Department of Civil Engineering, Gifu University, Gifu, Japan b Department of Civil Engineering, Tokyo City University, Tokyo, Japan

Transcription

1 This article was downloaded by: [Rungbanaphan, Pongwit] On: 1 February 2011 Access details: Access Details: [subscription number ] Publisher Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: Registered office: Mortimer House, Mortimer Street, London W1T 3JH, UK Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards Publication details, including instructions for authors and subscription information: Settlement prediction by spatial-temporal random process using Asaoka's method Pongwit Rungbanaphan a ; Yusuke Honjo a ; Ikumasa Yoshida b a Department of Civil Engineering, Gifu University, Gifu, Japan b Department of Civil Engineering, Tokyo City University, Tokyo, Japan Online publication date: 09 November 2010 To cite this Article Rungbanaphan, Pongwit, Honjo, Yusuke and Yoshida, Ikumasa(2010) 'Settlement prediction by spatialtemporal random process using Asaoka's method', Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, 4: 4, To link to this Article: DOI: / URL: PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: This article may be used for research, teaching and private study purposes. Any substantial or systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

2 Georisk Vol. 4, No. 4, December 2010, Settlement prediction by spatial-temporal random process using Asaoka s method Pongwit Rungbanaphan a *, Yusuke Honjo a and Ikumasa Yoshida b a Department of Civil Engineering, Gifu University, Gifu, Japan; b Department of Civil Engineering, Tokyo City University, Tokyo, Japan (Received 17 July 2009; final version received 10 November 2009) A methodology was presented for observation-based settlement prediction with consideration of the spatial correlation structure of soil. The spatial correlation is introduced among the settlement model parameters and the settlements at various points are spatially correlated through these geotechnical parameters, which naturally describe the phenomenon. The method is based on Bayesian estimation by considering both prior information, including spatial correlation and observed settlement, to search for the best estimates of the parameters at any arbitrary points on the ground. Within the Bayesian framework, the optimised selection of auto-correlation distance by Akaike s Bayesian Information Criterion (ABIC) is also proposed. The application of the proposed approach in consolidation settlement prediction using Asaoka s method is presented in this paper. Several case studies were carried out using simulated settlement data to investigate the performance the proposed approach. It is concluded that the accuracy of the settlement prediction can be improved by taking into account the spatial correlation structure and the proposed approach gives the rational prediction of the settlement at any location at any time with quantified uncertainty. Keywords: settlement prediction; Asaoka s method; spatial-temporal process; Bayesian estimation; random field 1. Introduction So far, all methods of predicting future settlement using past observations have been based solely on the temporal dependence of their quantity. However, the fact that soil properties tend to exhibit a spatial correlation structure has been clearly shown by several studies in the past, e.g. Lumb (1974), Vanmarcke (1977), DeGroot and Baecher (1993), Baecher and Christian (2003, 2008). It is therefore natural to expect that the accuracy of settlement prediction can be improved by taking into account the spatial correlation of ground properties, by which the observed settlement data from all observation points can be rationally utilised. Furthermore, by introducing spatial correlation, it is possible to rationally predict the ground settlement at any arbitrary points at any time based on the spatialtemporal structure. This study is actually an attempt to search for such an approach. Observation-based settlement prediction is considered to be a practical approach since it is independent of the initial and boundary conditions. Asaoka (1978) proposed a trend equation of time series data of settlement for a one-dimensional consolidation process based on an autoregressive model. The hyperbolic method, by which the curve of consolidation settlement with time is proved to be fitted very well by a rectangular hyperbola over a specific range of data, has also been proposed as a powerful approach for predicting settlement by several researchers (Sridharan et al. 1987, Tan 1994 etc.). For the current study, however, Asaoka s method was chosen since it is soundly founded on the physical processes of consolidation theory. Asaoka (1978) also proposed a set of formulas to estimate the posterior probability density function (PDF) of the model parameters concerning the prior PDF and the previously observed settlement data. This approach, however, gives only general estimation of the model parameters for the total area of observations. In other words, the settlement at a specific point on the ground cannot be predicted by this approach. In order to include spatial correlation into the settlement prediction model, a Bayesian approach is chosen for this research owing to its ability to systematically combine subjective information, i.e. the prior information including the spatial correlation, and objective information, i.e. the observation data. A general formulation is presented based on the Bayesian estimation concept to search for the best estimators of unknown parameters at any arbitrary points when the observation is made at discrete spatial points and discrete time. However, owing to *Corresponding author. rungbanaphanp@yahoo.co.jp ISSN print/issn online # 2010 Taylor & Francis DOI: /

