence and geometrical modifications within the technological parameters of the excavation. By combining these proven partial solutions, we tried to satisfactorily accurately solve the more complex problems of surface subsidence above longwall underground excavations, where coal is extracted both horizontally and vertically. In the presented case study, the observation area is subjected to the ongoing reclamation of the degraded terrain as the excavation progresses. Through the statistical analysis of such a stochastic process, a prediction of the time of consolidation was made, when further subsidence of the dumped soil (from terrain reclamation) is negligible. These calculations are possible for a single point on the surface observed in different epochs, where the object of study is the vertical coordinate of the point and the time of observation. Photogrammetric measurements with the help of an unmanned aerial vehicle (UAV) were used to acquire data in the form of a 3D point cloud of the surface observation area. To avoid numerous analyses of individual points in the cloud, the observation area was divided into several rectangular sectors of the same dimensions, each defined with its centroid to represent the impacts within the entire sector. By using rectangles instead of squares or triangles, the number of further computational steps along the length of the observed area was also significantly reduced. In the next step, planes were fitted to the data of all epochs within each sector and the plane centroids were calculated. Our proposed method allowed the point cloud to be reduced to a number of observation points (centroids) that can easily be studied. With the consolidation prognosis, an estimated time of active subsidence can be obtained and thus terrain reclamation can, to some extent, be controlled within sectors. The idea is to identify sectors that require additional soil dumping to achieve the estimated consolidation. In order to predict surface subsidence that occur under the influence of underground excavation by vertical extraction, we considered the real excavation height in each centroid above the excavation. Starting from the theoretical assumptions regarding the maximum subsidence over the excavations with vertical coal extraction, determination of total subsidence was made. Both the proposed consolidation and total surface subsidence prognosis can be used in the profile, point cloud or sector comparison, giving a concrete and valuable insight of surface subsidence caused by underground mining. A 3D point cloud comparison was also done to observe the intensity of the subsidence and identify sinking holes or areas of soil dumping.
4. Discussion
To study surface subsidence caused by mining in VCM, we used ten 3D point clouds acquired with UAV photogrammetric monitoring from 2 February to 1 August 2017 (
Table 3). In ten epochs, including the initial, vertical excavation heights were measured in 17 points along the excavation face (
Figure 4 and
Table 2). During monitoring and simultaneously as excavation progressed, ongoing reclamation of the degraded terrain was visually inspected and identified areas of intensive soil dumping were recorded (
Figure 5).
Each obtained point cloud consisted of 4,837,195 points, so the observation area was divided into 50 sectors with dimensions of 52.6 × 52.9 m
2 (
Figure 6a). By breaking the cloud into smaller parts, we obtained 50 points clouds per epoch, each with 96,944 points. Sector vertex points coordinates were used to calculate the centroid of each sector (
Figure 6b). With the centroids, five profiles across the width (
Figure 7a) and ten across the length (
Figure 7b) of the observation area were defined. Thus, allowing surface subsidence visualization and analysis by profiles. The centroid, being a geometrical center of the belonging sector, describes the occurred subsidence within the entire sector with one value, the height. To obtain centroid heights, planes that best fitted the according 3D point cloud were calculated along with their vertices and centroids. Fitted planes layouts were spatially observed with a five times larger z-axis scale (
Figure 9) and the layout of initial epoch was compared to the layout of last epoch (
Figure 10). Distances between centroids of both layouts are marked with red line segments to highlight the centroid height difference (subsidence or elevation). Two representative profiles Cw7 and CL3 were selected to show centroid points and their subsidence characteristics at different epochs (
Table 4).
The research task included the development of a predictive subsidence model of any observation point on the surface, which is influenced by underground mining and constant terrain reclamation, and the determination of the total impact of exploitation. After a critical review of modeling approaches, where the adaptability of mathematical models to real data was evaluated, we decided on a stochastic analysis using the FNSE model (Equation (1)). By fitting the model to a fixed number of observation points at which the subsidence dynamics is interpolated within the precision limits, the time and amount of active subsidence were extrapolated. Additionally, using the consolidation prognosis (
Figure 8) the time of consolidation was estimated. Centroid subsidence value, predicted to occur during active subsidence period, was then used to categorize each sector. An example of a consolidation prognosis of centroid C
33 is shown in
Figure 15, where it was calculated that, by the time of consolidation, occurred subsidence value from initial epoch will be 6.233 m. From last measurement to predicted time of consolidation, the subsidence of value 0.088 m is expected. Consolidation prognosis does not take into account the technological parameters of the excavation, and since it is known from practice that surface subsidence above the excavation will continue to occur for a longer duration, we used the predicted subsidence dynamics to categorize surface areas according to the degree of reclamation required to maintain terrain levels until consolidation time is reached. For centroid C
33 the estimated subsidence exceeds the value of 6 m and so sector 7-3 falls into category IV. Meaning that intensive terrain reclamation in the form of soil dumping and stiffening is required in sector 7-3 to limit further active subsidence within an estimated value of 0.088 m. Categorization of the overall layout of the sectors is shown in
Figure 16.
