Development of integrated earthquake simulation system for Istanbul

As explained previously, advanced numerical simulations may be used for predicting earthquake hazards and disasters. Large-scale numerical simulations may be carried out due to the developed computer science and technology. In this process, all phases of a possible earthquake should be simulated. Numerical simulation should give researchers further information about estimation of variability in earthquake disasters due to different earthquake scenarios. The amount and complexity of the data to be used for earthquake hazard and disaster simulations of large areas require the use of user-friendly and advanced graphical interface systems. With the advances in computer technologies, the computer-aided design (CAD) has been developed. MATLAB is one of the representatives of high-performance language for the CAD. It can be used for modeling, simulating, and analyzing dynamic systems. It supports linear and nonlinear systems, modeled in continuous and discrete time. Simulation is an interactive process, so the parameters may be altered, while the simulation is running and the system response may be immediately monitored.

A new user-friendly and graphical user interface-based system development for earthquake hazard and disaster simulation of urban areas has been started (Sahin 2014). The system is planned to include all phases of IES system. The final aim of the system is to simulate earthquake strong ground motion from source to site, structural damages, and human behavior during the earthquake. The long-term objective is simulation of emergency evacuation and short- and long-term social and economic recovery. It is possible to automatically construct a computer model for a city of some hundred kilometers scale using the latest geographical information system (GIS). It can be said that new IES system provides vital information to see the possible effects of future earthquake hazards and disasters.

High-performance computing enhancement was made, so that urban area models of higher fidelity were analyzed in shorter times and ensemble simulation which accounted for uncertainties in modeling the earthquake hazard and disaster processes could be made. The uncertainties in modeling the earthquake hazard and disaster processes are material or structural properties. It is not possible to remove these uncertainties, but we can account for it by providing a probabilistic distribution of seismic response carrying out ensemble computing. It should be noted that material property uncertainty is not significant for a structure if input ground motion is small. But the uncertainty provides wider range of responses if input ground motion is large and the structure has a chance of being damaged. Full-scale vibration tests will be carried out to eliminate structural uncertainties, and structural models will be developed.

It should be indicated that high-performance computing can be directly applied to the IES system developed on MATLAB by using Parallel Computing Toolbox. Multicore processors, GPUs, and computer clusters can be used to solve computationally and data-intensive problems. The applications developed in MATLAB can be parallelized without CUDA or MPI programming. High-level constructs parallel for loops, special array types, and parallelized numerical algorithms can be used for parallelization (Parallel Computing Toolbox™ User’s Guide 2015).

In IES system, physical modeling is used and more detailed modeling decreases the uncertainties. The soil models are directly developed from site observations, and seismic behavior of soil models can be observed. The uncertainties in soil models and hazard simulations have been eliminated significantly by integrating all soil test data such as borehole, ReMi, and PS Log. The integration algorithms are used for producing more accurate soil models. The mesh grid intervals could be shortened after considering all data sets with GIS format, and more reliable soil models could be generated.

In building models, it could be possible to predict detailed 3D FEM models from the estimated multidegree-of-freedom (MDOF) models. Detailed time history analysis of thousands of buildings could be possible owing to high-performance computing. As indicated before, more detailed models decrease the uncertainties in modeling.

Automatic soil model generation and earthquake hazard simulation tool—SoVeLAB

The ground motion prediction is the first step of earthquake damage assessment. This analysis generally consists of two processes. In the first process, the ground motion on seismic bedrock is calculated. In the second process, the soil effects are evaluated by considering seismic site amplification (site response analysis). The surface ground motion can be predicted by considering both processes. Wave propagation in seismic bedrock is evaluated in the first process, and wave propagation in soil surface is evaluated in the second process.

The seismic bedrock motion can be considered with three ways: (1) converting observed rock outcrop motion into bedrock motion, (2) directly applying a seismic record or random record as a bedrock motion, and (3) obtaining bedrock motion from the source by applying wave propagation techniques.

The surface ground layer model may be constructed by using a set of geotechnical data, which include boring, PS logging, or passive seismic methods such as microtremor and refraction microtremor (ReMi). An algorithm is used to automatically construct three-dimensional soil model. In the proposed algorithm, interpolation and extrapolation between neighboring ReMi and boreholes are performed.

