EFFECTS OF HIGH VARIANCE OF FRACTURE TRANSMISSIVITY ON TRANSPORT AND SORPTION AT DIFFERENT SCALES IN A DISCRETE MODEL FOR FRACTURED ROCKS

A three-dimensional (3-D) variable-aperture fracture network model for flow and transport in fractured crystalline rocks has been applied to study the effects of large variability in fracture transmissivity on non-sorbing and sorbing tracer transport, and scale effects in transpo rt distance. The variable-aperture character of the fractures is intro duced into a 3-D network model through a library of single-fracture pe rmeabilities and associated particle transport residence time spectra. Sorption onto the fracture walls is added by a mathematical model for linear sorption. The resulting variable-aperture fracture network mod el, VAPFRAC, can handle flow and transport from single-fracture scale to the multiple-fracture scale. The model produces multi-peak transpor t breakthrough curves even for relatively moderate values of the fract ure transmissivity variance. These breakthrough curves display dispers ion on two different scales in the same way as has been observed in se veral field experiments conducted in crystalline rocks. The multi-peak structure is due to so-called channeling. For high values of the frac ture transmissivity variance the solute transport is unevenly distribu ted and the channeling effects are more prominent. The effect of linea r sorption is not just a simple translation in mean residence time as in a homogeneous medium. The dispersion characteristics of the breakth rough curves also change when linear sorption is included. The degree of the change depends strongly on the fracture transmissivity variance , as does the translation, In particular, with a high fracture transmi ssivity variance the translation in mean residence time due to sorptio n is significantly smaller compared to the cases with a low fracture t ransmissivity variance. Finally, the high variability in the model out put data suggests that extrapolation of results from a particular trac er experiment will be highly uncertain.