3 Georisk 175 the fact that spatial correlation structure is controlled by auto-correlation distance, the estimation of this key parameter is necessary. By considering this as a model selection problem, it will be shown that, within the Bayesian framework, auto-correlation distance and observation error can be appropriately selected based on Akaike s Bayesian Information Criterion (Akaike 1980, Honjo and Kashiwagi 1999). Similar formulations have been proposed previously by the authors (Rungbanaphan et al., submitted May 2009). However, the former paper emphasised practical aspects of the proposed method in using actual observation data for the prediction of secondary compression in peat based on the linear relationship between settlement and logarithm of time, i.e. an S log(t) model, whereas the current paper focuses more on providing a sound and consistent theoretical framework of the spatial-temporal model and its application to primary consolidation settlement prediction using Asaoka s method. In fact, this paper emphasises that the estimation of the settlement of an arbitrary point at an arbitrary time can be given solely based on the Bayesian statistical framework. This estimation is actually equivalent to performing a simple Kriging method (Krige 1966, Matheron 1973, Wackernagel 1998) for parameter interpolation based on the estimates of the unknown parameters at the observation points. There is a misunderstanding that Kriging has nothing to do with the Bayesian approach. However, it is clear from the formulations proposed in the paper that Kriging can be included in a Bayesian approach. Therefore, prediction of settlement at an arbitrary point can be specified in the same format as that at observation points. In order to investigate the performance of the proposed approach in spatial-temporal settlement prediction, two-dimensional random fields of the model parameters are simulated by the frequency domain technique (Shinozuka 1971, Shinozuka and Jan 1972) and then the settlement data is generated using the simulated model parameters with considering the observation error. Based on this generated settlement data, the statistical inferences of the model parameters are back-calculated using the proposed approach. The result shows significant improvement in terms of parameter estimation error when spatial correlation is considered, compared to the case of ignoring spatial correlation. The sensitivity of this improvement for the variation of auto-correlation distance, spacing of observation points, and number of observation points is investigated. The estimation error of the auto-correlation distance, the standard deviation of observation error, and the final settlement at an arbitrary location is also discussed later in this paper. 2 Spatial-temporal process 2.1 Bayesian estimation considering spatial correlation structure In order to improve the estimation and to enable local estimation, utilisation of Bayesian estimation including spatial correlation is proposed in this paper. This approach uses prior information of the unknown parameters which characterise soil behaviour, e.g. model parameters or soil properties, and the observation data, e.g. observed settlement, from all observation points to search for the best estimates of the unknown parameters. The formulation consists of two statistical components, namely, the observation model and the prior information model. These two models will then be combined by Bayes theorem to obtain the solution Observation model This model relates the observation data to the unknown parameters which are defined in a multivariate stochastic Gaussian field Z(x)[z 1 (x), z 2 (x),..., z P (x)] T where x is a spatial vector coordinate; P is the total number of the unknown parameters; and z i (x) (for i1, 2,..., P) is a random function of an unknown parameter (e.g. soil or model parameter) at any location x in a specific domain. This paper proposes a method for identifying the best estimator of Z for a discrete spatial point field, x 1, x 2,..., x n, x n1,..., x m, which is defined as where Ẑ[ẑ 1 ; ẑ 2 ; ; ẑ P ] T (1) ẑ i [ẑ i (x 1 ); ẑ i (x 2 ); ; ẑ i (x m )]; i1; 2; ; P (2) Suppose that a set of observations Y k (e.g. ground settlement) at the discrete time step k, i.e., 2,..., K, has been obtained at n observation points x 1, x 2,...,x n. Y k is defined as Y k [y k (x 1 ); y k (x 2 ); ; y k (x n )] T (3) It should be noted that x n1, x n2,..., x m are defined as any arbitrary points at which the unknown parameters are supposed to be estimated, i.e. (m n) interpolation points. The general formulations of the observation model have been presented in, for example, Hoshiya and Yoshida (1996), Honjo and Kashiwagi (1999), etc. Here it is assumed that the observation Y k is expressed as a linear function of Z with observation error of o as follows: Y k M k Zo (4) where o is the Gaussian observation error vector which is assumed to follow N(0, V o ). V o is defined as a covariance matrix of o where V o s 2 o I n,n. s 2 o is the