The collected field data are scattered over the study area, and it is required to get them gridded with respect to generated mesh. This mesh generation is automatically done by developed system. Automatic interpolation is utilized, and soil profile can be obtained at any point on the produced mesh. The mesh generation process is summarized in Fig. 2.

Soil profiles at any point on mesh are generated by determining each layer’s top and bottom elevation. Borehole data are utilized for determining the top and bottom elevations of the layers. Figure 3 shows the procedure used to combine soil profiles at borehole locations. In the developed system, the procedure is conducted in three-dimensional domain. The upper and lower boundaries of the soil layers are combined. If any layer is not encountered in a borehole log, the following soil layer from the stratigraphy is assigned in place of it. These modifications mean that every formation exists in all borehole logs, but in some of them the nonexisting formations have zero thickness.

After that process layer thicknesses at any point in the study can be calculated by only interpolation. The same procedure is applied for generating the velocity model. The soil and velocity models are then integrated. The velocity information at any depth of ReMi test locations is obtained, and the same procedure is used to combine them.

To generate soil model of a selected urban area from borehole and refraction microtremor test data, a new MATLAB code, named as SoVeLAB (Soil Velocity Laboratory), was developed. The flow diagram given in Fig. 4 represents the algorithm of this program. Firstly, SoVeLAB reads the area data which includes the boundary coordinates of the region being studied. At this step, SoVeLAB generates a uniformly gridded mesh between the defined boundary coordinates. Secondly, SoVeLAB reads the borehole and refraction microtremor test data from the GIS database. Then, SoVeLAB visualizes the altitude map of the study region by utilizing the x, y coordinates and altitude of site test location. Interpolating the borehole test data, SoVeLAB produces the primer soil model including layer boundaries, while interpolating the layer depths, SoVeLAB uses a filter mask to filter out unrealistic results. MATLAB Image Processing Toolbox filtering functions are used for median filtering. Median filtering seems more effective than convolution technique if it is aimed to simultaneously reduce noise and preserve edges. After generation of primer soil model, SoVeLAB interpolates refraction microtremor test data and generates the velocity model. Firstly, the study area is divided into small cell and the data sets are combined by linear interpolation. The shear wave velocities at each 1 m depth for each observation stations are determined up to the maximum depth reached. Then, the velocity values over the study area for each 1 m depth are interpolated. This procedure successfully generates the three-dimensional digital velocity model. To ensure that unrealistic data are thrown out, velocity model is also filtered. The system is filtered according to median value of neighboring matrix elements. According to the Vs value representing the engineering bedrock, SoVeLAB determines the bedrock depth and integrates bedrock layer with primer soil model and generates main soil model. Finally, SoVeLAB combines velocity model and soil model to use in site response analyses.

It should be noted that two-dimensional median filtering is used for eliminating unrealistic test data. Median filtering is a nonlinear filtering operation and widely used in image processing. It can be seen that a median filter is very effective in automatic soil model generation with great variety and great number of data.

It should be emphasized that many comprehensive studies were carried out to satisfy the needs of loss assessment processes in Istanbul after the destructive Marmara earthquakes affected urban areas of Marmara Region in 1999. Japan International Cooperation Agency (JICA) prepared a “Study on Disaster Mitigation/prevention in Istanbul Including Seismic Microzonation” (Pacific Consultants International/OYO Corporation 2002). After this study, Earthquake Master Plan for Istanbul (EMPI) has been prepared by Istanbul Metropolitan Municipality in corporation with four leading Turkish universities. Due to the need of rapid earthquake loss estimation systems for large-scale areas, some software systems have been developed suitable to Istanbul database. Earthquake Loss Estimation Routine (ELER) has been developed within NERIES project JRA3 workpackage for rapid estimation of earthquake shaking and losses throughout the Euro-Mediterranean region (Erdik et al. 2010; Hancilar et al. 2010). HAZTURK (MAEviz-Istanbul) software has been developed based on the MAEviz of Mid-America Earthquake Center to visualize the earthquake risk and its possible damage to structures and people (Karaman et al. 2008a). The applications of these tools mentioned here have been conducted for some parts of Istanbul (Karaman et al. 2008b; Sesetyan et al. 2011). Additionally, many comprehensive studies have been carried out in order to assess the seismic risks in Istanbul (Atakan et al. 2002; Erdik et al. 2003; Bal et al. 2008; Ansal et al. 2009). However, none of the studies carried out for Istanbul were able to provide an integrated system that generates physical models of urban areas including soil and structural systems. The proposed system automatically generates virtual city and simulates earthquake hazard and disaster.

