Model Limitations


Groundwater flow models are necessarily simplified mathematical representations of complex natural systems. Because of this, there are limits to the accuracy with which groundwater systems can be simulated. These limitations must be known when using models and interpreting model results.


There are many sources of error and uncertainty in models. Model error commonly stems from practical limitations of grid spacing, time discretization, parameter structure, insufficient calibration data, and the effects of processes not simulated by the model. These factors, along with unavoidable error in observations, result in uncertainty in model predictions.


Specific sources of uncertainty in the upper Klamath Basin regional model include grid spacing and parameter structure. The 2,500 ft by 2,500 ft grid spacing of the upper Klamath Basin regional groundwater model limits its ability to simulate conditions on smaller spatial scales. For example, because heads are averaged over areas of roughly a quarter square mile, drawdown in response to pumping wells at distances smaller than about 2,500 ft cannot easily be simulated. Because of the vertical discretizaton, conditions such as head changes due to pumping or other stresses are similarly averaged over vertical distances, limiting the ability to simulate effects to specific strata. Because of the limited availability of subsurface geologic information, hydraulic conductivity is simulated as uniform over broad areas, as shown in figure 6, and does not reflect the true complexity of the geology. Other parameters, such as streambed and lakebed conductance, are also simplified because of the lack of information.


The formulation of streams in a manner that only simulates groundwater discharge to streams and not stream leakage to the aquifer system is another limitation and potential source of uncertainty. In general, groundwater/surface-water interaction in the upper Klamath Basin is overwhelmingly dominated by groundwater discharge to streams. Seepage run data indicate leakage from streams to the aquifer system does not occur to a measureable degree along any of the major streams, and stream leakage is not a significant source of recharge. Should simulated head changes result in groundwater levels dropping below stream elevations in normally-gaining reaches, the model would not simulate the possible addition of water to the groundwater system from stream leakage. Should this occur, it is not likely to affect simulation results on a regional scale, but could affect simulated conditions near the affected stream reach.


Model error and uncertainty are not uniformly distributed. The model fit to observations is best where there are abundant data. Simulated conditions are more uncertain where data are sparse, such as unpopulated upland areas. 


The upper Klamath Basin regional groundwater model was intended to simulate groundwater flow over an 8,000 mi² area. Groundwater management issues and specific questions continued to evolve after the model was constructed. Therefore, the model is not necessarily optimized to address all current groundwater management questions. As demonstrated in previous sections of this report, however, the model does a good job of simulating the spatial distribution of hydraulic head throughout the basin as well as the distribution of groundwater discharge to the stream network. The model also does a reasonable job of simulating the broad response of the groundwater system to climate influences such as decadal drought cycles and longer term trends, and is able to simulate the effects of large-scale irrigation pumping as observed in and around the Bureau of Reclamation’s Klamath Project area. Therefore, even with the limitations described above, the model can be a useful tool for informing groundwater management in the basin.


The numerical results of the flow model have an associated, but un-quantified, uncertainty. While it is possible to quantify model prediction uncertainty, that analysis is not included within the scope of this report. A sense of model uncertainty will develop as conditions are monitored in the future and compared to model predictions. For these reasons, continued monitoring of hydrologic conditions in the basin is crucial. For practical purposes it is advisable to maintain an adaptive approach whereby management strategies can shift if observations differ from model predictions. Water managers should also be on the lookout for local anomalies resulting from geologic complexity not represented in the model. Model error and uncertainty can be reduced in the future by further model refinements and collection of new calibration data. 


First posted May 5, 2012