4 176 P. Rungbanaphan et al. variance of the observation error and I n,n is an nn unit matrix. This implies that the observation errors are assumed to be spatially independent. M k is the n(p m) coefficient matrix, which is defined as M k [M 1 n;n 0 n;mn M 2 n;n 0 n;mn M P n;n 0 n;mn ] (5) where M i n,n denotes an nn coefficient matrix, relating ẑ i to Y k ;0 n,mn denotes an n(mn) zero matrix, attaching to each M i n,n to eliminate the unknown parameters at (m n) arbitrary points (i.e. x n1, x n2,..., x m ) from the observation model. Given Z and s 2 o, the predicted settlement distribution at any time step k can be represented by the following multivariate normal distribution p(y k jz; s 2 o )(2p)n=2 jv o j 1=2 exp 1 2 (Y k M k Z) T V 1 o (Y k M k Z) Prior information model It is assumed that the prior information of the unknown parameters has the following structure ZZ 0 d (7) where Z 0 is the prior mean vector (P m dimension), a deterministicvector; d is the uncertainty of the prior mean which is assumed to follow N(0, V Z ) where V Z is a prior covariance matrix. By introducing the spatial correlation structure in the formulation of V Z, we have 2 s 2 z1 V 3 C 0 s 2 z2 V Z V 6 C : 7 4 :: 5 (8) 0 s 2 zp V C where s 2 z1, s 2 z2,...,s 2 zp represent the prior variance of the unknown parameters z 1, z 2,..., z P, respectively. V C is the auto-covariance matrix which is defined as 2 3 r(j x 1 x 1 j) r(j x 1 x m j) 6 : V C n :: n (9) r(j x m x 1 j) r(j x m x m j) (6) r(jx - ix - jj) denotes the auto-correlation function where x - i, x - jspatial vector coordinate. Several analytical expressions have been proposed for the auto-correlation function but none of them can claim any fundamental basis (Vanmarcke 1977). The exponential type auto-correlation function is chosen for the current study because it is commonly used in geotechnical applications (e.g. Vanmarcke 1977, Fenton and Griffiths 2002, Griffiths and Fenton 2004 etc.). The function is given as r(j x i x j j)exp[j x i x j j=h] (10) where h auto-correlation distance. To emphasise, this parameter is assumed to be constant in any directions in the horizontal plane. This implies that the anisotropy of soil is not considered in this case. In addition, it should be kept in mind that this type of auto-correlation function is, in fact, chosen only as an example for an application of the proposed method. In practice, several types of autocorrelation functions may be tested and the one which best fits the observation should be used. It should also be noted that, for the sake of simplification, there are two important assumptions about the correlation structure for formulating the above covariance matrix (V Z ). First, the unknown parameters, z 1, z 2,..., z P, are assumed to be independent of each other. Second, the correlation structures of these parameters are identical, meaning that they share the same auto-correlation distance. In fact, these assumptions can be released without major changes of the formulation, if the observation data is available in the amount that the detail specification of the spatial correlation is possible. Given h, prior means, and prior variances of the unknown parameters, the prior distribution of the model parameters is also a multivariate normal distribution of the following form p(zjh)(2p) (P m)=2 jv Z j 1=2 exp 1 2 (ZZ 0 )T V 1 Z (ZZ 0 ) (11) Bayesian estimation Suppose that the set of observations Y k at the discrete time step, 2,..., K has already been obtained. By employing Bayes theorem, the posterior distribution of the state vector Z can be formulated as p(zjy; s 2 YK o ; h)c p(zjh) p(y k jz; s 2 o ) (12) where Y denotes the set of all observation data, i.e. Y(Y 1, Y 2,..., Y K ), and c denotes the normalising constant. By substituting Equations (6) and (11) into the above equation, we have p(zjy; s 2 o ; h)c (2p)[P mk n]=2 jv Z j 1=2 jv o j K=2 exp 1 (ZZ 0 ) T V 1 Z 2 (ZZ 0 ) XK (Y k M k Z) T V 1 o (Y k M k Z) (13) The Bayesian estimator of Z, i.e. Ẑ; is the one that maximises the above function. Therefore, it is

5 Georisk 177 equivalent to minimising the following objective function J(Zjs 2 o ; h)(zz 0 )T V 1 Z (ZZ 0 ) XK (Y k M k Z) T V 1 o (Y k M k Z) (14) It should be noted that s o 2 and h are assumed to be given in this case. The Bayesian method, however, does not provide the rational way to determine these values. In order to choose the most appropriate values of s o 2 and h based on the information in hand, Akaike s Bayesian Information Criterion (ABIC) is introduced and presented in the next section. 2.2 Model selection: Akaike s Bayesian information criterion Choosing appropriate values of s 2 o and h can be considered as the model selection problem. Akaike s Bayesian Information Criterion (ABIC) introduced in this study is developed on the same information theory principal as Akaike s Information Criterion (AIC) which is specifically used to selected the best model from several alternative models (Akaike 1980, Honjo and Kashiwagi 1999). By considering s 2 o and h as hyperparameters, the Bayesian likelihood can be formulated as follows: L(s 2 o ; hjy; Z) YK p(zjh) p(y g k jz; s 2 o ) dz (15) Substituting Equations (6) and (11) into the above equation, we obtain L(s 2 o ; hjy; Z) g (2p) [P mk n]=2 jv Z j 1=2 jv o j K=2 exp 1 2 XK (ZZ 0 ) T V 1 Z (ZZ 0 ) (Y k M k Z) T V 1 o (Y k M k Z) dz (16) By performing integration of the above equation, we have L(s 2 o ; hjy; Z)(2p)K n=2 jv Z j 1=2 jv o j K=2 j XK M T k V1 o exp 1 2 XK M k V 1 Z j 1=2 (ẐZ 0 ) T V 1 Z (ẐZ 0 ) (Y k M k Ẑ) T V 1 o (Y k