Ground Subsidence Hazard Analysis in an Abandoned Underground Coal Mine Area using Probabisltic and Logistic Regression Models

Saro Lee1, Kidong Kim2, Hyun-Joo Oh2 and No-Wook Park1

1

Geoscience Information Center, Korea Institute of Geoscience and Mineral Resources Daejeon, Korea E-mail: leesaro@kigam.re.kr 2 Department of Earth System Sciences, Yonsei University Seoul, Korea method, but above empirical methods were developed under condition of horizontal coal seam and long wall working which are dominant in Europe. In Korea, due to the mixed structure of geology there are various widths of coal seam, irregular inclined angle of coal seam and strata and slant-chute block caving method has been using. As a result, a sink-hole type subsidence is general, therefore another estimation of ground subsidence is necessary. Therefore, the aim of this study is to assess and predict of ground subsidence for hazard mapping of near near an abandoned underground coal mine area using GIS. To decide a study area, field investigation and reinforcement reports related with ground subsidence were carefully considered then select a site called Simpori where 48 ground subsidence signs had been found near AUCM at Samcheok city. The study area is located between longitude 12900 ~ 12903 and latitude 3711 ~ 3712. The coal resource of South Korea almost consists of anthracite and 85% of them had been deposited during the upper Paleozoic era and the lower Mesozoic era in the Jangseong Formation of the Pyeongan Supergroup. The study area is located around the Hanyang gallery on the Jangseong and Keumcheon Formation. Along the study area the Oship fault, Youngdong rail road, and no. 38 local road pass by [2]. The location map of this study appears in Fig. 1. II. SPATIAL DATABASE AND METHODOLOGY

AbstractFor quantitative analysis of presumptive ground subsidence near Abandoned Underground Coal Mine (AUCM), this study applied, verified and compared a probability model, a frequency ratio and statistical model, a logistic regression at Simpori of Samcheok city in Korea, using a Geographic Information System (GIS). To evaluate susceptibility of the subsidence, database was constructed using topographical map, geological map, mining tunnel map, GPS data, land use map, lineaments, DEM and borehole data. Then, the relationship between the factors and the existing subsidence were calculated using frequency ratio and logistic regression models. The relationships, frequency ratio and logistic regression coefficient, were overlaid to determine each factors rating for ground subsidence susceptibility mapping. Then the ground subsidence susceptibility map was compared with known ground subsidence locations and verified. The verification results showed very high prediction accuracy (about 90%). The frequency ratio model (93.29%) had higher prediction accuracy than logistic regression (89.35%). The maps can be used to reduce hazards associated with ground subsidence and to land cover planning. Keywords; ground subsidence; abandoned underground coal mine; frequency ratio; logistic regression; GIS

I.

INTRODUCTION

Since 1989, almost all of underground coal mines were abandoned and few are remaining, and the occurrence of ground subsidence around abandoned coal mine area has become a serious social problem in Korea recently. However, quantitative assessment of predicted ground subsidence area is difficult especially in coal mining area where structure of geology is very complicated. To predict the probability of ground subsidence empirically with (1) intact strength of the rock, (2) stress field, (3) geological structure of the rock, (4) depth of the mining horizon, (5) area extent of mining, and (6) volume extracted per unit area of mining was suggested within surprisingly narrow limits considering the form of the input data [6]. The National Coal Board published a basic technique to find out estimated area of influence by ground subsidence with height of cavity, width of mined panel and inclined angle of coal seam [4]. The prediction method of subsidence area is very dependent on a structure of local geology and coal mining

Many studies indicated important factors related with ground subsidence around coal mines such as followings [1], [2]: Depth and height of mining cavities, caving method, inclined degree of caving, scope of mining, structure of geology, flow of ground water and mechanical characteristics of rock mass(RMR). So, factor related to ground subsidence occurrence were constructed in a vector-type spatial database. These included 1:50,000-scale geological map, 1:5,000-scale topographic maps, 1:5,000-scale land use maps, 1:1,200-scale mined tunnel maps and borehole data. The geology was extracted 1:50,000 scale geological map and the distance from lineament were calculated using the lineament data. Contour and survey base points that had an elevation value read from the topographic map were extracted,

