This insight into groundwater uncertainty is written by Ines Epari, an experienced Groundwater Specialist and Pacific Environment’s Practice Leader – Land and Water.
What if random guesses are better than expert knowledge?
Remember Paul the Octopus? Paul predicted the outcomes of soccer matches better than the experts, which by our expectations should be the ones with the more accurate predictions. However, there are situations where random guesses can beat experience and knowledge. Groundwater modelling is just one of them. In May 2016, thought leaders in the area of uncertainty in groundwater modelling shared their insights in the IAH/NCGRT Modellers Forum: Uncertainty in Groundwater Modelling. This insight shares the challenges and solutions that were presented in that forum.
Replicating nature in groundwater models
Computer power has evolved over the last decade and so has the size of groundwater models. When groundwater modelling started out, people were thrilled about a 2D model with maybe 1,000 cells or nodes. Take my Master Thesis as an example, a regional groundwater model of the Murrumbidgee Irrigation Area in Griffith, NSW (2003). The grid cells were 1km by 1km and the model was divided into four layers, resulting in a “whopping” 170,000 cells. Nowadays, people would laugh about this number, as groundwater models now feature millions of cells. But by going with finer and finer grids (just because we can), are we actually gaining any additional certainty in our modelling results? It is not the grid size or cell count that will give the accurate predictions, but rather the conceptual model behind the numerical model and the parameters we choose.
Ideally, we want to create large and complex models that are almost identical to the processes and structures found in nature and the built environment. However, the greater the size of the model and level of detail, the greater the false sense of security in the outcomes of the model. In groundwater modelling, only little data is available to achieve a true image of nature, hence even the most elaborate model will be based on uncertain inputs and the results will have an unknown degree uncertainty.
Figure 1 – Geostatistical methods can be used for creating inhomogeneous input parameters for groundwater models
Solutions for the industry
Groundwater models are built for a specific purpose and the outputs of these models are often used as direct and quantitative inputs to a decision-making process. Together with the end-users, the model purpose needs to be well defined in the initial stages of a project. What level of accuracy is expected? Can this be achieved with the available data?
It is our task as modellers to communicate to our clients the uncertainty of the modelling inputs and results and if possible show other ways of using groundwater models. What if we can calculate ranges and trends instead of only one result? What if we can exclude certain outcomes and hence narrow down potential solutions? These discussions will help to find a modelling approach that the required level of certainty for the project.
Figure 2 – Modelling results for a series of models that have a decreasing level of detail (grid size doubled with each level, cell size reduced by factor 8).
Get certain about groundwater and move forward today
Has your geotechnical investigation report shown that groundwater might be intercepted? Contact us today so that we can find a modelling approach that will specifically suit your operation and help you move ahead.