This section presents examples of applying the developed system to Istanbul for prediction of earthquake hazards and disasters. The virtual city is a computer model of a real town, which is constructed in Zeytinburnu district of Istanbul. The size of the target area is 1000 m × 1000 m as shown in Fig. 5.

The virtual city is constructed by using the data stored in several GIS’s, which are available for Zeytinburnu district. The location of available borehole and ReMi measurements in Zeytinburnu district is presented in Fig. 6. These data are available in Istanbul Metropolitan Municipality (IMM) servers, and the models are developed by using the database in the IMM servers.

The initial three-dimensional (3D) soil model is automatically generated by interpolating the borehole data for the developed grid points. The boundaries of each layer are determined, and the initial parameters are assigned accordingly. The 3D velocity model of the selected area is also automatically generated by interpolating and extrapolating the observed ReMi data. The generated velocity model is filtered according to median value of neighboring matrix elements by using nonlinear median filtering.

The velocity distribution at the center of the model is shown in Fig. 7: The X–Z section at Y = 4,543,550 m and the Y–Z section at X = 408,250 m of the developed model are presented in this figure. It should be noted that “Z” represents the altitude of the soil model.

The developed initial soil model and velocity model are combined, and the final soil model is obtained as given in Fig. 8. The bedrock layer is determined from the velocity distribution. In this study the velocity of engineering bedrock is assumed as 760 m/s. The bedrock surface layer is presented in Fig. 9.

In the next step, the virtual city is automatically generated. In the virtual city, there are 417 residential buildings where a MDOF system is constructed for each building. Data for the structure type, floor number, land usage, and the location of residential buildings are extracted from GIS database. Data collected from GIS database are automatically loaded into the system, and MDOF system is constructed as shown in Fig. 10.

Input ground motion

Strong ground motion which hits the virtual city is simulated by using one-dimensional site response analysis. The developed soil model is divided into same sized mesh grids, and the strong ground motion at each grid point is calculated. Time series ground motion data recorded at Zeytinburnu station in 1999 Kocaeli Earthquake that struck northwestern Turkey is used as input for site response analysis. The geodetic position of the observation station is 28.9080° longitude and 40.7270° latitude as shown in Fig. 11. These seismic records are downloaded from Pacific Earthquake Engineering Research Center—PEER ground motion database (http://ngawest2.berkeley.edu/). The ellipsoidal coordinates converted into the Cartesian coordinates are introduced to developed system for conversion. The location of the observation station in Zeytinburnu district is also presented in the altitude map as shown in Fig. 11.

The soil profile of the observation station is automatically generated as shown in Fig. 12, and the measured records are converted into bedrock motion. The east–west and the north–south components of the observed record and produced bedrock motions are given in Figs. 13 and 14, respectively. The obtained bedrock motion is transferred to the selected area, and 1D site amplification is calculated for all grid points generated in the virtual city.

As an example of the synthesized seismic motion at the surface, the distribution of the maximum displacements, velocities, and accelerations for the east–west and north–south directions is given in Figs. 15, 16, 17. The maximum displacement in the north–south direction is 8.79 cm, and it is obtained at t = 52.54 s. As shown in this figure, the strong ground motion is far from being uniform, and some concentration of large ground motion is observed.

Seismic structure response

Linear time history analysis is carried out for 417 MDOF models of the residential buildings which are located in the virtual city. The simulated strong ground motion (SGM) at the site of each building is used as an input wave. Note that the MDOF models are isolated and no interactions among nearby buildings are considered as well as no soil–structure interaction. In Figs. 18, 19, 20, and 21, earthquake hazards and disasters in E–W and N–S directions, which are predicted for the virtual town, are presented as a snapshot of the perspective of the virtual city. In these figures, strong ground motion distribution in terms of peak ground displacement (PGD) and structure response of all residential buildings is presented for each 10 s. These images are the output of the visualization tool that is implemented in the recently developed IES system.