By introducing the technological parameters of the excavation into the prognosis, a prediction of total subsidence in one and two years after the completion of excavation was also carried out. In this approach, we focused only on the centroids above the excavation area. Considering that 75% (one-year prediction) and 90% (two-year prediction) of the total excavation height represent the maximum draft, the initial centroid heights were modified by H0–0.75hv or H0–0.9hv. Assessing the final state, a rough assumption was made on how much subsidence is expected in the coming years and how much material is needed to maintain terrain level during active subsidence period.
Calculated centroid heights based on actual data, estimated heights at time of consolidation, and final subsidence along profiles Cw7 and CL3 are shown in
Figure 11,
Figure 12,
Figure 13 and
Figure 14. Calculated values are also presented in
Table 6. Comparing the calculated height of the centroids and the estimated consolidation, we can see that the trend is well matched. Due to terrain diversity, ongoing reclamation, and variable excavation heights, the bottom of the developed subsidence trough is not flat or the maximum subsidence is not constant.
Figure 12 shows agreeable consolidation prognosis of centroids directly above the excavation (C
23, C
28, C
33, C
38, C
43, C
48) and according to calculated heights in the last epoch. With exceptions of centroids C
3, C
8, C
13, and C
18, where some deviations can be noted. The reason for this is the influence of a previously mined longwall panel on a higher level located more to the south of the observed excavation. Which is also reflected in intensive terrain reclamation operations in area (3) shown in
Figure 5. The CL3 profile in
Figure 12 shows that the terrain level between the C
3 and C
13 centroids was maintained at a constant level, but due to enhanced subsidence, the terrain is much lower in the last epoch. Therefore, the consolidation prognosis for this area (3) also included the long-term impacts of a previously mined longwall panel. Estimated centroid heights at the time of consolidation can then be accepted as final heights, supposing total subsidence of a previous excavation in this area has been reached. For all remaining centroids located northeast of the Cw3 profile (
Figure 7a), the heights were modified to enable determination of 75% and 90% of total subsidence. Total subsidence prognosis along profiles Cw7 and CL3 is shown in
Figure 13 and
Figure 14. Estimated centroid heights using the consolidation and total subsidence prognosis were analyzed with CloudCompare. This made it possible to compare the results of both approaches to the 3D point cloud of the initial epoch and calculate the volume between consolidated terrain and final terrain (
Figure 20,
Figure 21 and
Figure 22).
Results of the C2C analysis of 3D point cloud from the initial epoch and the estimated consolidated heights of the centroids (meshed) indicate that a sinking hole may occur in the northern part (a) of the observation area (darker blue) and an increase of subsidence in zones (b) and (c) (
Figure 21). At the same time, we see that there is a correlation between the color pattern of the C2C analysis and sectors layout categorization (
Figure 16).
The results of the analogue C2C analysis for the determined 75% value of the total subsidence, which is to take place in the first year after the end of the excavation, i.e., at a similar time as the consolidation is to be achieved, are shown in
Figure 22. It can be observed that, considering the real excavation height, the bottom of the surface subsidence trough cannot be flat if the excavation height is not constant. The largest subsidence will occur in the middle and along the length of the excavation, where the excavation height is maximum (
Figure 4). The smallest elevation changes are in the area that is not covered by the total prognosis (red-yellow). The volume between consolidated and final terrain is 733,825 m
3, which provides a robust assumption of the material that will be required for ongoing reclamation.
In order to verify the proposed prognosis or categorization approaches, nine different cloud comparisons were generated by using UAV photogrammetric data (
Figure 17,
Figure 18 and
Figure 19). Four sinking holes were identified when comparing point clouds from the initial and last epochs (
Figure 19c). Depending on the estimated consolidation, the S1 sinking hole in zone (a) (
Figure 21) can be confirmed. Sinking holes S3 and S4 lie in areas of increased subsidence (b), while S2 was not detected. By observing the categorized sectors layout in
Figure 16 and the results of the C2C analysis, the correlation between actual subsidence and sector categories can be observed. Where actual subsidence is greater, the categories of sectors are higher and vice versa.