M k Ẑ) (17) Note that Zˆ denotes the Bayesian estimator of Z, by minimising the objective function in Equation (14). By substituting Equation (14) into the above equation and taking the natural logarithm, we have the log Bayesian likelihood l(s 2 o ; hjy; Z)1 2 lnjv Z jk 2 lnjv o j 1 2 ln j XK M T k V1 o M k V 1 Z j 1 2 [J(Ẑjs2 o ; h)]const (18) The general definition of ABIC (Akaike 1980) is given as ABIC(2) log (maximum Bayesian likelihood) 2 (number of hyperparameters). In this case, the number of hyperparameters is fixed as 2. Thus, we have ABIClnjV Z jk lnjv o jlnj XK M T k V1 o M k V 1 Z [J(Ẑjs 2 o ; h)] const (19) By optimising the above equation, the values of s 2 o, h, and the corresponding Ẑ that gives the minimum value of the ABIC, can be obtained. This can be considered as the optimised selection of these parameters. 2.3 Settlement prediction model The basicmodel used for settlement prediction in this paper is the first-order autoregressive model proposed by Asaoka (1978). The model is applicable for one-dimensional consolidation and is used for predicting the primary consolidation settlement based on the previously observed settlement data, as follows: S k b 0 b 1 S o k (20) where S k observed settlement at kth step of observation, b 0 and b 1 constant parameters for the Asaoka s model, and o k observation error. It is assumed that o k follows N(0, s o ), where s o denotes standard deviation of the observation error. In practice, one of the most required pieces of information is the final settlement (S f ). This can be simply estimated based on the parameters of Asaoka s model. By defining the final state as the state at which the settlement stops increasing with time (e.g. the stable state, at k 0), it can be concluded that, at this state, S k S S f, where S f refers to the final settlement. By substituting this in to Equation (20) and theoretically assuming o k 0, we have S f b 0 b 1 S f. Thus, the final settlement j

6 178 P. Rungbanaphan et al. S f b 0 (21) 1 b 1 Suppose that the consolidation settlement at the nth observation point, x 1, x 2,..., x n, has been sequentially observed at discrete time step k, k 0, 1,..., K. Based on Asaoka s model, the components of the observation model equations given in section can be defined as follows: Ẑ[bˆ 1(x 1 ); bˆ 1(x 2 ); ; bˆ 1(x m ); bˆ 0(x 1 ); bˆ 0(x 2 ); ; bˆ 0(x m )] T (22) Y k [S k (x 1 ); S k (x 2 ); ; S k (x n )] T (23) 2 3 S (x 1 ) 0 n : M k 4 :: 0 n;mn n I n;n 0 n;mn 5 0 S (x n ) n (24) where I n,n denotes an nn unit matrix. Note that, in this case, the total number of unknown parameters (P) is 2, while ẑ 1 and ẑ 2 are represented by ˆb 1 and ˆb 0 ; respectively. In the same way, the components of the prior information model equations given in section can be defined as follows: Z 0 [ ˆb 1;0 (x 1 ); ˆb 1;0 (x 2 ); ; ˆb 1;0 (x m ); ˆb 0;0 (x 1 ); ˆb 0;0 (x 2 ); ; ˆb 0;0 (x m )] T (25) V Z s2 b1;0 V C 0 n;n 0 n;n s 2 b0;0 V (26) C where ˆb 1;0 (x i ) and ˆb 0;0 (x i ) denote the prior means at observation point x i of b 1 and b 0, respectively. s 2 b1,0 and s 2 b 0, 0 represent the prior variances of b 1 and b 0, respectively. 0 n,n denotes an nn zero matrix. From the above definitions, it is clear that the spatial correlation of soil properties is included in the form of the spatial correlation of b 1 and b 0 instead of that of settlements. The authors believe that this is the most suitable way to introduce the spatial correlation structure to the settlement prediction model owing to the fact that the physical correlation of ground settlement actually results from the spatial correlation of soil properties. 3. Simulation experiments 3.1 Random field generation by frequency-domain technique To investigate the performance of the proposed approach, a two-dimensional random field of the model parameters is generated based on the assumed mean, variance, and auto-correlation distance. The observed settlement data is then calculated by Equation (20), using the generated parameters and the assumed standard deviation of the observation error, s o. It should be emphasised that the generated settlement data simulates the field observation data of an area with a predetermined spatial correlation structure. Performing the spatial-temporal updating procedure previously stated in Section 2.1 based on the generated observation data, the statistical inferences of the model parameters at each point can be back-calculated. A comparison of these inferences with the simulated ones, namely the true values, reveals the efficiency of the procedure. Various techniques have been proposed by several authors for random field generation, e.g. the turning bands method (Matheron 1973), the frequency domain technique (Shinozuka 1971, Shinozuka and Jan 1972), and the local average subdivision method (Fenton and Vanmarcke 1990). The frequency domain technique is chosen for this study to avoid the streaking problem which is found in the turning bands method, and implementing difficulties which are common issues for the local average subdivision method (Fenton 1994). This technique concentrates on the spectral density function (SDF) of the process, which is defined as the Fourier transform of the autocorrelation function. For an exponential type autocorrelation function, it can be proved that the SDF 1 S(v 1 ; v 2 ) 1 2ph h 2 (v2 1 v2 2 ) 3 =2 (27) which is a function of the frequency domain, v 1 and v 2. Assuming that the power of the employed SDF is