The simulation was run at steady-state; stress on the system was provided by the water supply wells, using pumping rates provided by the City of Hays (Table 2). Water input to the system occurred through recharge to the exposed portions of all three layers, and through constant-head nodes in Layers 1 and 2 at the NE and SW grid corners (representing along-valley underflow from outside the grid, see Fig. 13). Layer conductivities and recharge distribution and magnitude were adjusted during model calibration to produce the best fit (smallest RMSE, or Root-Mean Square Error [Anderson and Woessner, 1992] compared to observed water levels within Hays city limit) and distribution of errors in head residual (difference between modeled head and observed head at the time of drilling for boreholes within Hays city limits). The average head residual for the best-fit model within the 1987 Hays City limit was 2.11 meters, with a standard deviation (or RMSE) of 5.97 m. Models with lower mean residual (as low as 0.4 m) were rejected because they exhibited excessive residual (-10 m) in northeast Hays, in the vicinity of Chetolah Creek. A major difficulty with the RSKERC Hays well database is that it includes observations made by a variety of workers over a period of 50 years. As such it contains a significant amount of variability that can be attributed to climatic and seasonal changes, and the cumulative effect of pumping from the alluvial aquifers. Some synoptic (short time period) observations have been made [Schmidt, 1995] but are too limited in geographic distribution to provide a suitable calibration dataset. In summary, the model was calibrated primarily on layer hydraulic conductivity, which is effectively a calibration on cell transmissivity, since layer thickness was variable based on interpolated well logs; however, inter-cell variation in transmissivity remained fixed by the layer thickness. The model was calibrated secondarily on total recharge, with inter-cell variation in recharge fixed during calibration (specifying a factor of 8 increase in recharge for cells adjacent to the Kc contact). Final model adjustment was made by increasing the recharge adjustment factor in cells adjacent to the Kc contact along Chetolah Creek. Lower cell transmissivities in northeastern Hays could have achieved the same quality of fit with observed heads, but were not indicated by well tests or the lithologic model.