A Manifesto for Decision-Support Groundwater Modelling

Human management of an environmental system changes that system. Groundwater systems are complex beyond measure. Hence the nature and magnitude of human-induced changes cannot be known with certainty. Nevertheless, it may be possible to place limits on our assessment of what these changes will be, based on scientific knowledge of system processes and properties.

System processes follow the laws of nature. They can be described using equations that can be solved numerically. However, these equations are only approximate at the scale at which they are applied. Furthermore, the properties of the aquifers and aquitards which are featured in these equations, and which control the flow of groundwater and the transport of dissolved constituents, can be measured at only a handful of locations. Extrapolation between these locations fails to capture the profoundly heterogeneous nature of geological media.

Because groundwater behaviour is determined by aquifer and aquitard properties, field measurements of its behaviour, both present-day and historical, can inform these properties. However, this information is rarely sufficient to allow accurate assessment of the range of predicted consequences of a contemplated course of management action.

Integrity of environmental management demands that decision-support modelling expose the uncertainties of decision-critical predictions. It also requires the reduction of these uncertainties where information from one or multiple sources allows this, particularly if the risks associated with management outcomes are high.

Quantification and reduction of model predictive uncertainty requires that numerical simulators be used in conjunction with other software packages. These packages must interact closely with simulators. They must constrain a simulator’s representation of aquifer properties in order to ensure that its numerical outcomes replicate past system behaviour. At the same time, they must support simulator exploration of predictive possibilities that are compatible with system properties, insofar as these can be measured or constrained. Ideally, they should enable formulation of management strategies which maximise management utility while ensuring that environmental constraints are respected, taking account of uncertainties associated with predictions of the latter.

Natural systems are far too complex to replicate numerically. Nevertheless, numerical modelling can support their management. Its ability to do this rests on the following premises:

  1. Numerical simulation of environmental processes is imperfect.
  2. The metric by which decision-support modelling should be judged is not simulation perfection, it is its ability to extract information from all available sources in order to quantify and constrain the uncertainties of decision-critical predictions.
  3. Model-partner software plays an essential role in satisfying this metric.