Laserfiche WebLink
Model calibration DRAFT 18 <br />One measure to evaluate the quality of the calibration is in terms of 'average residual <br />mean error'. This is the average absolute value of the difference between observed and <br />simulated water levels (i.e., differences may be positive or negative but the absolute <br />value is used in the mean). For this model, the average residual mean error is about <br />54 ft for all monitoring wells and about 64 ft the F/G sequence wells alone. Although <br />this value may initially seem high, it is reasonable given the extent of the regional model <br />and the data used for calibration. The model provides a simplified representation of <br />mountainous terrain where water levels vary both horizontally and vertically over <br />relatively short distances. The wells used to obtain water levels are completed across <br />multiple intervals in some cases, particularly the older wells installed by Utah <br />International. In addition, the model must combine numerous individual stratigraphic <br />layers into a fewer number of discrete model layers. As a result, it is common that the <br />average residual mean error is in the range calculated for a model of this type. <br />A second measure to evaluate the quality of the calibration is `bias'. Bias is determined <br />by calculating the average of the differences between simulated and observed water <br />levels at all the calibrations points. Differences are negative if the measured water level <br />is lower than the simulated water levels and are positive if the measured water level is <br />higher then the simulated water level. Bias provides an indication of whether simulated <br />water levels are generally higher or generally lower than measured water levels. For the <br />steady-state calibration, the calculated bias is -2.2 ft for the F/G sequence wells and - <br />11 ft for all wells. This indicates that although simulated water levels at specific <br />locations may not be accurate, the model simulates the general behavior of the <br />40 groundwater system very well. The bias is considered a better measure of calibration <br />for this model than the average residual mean. <br />The `residual standard deviation' of simulated compared to measured water levels is <br />another measure that can be used to evaluate the quality of the calibration. The <br />residual standard deviation is the ratio of the standard deviation of the differences <br />between the measured and simulated water levels (82 ft for the F/G sequence wells, <br />68 ft for all wells) to the total water range of water levels across the model area (331 ft <br />for both F/G sequence wells alone and all wells). In areas of complex geology and <br />mountainous terrain high residual standard deviation values are expected with a <br />regional model. In areas where the water table is relatively planar, and geology is <br />simple, the residual standard deviation would be expected to be small. The residual <br />standard deviation for the steady-state model is 0.25 for the F/G sequence wells and 0.2 <br />for all monitoring wells. These values are considered to be reasonable considering the <br />complexity of the geology, the nature of the terrain, and the variable completions of the <br />monitoring wells, which could not be simulated in detail given the resolution of the <br />model. <br />The internal mass balance of the calibrated steady-state model was also evaluated. <br />The difference between fluxes into and out of the model was less than 0.1 %, indicating <br />excellent internal consistency. <br />• <br />2572-R3 Colowyo Coal Company, L.P. <br />Water Management Consultants