The maximum displacement of each MDOF model in E–W direction, N–S direction, and combined direction is presented in Figs. 22, 23, 24. The related soil motions in E–W direction and N–S direction at the same times are also given in Figs. 22, 23. As expected, the maximum displacement becomes larger for buildings which are higher than others. It can be observed from the analysis results that the buildings have not caused seismic amplification very much. The maximum displacement values of the top of the buildings and the soil surface are very close to each other.

The virtual town is constructed for the selected part of Zeytinburnu district, with MDOF models for 417 buildings. Topographical effects may cause concentration of ground motion and induces larger response of the MDOF models. Further studies including two- or three-dimensional soil model analysis are surely necessary to clarify structure damage induced by such topographical effects. It is the core functionality of new IES that makes it possible to evaluate possible damages in each residential building. For mode exclusive information about possible earthquake hazard and disaster which may happen in Istanbul, an urban area model needs to be extended to all Istanbul, by using available GIS. It is also necessary to employ a more sophisticated model for residential building if sufficient data are stored in GIS or reasonable guess of structure properties is allowed.

It should be emphasized that the initial version of the software module has been presented and now 1D site amplification analysis is ready. In the next stage of the project, we plan to develop 2D and 3D analysis modules. 3D soil models are generated for further studies which takes layer slope effects into account, since the studied area that has a smooth ground surface using 1D site amplification was very efficient and sufficient.

In this study, IES on MATLAB is introduced for Istanbul with ground motion and building response modules. An application is presented herein where the results are expressed in terms of spatial distribution of peak ground motion parameters as well as dynamic building responses. As of now, IES on MATLAB uses the same bedrock motion at all grid points and employs a deconvolved record measured in Istanbul during the 1999 Kocaeli earthquake as input to the site response analyses.

At the end of the analyses, it is seen that the displacement distribution of the virtual city is very limited although some tall buildings make very high displacements. Since the obtained ground motion distribution and peak values are not very high, no damage has been observed in the virtual city under the current simulation. The most active areas depending on the site response simulations have been determined, and the critical buildings that have maximum displacement during the earthquake motion are detected. The structural damage distribution in virtual city can be observed, and loss estimation can be made by using the produced soil and structural response data under earthquake records.

In current study, the structural models are automatically developed by assuming that all buildings comply with seismic design codes. For the case of Istanbul, buildings having severe irregularities (e.g., soft/weak story, discontinuous framing) and those that do not comply with seismic design codes, constitute the largest portion of the seismic risk. In future part of this study, site vibration tests will be made for hundreds of buildings and new models will be developed for Istanbul building inventory.

In its current situation, IES system on MATLAB does not yet include the wave propagation from the fault plane to each site of interest. At every site, however, one-dimensional site response analysis is already implemented to account for the local soil amplifications. As input to these site response analyses, either a Ricker pulse or an observed velocity time history is employed. The observed time history is measured on a rock site during a previous earthquake that occurred in Istanbul. However, in order to model the effects of fault rupture process of past earthquakes as well as scenario events, a simulation tool to model the wave propagation from the rupture plane to each site is required. Since one-dimensional site response is already included, the basic goal in this section is to select and apply a robust approach for IES on MATLAB to model the generation and propagation of the waves from the source to each site of interest. This way, the model will be complete and physical in terms of all of the source-to-structure components.

A hybrid model composed of both deterministic (Spudich and Xu 2003) and stochastic (e.g., Motazedian and Atkinson 2005; Askan et al. 2013) approaches for waveform modeling will be employed. Validation of our ground motion simulations will be performed against previous events in the region. Since there is no large event in Istanbul that occurred in the instrumental era, the records are obtained only from the small events. Finally, after validations, we will perform finite-fault simulations for potential large earthquakes in Istanbul in the final form of IES on MATLAB. Results will be presented as ground motions on free surface as well as building responses. Spatial distributions of ground and structural responses for different earthquake scenarios could be employed for a range of purposes including disaster mitigation and emergency management.