UNCERT is an uncertainty analysis and geostatistical software package. It was developed for evaluating the inherent uncertainty in describing subsurface geology, hydraulic properties, and the migration of hazardous contaminants in groundwater flow systems. The package is well suited for evaluating hazardous waste sites and evaluating remediation methods, but the package also has many general modules which are usable by researchers from a wide range of disciplines. At the Colorado School of Mines, it was written for use by Geological Engineers interested in Groundwater, but many of the tools are being used by researchers from mining, mathematics, chemistry, and geophysics. Although the whole package may not be useful for you, examine the different modules; many are of general use and may serve your purposes.
UNCERT is a software package designed to aid hydrogeologists in using computers to simulate the distribution of materials and material properties in the subsurface, evaluate groundwater flow and contaminant transport, and design and evaluate alternative contaminant remediation designs.
This package is also designed so that users in other disciplines of research may find various portions of the package useful, and handles will be attached so that new software packages may be easily incorporated as new developments are made and the need arises.
There are a number of software modules associated within this package. These modules allow the modeler to 1) input raw field data or data from a pre-existing database, 2) analyze the data using classical statistics, 3) evaluate trends, 4) evaluate the data using geostatistical techniques such as semivariogram analysis, various kriging techniques (simple, ordinary, indicator, and Bayesian), and stochastic simulation. When the data are analyzed, or when data are prepared from other sources, graphical tools are available to view the results in two-, two-and-a-half-, and three-dimensions. Once the spatial variation of materials has been determined, tools will be available to automatically generate finite-difference grids for groundwater flow and contaminant transport models such as MODFLOW and MT3D, 5) run these models, and 6) evaluate both the results of individual runs, as well as the composite results of multiple model simulations.
Development of this toolkit is important because of the inherent difficulty in describing the subsurface. For any given set of data there are a multitude of possible interpretations of the subsurface which honor the raw data. To evaluate the alternatives manually would take considerable time and still only a small portion of the possibilities could be evaluated. This is true even when the subsurface configuration is relatively simple. In Figure 1.1, for example, where there are only two materials present (sand and silt), three alternate interpretations are suggested based on data from two wells. Each description honors the raw data exactly, but groundwater flow and contaminant transport through each would vary significantly. In more complicated situations where materials grade into one another (Figure 1.2) alternative interpretations are much more complicated and varied, but still honor the data. In order to evaluate this inherent uncertainty, computers can be used to create the multiple alternative realizations of the subsurface. The process can be forced to honor the hard data (well logs, etc.) by using indicator kriging techniques, and incorporate more uncertain data (soft data - data with a range of uncertainty, e.g. seismic information, geophysical well logs, expert opinion) through Bayesian kriging. By automating this process, much of the uncertainty can be characterized with comparatively little time invested by the hydrogeologist. Once multiple realizations are created, groundwater flow and contaminant transport models can be executed to compare modeled and field conditions. When a model response clearly doesn't match field observations, this possible subsurface configuration can be disregarded; of the remaining realizations (invalidating 90% of the realizations might not be unreasonable) that appear reasonable, the distribution of contaminants may be evaluated, for the time already modeled, or for future conditions. Based on the results of flow and transport modeling in these remaining configurations, the probable locations of contamination may be identified. Also, the probable effectiveness of remediation facilities designed to contain the contamination, can be evaluated. A computer can evaluate only a limited number of realizations, but the number is so large, relative to that which can be accomplished manually, that a representative assessment of the reasonable alternatives will be realized.
This process is illustrated in Figures 1.3, 1.4 and 1.5. At this hypothetical site there are two leaking storage tanks. For simplicity we will assume a two-dimension system (i.e., lithology is constant with depth) and assume exploratory drilling has cored six clay holes and five gravel wells. One interpretation a hydrogeologist might make appears in Figure 1.3a. The material between wells is assumed to be fairly uniform; this yields several large clay zones separated by a gravel channel. Contaminants migrating through the subsurface might form a plume similar to that in Figure 1.3b over a given time interval. On the other hand, a less intuitive description of the subsurface is offered in Figure 1.4a. This realization exactly honors the data as in the first example, but is substantially different. In this case it was assumed that the spatial continuity of units was not as great, and as a result the contaminant plume would be significantly different (Figure 1.4b) over the same time interval. With no further data, both these solutions are equally probable, but a hydrologist probably would not equally consider both, and yet there are many more possibilities to be considered. This software package is designed to help identify these possibilities in a timely fashion. Finally, based on the multiple possibilities, maps can be made showing the probability that the contaminant plume will exceed a given concentration at a particular location at a given time (Figure 1.5). From this risk map, remediation facilities can be designed, and going through a similar process, the likely effectiveness of each design can be evaluated.
Figure 1.3a and 1.3b, and
Figure 1.4a and 1.4b, and
In this chapter several points have been made about the types of data that can be evaluated , and how they are integrated into the characterization of the subsurface. One of the goals of this project is to allow the model reasonable variation in the subsurface while constraining results as much as possible with available data. This can be done by integrating the available data, and although one particular data set may suggest a wide range of alternatives, when all the available data is combined, the possible solution population should be greatly reduced (Figure 1.6). The data may be divided into two basic types; "hard" data and "soft" data. Hard data are information that can be directly examined and evaluated, drill core data are an example of data that explicitly define material types. Soft data are less precise data; there is uncertainty associated with the values. Seismic data are an example; seismic exploration measures the velocity of shock waves through materials, but because different materials have similar velocities, and the degree of fluid saturation in a single material also effects velocity, only estimates can be made about material type and location. As a result, there is error associated with the interpretation of seismic data (there are also errors in hard data, but they are considered small enough to be ignored).
To accomplish the tasks of data entry, data evaluation, subsurface characterization, groundwater flow and contaminant transport modeling, and data visualization, many steps are required. A simplified flow chart (Figure 1.7) outlining these steps shows how a modeler can start with field data, or data prepared by other products, and be guided through statistical analysis of the data, generation of multiple realizations of each data property, development of model grids, kriging of data properties into model grids, generation of input files for flow and contaminant transport models, execution of models, and visualization of model results. A more complete flow chart is presented in Figure 1.8, but it will not be discussed in detail.