Estimation of block conductivities from hydrologically calibrated fracture networks - description of methodology and application to Romuvaara investigation area

This study looks at heterogeneity in hydraulic conductivity at Romuvaara site. It concentrates on the av-erage rock outside the deterministic fracture zones, especially in the deeper parts of the bedrock. A large number of stochastic fracture networks is generated based on geometrical data on fracture ge-ometry from the site. The hydraulic properties of the fractures are determined by calibrating the networks against well test data. The calibration is done by starting from an initial estimate for fracture transmissiv-ity distribution based on 2 m interval flow meter data, simulating the 10 m constant head injection test behaviour in a number of fracture network realisations and comparing the simulated well tests statistics to the measured ones. A large number of possible combinations of mean and standard deviation of fracture transmissivities are tested and the goodness-of-fit between the measured and simulated results determined by means of the bootstrapping method. As the result, a range of acceptable fracture trans-missivity distribution parameters is obtained. In the accepted range, the mean of log transmissivity varies between -13.9 and -15.3 and standard deviation between 4.0 and 3.2, with increase in standard devia-tion compensating for decrease in mean. The effect of spatial autocorrelation was not simulated. The variogram analysis did, however, give indications that an autocorrelation range of the order of 10 m might be realistic for the present data. Based on the calibrated fracture networks, equivalent continuum conductivities of the calibrated 30 m x 30 m x 30 m conductivity blocks were determined. For each realisation, three sets of simulations was carried out with the main gradient in x, y and z directions, respectively. Based on these results the com-ponents of conductivity tensor were determined. Such data can be used e.g. for stochastic continuum type Monte Carlo simulations with larger scale models. The hydraulic conductivities in the direction of the imposed gradient (Kg) - determined based on flow in and out of the opposite ends of the simulation cube - differed significantly in most realisations. This means that the Darcy flux in the direction of question differs with location inside the cube. In such a case the behaviour of the cube cannot be well presented with a continuum tensor. A finer rotation angle could be used to see whether the discontinuity in Kg-values is due to the sparse rotation density. The fact that the non-continuity was observed in most realisations indicates, however, that the result reflects the be-haviour of the rock in question rather than an 'unfortunate' selection of simulation flow direction. Due to the spatially varying Darcy flux components inside the cube, the least square approximation used for determining the continuum conductivity tensors is not the best estimate for the present data. Alterna-tive averaging methods were, however, not investigated as it appears the at the continuum approxima-tion may not be a good representation for the data and scale anyway. Regardless of the validity of the continuum approximation, the results of the 30 m fracture network blocks give us valuable information on the flow distribution at this scale. Firstly, the results can be used to evaluate possible anisotropy, which cannot be seen from borehole well test data alone. Inspection of the simulated results showed that there appear to be no anisotropic effects. Secondly, the surface area conductivities can be compared to well test conductivities of the same scale. This comparison indicates that the conductivities determined from the well tests are clearly lower. The result is preliminary in nature, and may be case- and data-specific, but it may also be more general and related to e.g. differences in flow and observation geometry between the two situations.