negligible outside the interval [v 1,0,v 1,0 ] and [v 2,0,v 2,0 ], the simulated stationary Gaussian random field at any coordinate (x, y) can be expressed as the following series of cosine functions X(x; y) XM 2 X M 1 j1 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2S(v 1j ; v 2k )Dv 1 ; Dv 2 cos(v 1j xv 2k yf jk ) (28) where Dv 1 2v 1,0 /M 1, Dv 2 2v 2, 0 /M 2, v 1j v 1,0 (j1/2)dv 1, v 2k v 2,0 (/2)Dv 2, and f jk random phase angles which uniformly and independently distribute in the interval (0,2p). M 1 and M 2 are the number of equally divided intervals of the range [v 1,0,v 1,0 ] and [v 2,0,v 2,0 ], respectively. Care must be taken when selecting these ranges and discretisation intervals to ensure that the spectral density function is adequately approximated. Based on the above equations, a realisation of a random parameter with the specified mean and standard deviation at any coordinate (x, y) can be calculated by considering X(x, y) as a realisation of

7 standard normal distribution, i.e. X(x, y) follows N(0,1). Georisk Improvement of the estimation by considering spatial correlation structure A series of simulation experiments was performed based on the aforementioned procedure. For simulation of the model parameters, it is assumed that the mean and standard deviation of the random field of b 1 are and and those of b 0 are 6.94 (cm) and 0.59 (cm), respectively. These values are chosen from the data presented by Asaoka (1978) based on the observations of Kobe Port No. 3. It should be noted that, based on Equation (21), the first-order approximate mean and standard deviation (Ang & Tang 2007) of the final settlement, S f, are 332 (cm) and 53 (cm), respectively, assuming statistical independency between b 1 and b 0. This level of uncertainty for final settlement estimation is considered to be common in engineering practice. The initial settlement (S 0 ) is set as zero for every observation point. For the current study, the observation error (s o ) is assumed to be 1.0 (cm). By assigning the desired values of auto-correlation distance (h), random values of the model parameters together with the observed settlements at each observation point can be generated, as described in Section 3.1. To investigate the effect of sampling size, three different layouts of observation plans, with 16, 36, and 64 observation points (n), are set. All of these are arranged in a square grid pattern with even spacing of s and a total width of L, as shown in Figure 1. Based on the generated observation data, the procedure proposed in Section 2.1 is performed by assuming that the prior mean and standard deviation of the model parameters (see Equations (25) and (26)) are equal to those used for the data generation, i.e. ˆb 1;0 (x 1 )/ ˆb 1;0 (x 2 ).../ ˆb 1;0 (x m )0.9791, s b1, , and ˆb 0;0 (x 1 )/ ˆb 0;0 (x 2 ).../ ˆb 0;0 (x m )6.94 (cm), s b0, (cm). In practice, these statistics can be chosen based on field observation data, laboratory test results, or even judgement of an experienced engineer. For example, one can estimate the value of b 1 using coefficient of consolidation (C v ), which can be obtained from consolidation test, based on the relationship between these two parameters (Asaoka 1978). In this case, the auto-correlation distance and the observation error are assigned the same values as those used for generating the simulated data, namely the true values, in order to focus on the effect of considering spatial correlation. In other words, the model selection process presented in Section 2.2 is not included in the calculations in this section. Figure 1. Layout of observation plans. To examine the advantages of considering spatial correlation structure, the Bayesian estimation, using the observed settlement of an observation point to estimate the model parameters at that point itself, i.e. ignoring spatial correlation structure, is also performed based on the same conditions. In fact, the results of the calculations in the case with considering spatial correlation will converge to those of this case if very short spatial correlation distance is assumed. The estimations based on these two different conditions are compared and presented in Figure 2, Figure 3 and Table 1. Figure 2 demonstrates the stepwise updated estimates of the model parameters and the final settlement at point A in Figure 1 for n 36 and s/h 0.5, i.e. the spacing of the observation points is half of auto-correlation distance, based on the generated settlement data from only one time simulation. The estimated values and true values of final settlement (S f ) are calculated based on Equation (21) using the corresponding values of the model parameters (b 1, b 0 ). It can be seen that, with consideration of spatial correlation, the estimation is improved, i.e. closer to the true values. To confirm this conclusion, different sets of model parameters are randomly generated 100 times (N sim 100) and the estimation errors at selected time steps are calculated based on the estimated values of all observation points and all simulations. These errors are represented by the term

