Premium
Predictive Accuracy of a Ground–Water Model — Lessons from a Postaudit
Author(s) -
Konikow Leonard F.
Publication year - 1986
Publication title -
groundwater
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.84
H-Index - 94
eISSN - 1745-6584
pISSN - 0017-467X
DOI - 10.1111/j.1745-6584.1986.tb00993.x
Subject(s) - hydrogeology , subsidence , standard deviation , calibration , hydrology (agriculture) , groundwater , geology , environmental science , statistics , mathematics , geotechnical engineering , structural basin , geomorphology
Hydrogeologic studies commonly include the development, calibration, and application of a deterministic simulation model. To help assess the value of using such models to make predictions, a postaudit was conducted on a previously studied area in the Salt River and lower Santa Cruz River basins in central Arizona. A deterministic, distributed‐parameter model of the ground‐water system in these alluvial basins was calibrated by Anderson (1968) using about 40 years of data (1923–64). The calibrated model was then used to predict future water‐level changes during the next 10 years (1965–74). Examination of actual water‐level changes in 77 wells from 1965–74 indicates a poor correlation between observed and predicted water‐level changes. The differences have a mean of 73 ft that is, predicted declines consistently exceeded those observed and a standard deviation of 47 ft. The bias in the predicted water‐level change can be accounted for by the large error in the assumed total pumpage during the prediction period. However, the spatial distribution of errors in predicted water‐level change does not correlate with the spatial distribution of errors in pumpage. Consequently, the lack of precision probably is not related only to errors in assumed pumpage, but may indicate the presence of other sources of error in the model, such as the two‐dimensional representation of a three‐dimensional problem or the lack of consideration of land‐subsidence processes. This type of postaudit is a valuable method of verifying a model, and an evaluation of predictive errors can provide an increased understanding of the system and aid in assessing the value of undertaking development of a revised model.