These impressive gains in hardware and algorithms have enabled scientists to numerically model physical phenomena that only a few years ago were beyond the limits of our computational resources. In the area of water resources research, for example, transient, three dimensional groundwater-contaminant models incorporating both saturated and partially saturated flow and multiple contaminants involving millions of variables over many thousands of time steps are today considered leading edge, but fairly routine.

From this increase in computing power and improved algorithms, a new generation of data hungry models has emerged. For example, in a contaminant transport model of a three dimensional nonhomogeneous aquifer involving a million finite difference blocks, the task of setting values of conductivity, specific yield, transmissivity, initial heads and initial concentrations presents the modeler with upwards of 15 million data values to set---a daunting task indeed! If the region to be modeled requires an irregular finite element or finite difference mesh, the data entry problem is compounded. If the model is to be calibrated against some observations, the data entry loop will have to be performed many times.