8 180 P. Rungbanaphan et al. Figure 2. Stepwise updating for estimation of the model parameters and final settlement at point A (see Figure 1) based on one random simulation, for n36 and s/h0.5: (a) b 1 estimation; (b) b 0 estimation; (c) final settlement estimation. Figure 3. Comparison of estimation error between the cases with considering and ignoring spatial correlation at different time factor, for n36 and s/h0.5: (a) b 1 estimation; (b) b 0 estimation; (c) final settlement estimation. of mean and standard deviation (SD) of the error ratio, E r, which is defined as E r ln Xest X true (29) where X est and X true denote the estimated value and true value, respectively, of the parameter to be estimated at each point for each simulation. Note that, because only estimation at the observation points is considered in this case, the total number of estimated values used for calculation of mean and standard deviation of E r is nn sim, i.e In this study, it is found that 100 simulations are sufficient for obtaining stable statistics for the results. Figure 3 illustrates the plots of the mean and standard deviation of E r of the model parameter and final settlement estimation against time factor, T v, for n36 and s/h 0.5. The time factor is chosen to represent the observation time in place of the observation time step because it provides more information relating to the degree of consolidation. The time factor at a specific time step can be calculated based on the relationship between b 1 and the observation time interval derived from Asaoka s formulation (Asaoka, 1978) and the approximated

9 Georisk 181 one-dimensional consolidation equation. The equation is given as DT v 4ð1 b 1 Þ (30) p 2 b 1 where DT v denotes the time factor interval which corresponds to the constant time interval between observations. T v can then be obtained by multiplying DT v by the time step. b 1 is assumed to be the mean of the random field used in data generation process, i.e. b Figure 3 clearly shows that the standard deviations of the error ratio, E r, for the cases of considering the spatial correlation structure are lower than those of ignoring spatial correlation structure, regardless of the observation time. This confirms that the estimation can be improved by taking into account the spatial correlation structure in terms of Table 1. Comparison of estimation error between the cases with considering and ignoring spatial correlation, for estimation at T v 0.164: (a) b 1 estimation; (b) b 0 estimation; (c) final settlement estimation (a) b 1 estimation E r of b 1 estimation Ignoring spatial corr. Considering spatial corr. n s/h Mean SD (1) Mean SD (2) Improvement a (%) E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E a Improvement (%)[(1)(2)]100/(1) (b) b 0 estimation E r of b 0 estimation Ignoring spatial corr. Considering spatial corr. n s/h Mean SD (1) Mean SD (2) Improvement a (%) E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E a Improvement (%)[(1)(2)]100/(1)

10 182 P. Rungbanaphan et al. Table 1. (continued). (c) final settlement estimation E r of S f estimation Ignoring spatial corr. Considering spatial corr. n s/h Mean SD (1) Mean SD (2) Improvement a (%) E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E a Improvement (%)[(1)(2)]100/(1) reduction of the estimation error. Moreover, with insignificantly low values of the means of E r, it can be concluded that the bias of the estimation is negligible. These trends are the same for both model parameters and the final settlement estimation. It should be noted that E r 0.1 or 0.1 imply the X est /X true ratios of about 1.1 or 0.9, respectively. To investigate the sensitivity of this improvement for different soil and observation conditions, the same calculations for several sampling sizes (n), and ratios of observation spacing to auto-correlation distance (s/h) are performed. Then, the mean and standard deviation of the estimation errors at the 20th time step, which corresponds to T v of 0.164, are calculated and summarised in Table 1. The improvement columns in this table present the reduction (in percent) of the standard deviation of E r by considering spatial correlation structure, compared with the case of ignoring spatial correlation structure. This value represents the level of improvement of parameter estimation by taking into account the spatial correlation structure. It can be seen from Table 1 that the improvement values increase with the reduction of the s/h ratio. This leads us to conclude that a stronger spatial correlation gives a more accurate estimation. According to the simulations, significant improvement seems to be found when the spacing of the observation points is shorter than half of the auto-correlation distance. On the other hand, for the case that the spatial correlation distance is relatively short comparing to the observation spacing, i.e. s/h ] 0.5, enlarging the sampling size with constant spatial correlation structure does not greatly improve the accuracy of the estimation. This result can be expected because, for the site with relatively weak spatial correlation, it is only neighbouring observation which contributes to the improvement of the estimation. These conclusions are the same for both model parameter and final settlement estimations. The biases of the estimation, which are reflected by the means of E r, are relatively low for all estimations. 3.3 Estimation of auto-correlation distance and observation error In the previous section, the true values of autocorrelation distance (h) and standard deviation of the observation error (s o ) are assumed to be known and are used in the estimation procedure. In practice, however, these parameters are unknown and needed to be estimated based on the observation data. It was proposed in Section 2.2 that these parameters can be appropriately selected by an optimization procedure based on Akaike s Bayesian Information Criterion. By performing several numerical experiments, the statistical inferences of the error ratio, E r, of this estimation for several conditions can also be investigated. Table 2 summarises the mean and standard deviation of the E r of auto-correlation distance and observation error estimation for sampling size (n) of 16, 36 and 64 (see Figure 1) at the 20th time step (T v 0.164). L is the total width of the group of observation points as shown in Figure 1. The number