0-7803-9510-7/06/$20.00 2006 IEEE

1549

and a Digital Elevation Model (DEM) was constructed. Using the DEM the slope gradient was calculated. There are 12 classes of landuse which is extracted from land use map of National Geographic Institute. Most literatures said that major factor of ground subsidence is the scope of mined cavities. Therefore, constructing database of depth and width of mined cavities is very important work during this study. To achieve the object, (1) GPS measurement was used to find out an exact position of a mine head, (2) vectorizing a hard copy of mined tunnel map with it, and (3) after converting the vectorized mined tunnel map to an ASCII grid file, minus it with DEM raster data. There were 37 boreholes at the study area but some boreholes did not have values, so Inverse Distance Weighting (IDW) interpolation method was used to contour the value of RMR, groundwater level and permeability factors and reclassified it using ArcGIS. This study was conducted using GIS, a frequency ratio, and a logistic regression with factors that may caused ground subsidence. Image database and attribute database concerning about ground subsidence were constructed using ArcGIS, and weights for each database were determined through logistic regression. A key assumption using this approach is that the potential (occurrence possibility) of ground subsidence will be comparable to the actual frequency of ground subsidence. After the study area was selected, ground subsidence occurrence areas were detected in the study area by field surveys. A map of existing ground subsidence was developed with a GIS, and this was used to evaluate the frequency and distribution of ground subsidence in the study area. To represent the distinction quantitatively, frequency ratio and logistic regression models were used. For this analysis, the calculated and extracted factors were mapped to a 1 m resolution grid. The raster data were converted for the statistical program used. Then, using the frequency ratio and logistic regression models, the spatial relationships between the ground subsidence area and each ground subsidence related factor, such as topography, depth of mined tunnel, borehole data, geology and land use were analyzed in the statistical program, and a formula of ground subsidence occurrence possibility was extracted using the relationships. The formula was used for calculating the subsidence hazard index, which was mapped to each grid cell. Finally, the hazard map was verified using known ground subsidence areas and success rates were calculated for quantitative verification. In this study, GIS software, ArcView 3.3 and ARC/INFO 9.0 version, and statistical software, SPSS 12.0, were used as the basic analysis tools for spatial management and data manipulation. Frequency ratio approaches are based on the observed relationships between distribution of ground subsidence areas and each subsidence-related factor, to reveal the correlation between ground subsidence locations and the factors in the study area. Using the frequency ratio model, the spatial relationships between subsidence occurrence location and each factors contributing ground subsidence occurrence were derived. The frequency is calculated from analysis of the relation between ground subsidence and the considered factors. Therefore, the frequency ratios of each factors type or range were calculated from their relationship with ground subsidence events. In the relation analysis, the ratio is that of the area

where ground subsidence occurred to the total area, so that a value of 1 is an average value. If the value is greater than 1, it means a higher correlation, and value lower than 1 means lower correlation. Logistic regression allows one to form a multivariate regression relation between a dependent variable and several independent variables. Logistic regression, which is one of the multivariate analysis models, is useful for predicting the presence or absence of a characteristic or outcome based on values of a set of predictor variables. The advantage of logistic regression is that, through the addition of an appropriate link function to the usual linear regression model, the variables may be either continuous or discrete, or any combination of both types and they do not necessarily have normal distributions. In the case of multi-regression analysis, the factors must be numerical, and in the case of a similar statistical model, discriminant analysis, the variables must have a normal distribution. In the present situation, the dependent variable is a binary variable representing presence or absence of ground subsidence. Where the dependent variable is binary, the logistic link function is applicable [5]. For this study, the dependent variable must be input as either 0 or 1, so the model applies well to ground subsidence possibility analysis. Logistic regression coefficients can be used to estimate ratios for each of the independent variables in the model. Quantitatively, the relationship between the occurrence and its dependency on several variables can be expressed as: p = 1 / (1 + e-z) (1)

