GNGTS 2016 - Atti del 35° Convegno Nazionale

594 GNGTS 2016 S essione 3.3 brine. Furthermore, the grains consist of a mixture of quartz and clay. Porosity and permeability vary continuously in the domain; the latter ranging from 10% in the shales bounding the aquifer to 35 % in the porous sands in the anticlinal structure; the former from 10 -17 m 2 to 10 -12 m 2 . The geological model, together with the distribution of the porosity, permeability and clay content can be seen in Fig. 1. The domain was computed on a regular grid of 315x315 nodes, which implies a node size of 2.5 m. This had to modified for the fluid flow simulation for numerical stability. In the z direction, the domain was divided in 244 nodes, with a spacing of 2 m in the area of the complex aquifer and 8 m where the medium consists of homogenous shale. In order to minimize the effects of the boundary conditions and enhance the numerical stability of the fluid flow simulation, the domain has been extended approximately 25 km to the east and west. Then, the domain was divided in 4 different zones; the first is the area of the aquifer, from 0 to 500 m, in which nodes are 2 m wide. In the other areas, the boundaries of which are located at 0.5, 6 and 20 km from the well, node size is let increase logarithmically. Therefore, node space increases from 2 m close to the well to 20 km at the furthermost node. This allows to have the necessary high resolution close the well and far enough boundary conditions using only 247 nodes. Results of the fluid flow simulation were then interpolated using a natural neighbor interpo- lation algorithm to match the original 315x315 grid, on which the forward seismic modeling was launched. Fluid-flow simulation. The STOMP-CO2 (White and Freedman, 2007), developed at the Pacific Northwest National Laboratories, commercial software is used to model the fluid tran- sport equation in isothermal mode. The code solves the mass balance equation for the non wetting (CO 2 ) and wetting phase fluid (brine) and for the dissolved salt. The fluxes are computed with the Darcy equation, considering two different relative per- meabilities for each of the two fluids. Since we are dealing with a two phase flow, the interfacial forces between the wetting and non-wetting phase fluids will give place to capillary pressure. We use a Van Genuchten (van Genuchten, 1980) relation for both capillary pressure and relative permeability curves. In this simulation STOMP considers capillary trapping mechanisms due to hysteresis of the relative permeability curve. Temperature was considered constant at 37°C for the entire simulation time, the aquifer was considered to be fully saturated with brine before the onset of the injection. Initial pressure is considered to be hydrostatic with a brine density of 1030 kg/m 3 (~15 MPa at the bottom of the formation), while salinity is set to 3.2%. Zero flux boundary conditions are set at all boundaries, even though the same results can be obtained if Dirichlet boundary conditions are set, given their distance from the injection well. Van Genuchten’s pore size distribution parameter m is set to 0.4 �� ��� ������ ������� �� on the entire domain. We consider the Mualem approximation for which n=1-1/m=1.67. Entry capillary head α is set to 0.16 in the shales and 2.84 in the sandstone. Grain density is set uniformly to 2600 kg/m 3 , its compressibility to 4.5x10 -10 Pa -1 and a Mil- lington and Quirk tortuosity function is used. We inject 0.31 kg/s of pure CO 2 for 2 years in a well located at point (501 m,-1030 m) of the formation. STOMP is asked to give outputs at 0.5, 1 ,1.5 and 2 years after start of injection. In Figs. 2a and 2b we show the distribution of CO 2 saturation and pressure and density after 1 year injection respectively. It has to be stressed that we incurred in strong numerical instabilities, convergence failures and crashes while performing this simulation. For this reason, we compared the results with those obtained with TOUGH2-ECO2N software used at INGV Catania. This other simulator showed similar instabilities, but finally we found a grid and a set of parameters with which both simulator showed similar results.