11 Georisk 183 Table 2. Estimation error of auto-correlation distance and standard deviation of observation error, for estimation at T v E r of h estimation E r of s o estimation n L/h s/h Mean SD Mean SD of simulations for each trial is 100. The other parameters, such as ˆb 1;0 ; ˆb 0;0 ; s b1,0, s b0,0, s o, and S 0, are assigned the same values as those stated in Section 3.2. Table 2 clearly shows that the error of h estimation is much higher than that of s o estimation. With the standard deviation of E r below 0.04, it is concluded that s o can be accurately estimated by the proposed approach. Judging from the high positive values of the means of E r, a significant level of bias is found in h estimation. However, the error of h estimation tends to reduce with an increasing L/h ratio. Increasing the number of observation points does not greatly improve the accuracy of h estimation in this case. In addition, it should be noted that Table 2 shows only the estimation errors at the 20 th time step. Any estimation at a later stage can give a lower level of error owing to a larger amount of observation data included in the calculation. With the significant error of h estimation, the sensitivity of the settlement predictions to this error is of interest. This will be investigated and discussed in the next section. 3.4 Estimation of final settlement at an arbitrary location As mentioned previously, one of the advantages of the proposed method is its ability to estimate the settlement at any arbitrary location and at any arbitrary time. It is shown in section 2.1 that the components of the unknown parameters at any unobserved points, i.e. x n1, x n2,..., x m, are included in the formulation and the estimates of the parameters at these points will be calculated, together with those at the observation points, using the optimisation process based on a Bayesian approach. Then, the final settlement at these unobserved points can be predicted using Equation (21). To investigate the level of error for this prediction, a series of numerical examples was performed, the results of which are shown and discussed in this section. Figure 4 shows the plan of the observation points and the location of the points to be considered for settlement prediction. Several calculations are performed based on different values of the s/h ratio and the observation period (T v ), the results of which are summarised in Table 3. The number of simulations for each calculation is 100. The values of other parameters are also the same as those assigned in Section 3.2. Table 3 summarises the standard deviation of the estimation error for the final settlement, S f. The estimation error is also represented by the error ratio, defined in Equation (29), with the true values at an arbitrary point determined by the simulated model parameters at that point based on the same random field with those of the observation points. The means of E r are also found to be negligible in this case; therefore, they are not shown in this table. Concerning the error of h estimation as discussed in the previous section, Table 3 also shows the comparison between the estimation using the true value of both h and s o (Case A) and that using the estimated values of these parameters (Case B). The advantages of including the spatial correlation structure into the settlement estimation can clearly be seen from Table 3. For the site that the spatial correlation of the soil parameters is relatively strong, i.e. s/h 0.25, the final settlement can be predicted with a similar level of accuracy at the point located within the group of observation points or within the

12 184 P. Rungbanaphan et al. Figure 4. Observation plan and locations of the points to be estimated. length of auto-correlation distance around the group, i.e. at points 1, 2, and 3. This level of accuracy will be reduced with the increase of the distance from the group of observation points to the point to be considered. On the other hands, for the site that the soil parameters tend to be independent, i.e. s/h5, the errors of the settlement estimation by the proposed method are similar, regardless of the locations. The difference between the errors for the estimations at the earlier stage, i.e. at T v (the 20th time step), and at the later stage, i.e. at T v (the 50th time step), is noticeable only for the case that the spatial correlation is strong and, especially, at the points within the range of spatial correlation distance. These results emphasise the merit of considering spatial correlation structure for the local estimation, especially, when the soil parameters are strongly correlated in space, which is quite usual. Furthermore, the estimation errors for Case A and Case B are similar in any conditions. This confirms the insensibility of the proposed approach with the value of auto-correlation distance, which makes the approach practical even though the true value of the auto-correlation distance is, in some cases, difficult to be obtained. Table 3. Estimation error of final settlement at several unobserved points (point 1 to 5 in Fig. 4), for n36 Standard deviation (SD) of E r of S f estimation at T v at T v s/h L/h Point Case A a Case B b Case A a Case B b a Case A refers to the case which the calculations are performed based on the true values of both h and s o b Case B refers to the case which the calculations are performed based on the estimated values of both h and s o