where p is the probability of an event occurring. In the present situation, the value p is the estimated probability of subsidence occurrence. The probability varies from 0 to 1 on an S-shaped curve and z is the linear combination. It follows that logistic regression involves fitting an equation of the following form to the data: z = b0 + b1x1 + b2x2 + + bnxn (2) where b0 is the intercept of the model, the bi (i = 0, 1, 2, , n) are the slope coefficients of the logistic regression model, and the xi (i = 0, 1, 2, , n) are the independent variables. The linear model formed is then a logistic regression of presence or absence of ground subsidence (present conditions) on the independent variables (pre-failure conditions). APPLICATION OF THE LIKELIHOOD RATIO AND LOGISTIC REGRESSION MODELS Likelihood ratios for the class or type of each factor were calculated by dividing the ground subsidence occurrence ratio by the area ratio. The likelihood ratios of each factors type or class were summed to calculate the subsidence hazard index (SHI), as shown in Equation (3) SHI = Fr (3) where Fr is the frequency ratio of each factors type or class. If the SHI value is high, it means a higher hazard to l ground subsidence; a lower value means a lower hazard to III.

0-7803-9510-7/06/$20.00 2006 IEEE

1550

ground subsidence. The SHI value index is shown in Fig. 2 for interpretation. The index was classified into five classes based on equal area for visual and easy interpretation. The minimum value is 1.27 and maximum value is 52.21, the mean value is 8.996 and the standard deviation value is 6.773. In the logistic regression model, The spatial databases of each factor were converted to ASCII format files for use in the statistical package, and the correlations between ground subsidence and each factor were calculated. In addition, logistic regression formulae were created as shown in equations (4). Finally, the probability that predicts the possibility of subsidence-occurrence was calculated using the spatial database, data from and equations (1) and (4): z = (-0.034 SLOPE) + (-0.095 DISTDRIFT ) + (-0.079 WATERLEVEL ) + (-0.036 DEM) + (-1.338 PERMAG ) + (0.368 RMR) + (0.001 DEPTHDRIFT ) + GEOLb + LANDUSEb + 3.952 (4) where SLOPE is slope value; DISTDRIFT is distance from drift value; WATERLEVEL is depth to ground water level value; DEM is dem value; DEPTHDRIFT is depth of drift value; PERMAG is permeability value; RMR is RMR value; GEOLb LANDUSEb are logistic regression coefficient; z is a parameter) Using formula (1) and (4), the possibility of subsidence occurrence was calculated and using the possibility, a subsidence hazard map was made. The distribution of calculated possibility is shown in Fig. 3. The possibility was classified into five classes based on equal area for visual and easy interpretation. The minimum value is 0.00 and maximum value is 0.64744. The mean value is 0.008961 and the standard deviation value is 0.03482 The subsidence hazard analysis result was validated using known ground subsidence locations. Validation was performed by comparing the known ground subsidence location data with the subsidence hazard map. Each factor used and frequency ratio was compared. The rate curves were created and its areas of the under curve were calculated for all cases. The rate explains how well the model and factor predict the subsidence. So, the area under curve can assess the prediction accuracy qualitatively. To obtain the relative ranks for each prediction pattern, the calculated index values of all cells in the study area were sorted in descending order. Then the ordered cell values were divided into 100 classes, with accumulated 1% intervals. The rate verification results appear as a line in Fig. 4. For example, in the case of frequency model used, 90 to 100% (10%) class of the study area where the subsidence hazard index had a higher rank could explain 73% of all the subsidence. In addition, the 80 to 100% (20%) class of the study area where the subsidence hazard index had a higher rank could explain 85% of the subsidence. In the case of logistic regression model used, 90 to 100% (10%) class of the study area where the subsidence hazard index had a higher rank could explain 79% of all the subsidence. In addition, the 80 to 100% (20%) class of the study area where the subsidence hazard index had a higher rank could explain 96% of the subsidence. To compare the result quantitative, the areas under