13 Georisk Conclusion A systematic approach for spatial-temporal prediction of consolidation settlement based on Asaoka s method is proposed. Both prior information of the settlement model parameters and the observed settlement are used for settlement estimation using Bayesian approach. The spatial correlation structure of soil properties is introduced to the settlement prediction model through the spatial correlation of the model parameters. On this basis, an approach for estimating the auto-correlation distance and observation error using Akaike s Bayesian Information Criterion (ABIC) is also proposed. By performing a series of simulation examples for consolidation settlement prediction, the proposed spatial-temporal formulation is considered to have the following two main advantages: (1) The settlement prediction can be improved by considering the spatial correlation structure in term of reduction of the estimation error. According to the simulation, the improvement becomes more apparent when the spacing of the observation points is shorter than half of the auto-correlation distance. (2) The rational prediction of settlement at any arbitrary point becomes possible by introducing spatial correlation into the settlement prediction model. In addition, it was found that, while the standard deviation of observation error can be estimated accurately, the error of auto-correlation distance estimation is relatively high. The error will be even larger when the ratio of total width of the observations to the auto-correlation distance reduces. However, it was proved that the accuracy of settlement prediction is relatively insensitive to changes of auto-correlation distance. Therefore, it can be concluded that the proposed method is suitable for practical uses. References Akaike, H., Likelihood and Bayes procedure with discussion. In: Bernardo et al., ed. Bayesian Statistics. J. M. Valencia Spain: University Press, , Ang, A. H-S. and Tang, W.H., Probability Concepts in engineering emphasis on applications in civil & environmental engineering. 2nd ed. U.S.: John Wiley, Asaoka, A., Observational procedure of settlement prediction. Soil and Foundations, 18 (4), Baecher, G.B. and Christian, J.T., Reliability and statistics in geotechnical engineering. Chichester: John Wiley. Baecher, G.B. and Christian, J.T., Spatial variability and geotechnical reliability. Reliability-based design in geotechnical engineering. Oxon: Taylor & Francis, DeGroot, D.J. and Baecher, G.B., Estimating autocovariance of in-situ soil properties. Journal of Geotechnical Engineering, 119 (1), Fenton, G.A., Error evaluation of three random-field generators. Journal of Engineering Mechanics, ASCE, 120 (12), Fenton, G.A. and Griffiths, D.V., Probabilistic foundation settlement on spatially random soil. Journal of Geotechnical and Geoenvironmental Engineering, ASCE, 128 (5), Fenton, G.A. and Vanmarcke, E.H., Simulation of Random fields via local average subdivision. Journal of Engineering Mechanics, ASCE, 116 (8), Griffiths, D.V. and Fenton, G.A., Probabilisticslope stability by finite elements. Journal of Geotechnical and Geoenvironmental Engineering, ASCE, 130 (5), Honjo, Y. and Kashiwagi, N., Matching objective and subjective information in groundwater inverse analysis by Akaike s Bayesian Information criterion. Water Resources Research, 35 (2), Hoshiya, M. and Yoshida, I., Identification of conditional stochastic Gaussian field. Journal of Engineering Mechanics, ASCE, 122 (2), Krige, D.G., Two dimensional weighted moving averaging trend surfaces for ore evaluation. Journal of the South African Institute of Mining and Metallurgy, Proceedings of Symposium on Mathematics, Statistics and Computer Application for Ore Evaluation, 66, Lumb, P., Application of statistics in soil mechanics. In: Lee, I.K., editor. Soil mechanics: New horizons. London: Newnes-Butterworth, 44112, Matheron, G., The intrinsicrandom functions and their applications. Advances in Applied Probability, 5. Rungbanaphan, P., Honjo, Y., and Yoshida, I., May, Spatial-temporal prediction of secondary compression using random field theory. Submitted to Ge otechnique. Shinozuka, M., Simulation of multivariate and multidimensional random processes. Journal of the Acoustical Society of America, 49 (1, part 2), Shinozuka, M. and Jan, C.-M., Digital Simulation of random processes and its applications. Journal of Sound and Vibration, 25 (1), Sridharan, A., Murthy, N.S., and Prakash, K., Rectangular hyperbolar method of consolidation analysis. Ge otechnique, 37 (3), Tan, S.A., HyperbolicMethod For Settlements In Clays With Vertical Drains. Canadian Geotechnical Journal, 31, Vanmarcke, E.H., Probabilistic modeling of soil profiles. Journal of the Geotechnical Engineering Division. ASCE, 103 (GT11), Wackernagel, H., Multivariate geostatistics: An introduction with applications. 2nd ed. Germany: Springer-Verlag Berlin Heidelberg.