the curve were re-calculated as the total area is 1 which means perfect prediction accuracy. So, the area under a curve can be used to assess the prediction accuracy qualitatively. The area under the curve is shown in Fig 4. In the case of logistic regression model used, the area ratio was 0.9501 and we could say the prediction accuracy is 95.01%. In the case of frequency ratio model used, the area ratio was 0.9329 and we could say the prediction accuracy is 93.29%. Overall the case of all factor and logistic regression model used showed a higher accuracy than cases of each factor and logistic regression used and all factor and frequency ratio model used. IV. CONCLUSIIONS AND DISCUSSION Ground subsidence is among the most hazardous of artificial disasters. Government and research institutions worldwide have attempted for years to assess subsidence hazards and risks and to show their spatial distribution. In this study, a statistical approach for identifying the hazardous area of subsidence using GIS shows considerable promise. As the result, the ground subsidence maps were made using frequency ration and logistic regression models and they showed prediction accuracy 90 % in frequency ratio model and 90% in logistic regression model. In the case, the logistic regression model showed the better result than frequency ratio model. The subsidence areas of this study are located around railroad, road, and other facilities above shallow mined tunnel. Therefore, low elevation and depth of mined tunnel become important factors as well as groundwater level. The data of groundwater level were obtained during field survey without considering an amount of rainfall at that time, but it has a meaningful value and should be considered to calculate safe rate of a base rock. For further study, exact maps of mined tunnel, lots of borehole data and geophysical data will be needed to design an underground model of this study area to analyze ground subsidence more quantitatively. The frequency ratio model is simple and the process of input, calculation and output can be understood easily. Moreover, because the likelihood ratio value can be used as a rating, there is no need to convert the database to another format, such as ASCII. A large amount of data can be processed in a GIS environment quickly and easily. The logistic regression model requires data to be converted to ASCII format for use in the statistical package and later reconverted to incorporate it into the GIS database. Moreover, the large amount of data cannot be processed in the statistical package quickly and easily. However, degree of ground subsidence hazard can be analyzed qualitatively. Statistical packages can allow analysis of subsidence hazard, but they are inconvenient for management of spatial data. A GIS has few if any functions for statistical analyses, but has many functions for database construction, display, printing, management and spatial analysis. Therefore it is necessary to integrate the GIS and the statistics to reduce the restrictions of using the two applications separately. The benefits of integrating GIS and statistical programs are efficiency and ease

0-7803-9510-7/06/$20.00 2006 IEEE

1551

of management, input, display and analysis of spatial data for subsidence hazard. ACKNOWLEDGMENT The Coal Industry Promotion Board provided whole investigation reports and basic GIS database. REFERENCES
[1] [2] [3] A. C. Waltham, Ground Subsidence, Blackie & Son Ltd: New York, 1989, pp. 49-97. Coal Industry Promotion Board, Fundamental investigation report of the stability test for Simpori, vol. 97, pp. 5-38, 1999. Coal Industry Promotion Board A study on the Mechanism of subsidence over abandoned mine area and the Construction method of subsidence prevention, Coal Industry Promoton Board, vol. 97, pp. 167, 1975. National Coal Board, Subsidence Engineers Handbook, National Coal Board Mining Department, London, 1975, pp. 111. P. M, Atkinson and R. Massari, Generalized linear modeling of susceptibility to landsliding in the central Apennines Italy, Computers & Geosciences, vol. 24, pp. 373-385, 1998. S.C. Goel and C. H. Page, An Empirical Method for Predicting the Probability of Chimney Cave Occurrence over a Minig Area, Int. J. Rock Mech. Min, Sci. & Geomech. Abstr. Vol. 19, pp. 325-337, 1982.

Figure 2. Ground subsidence hazard map using frequency ratio model.

[4] [5]

[6]

Figure 3. Ground subsidence hazard map using logistic regression model.

Figure 4. Cumulative frequency diagram showing ground subsidence hazard rank occurring in cumulative percent of ground subsidence occurrence. Figure 1. Ground subsidence in study area

0-7803-9510-7/06/$20.00 2006 IEEE

1552