Data interpolation for groundwater modelling : How choice of interpolation method and sample size affect the modelling results

Sammanfattning: Over the past several decades, the use of groundwater modelling has been increasing in order to better evaluate the complexities inherent in hydrogeological calculations. Information required for groundwater modelling is for example elevation of soil and bedrock layers, which often is collected by drilling. This is both time consuming and expensive, making it impossible to collect an unlimited number of data points. It is not uncommon that smaller hydrogeological investigations are based on only three or four sample points. To approximate values of unknown points in the study area, the known values of the measured data is interpolated. The interpolation can be done with different methods, and the estimations of the elevation of geological layers at the unknown points might vary with different methods. How the uncertainty of the interpolations then affects the modelling results is generally unknown when simulating groundwater flow. The aim of the thesis work is to investigate how hydrogeological results from groundwater models are affected by choice of interpolation method and by sample size of which the interpolation is based on. The hydrogeological models were simulated within the framework of a Swedish railway project where an unusually large number of probing data was available. The study focuses on two-dimensional, steady-state groundwater flow modelled in the software SEEP/W. Consequently, the objective of the modelling work was to simulate groundwater flow where the difference between the models was how the geometry of the geological layers was defined. The definition of the geometry was done with interpolations of the probing data with different interpolation methods and with interpolation based on different sample sizes. Twelve different interpolation methods in the software Surfer was used to interpolate the total of 357 data points. The interpolations that were estimated to be reliable were then used in the groundwater modelling. Groundwater models were also simulated based on a reduced amount of data in order to investigate the importance of sample size. The total amount of data points was reduced to 50%, 25%, 5% and 1% of the initial sample size before it was interpolated and then used to define the geometry in the groundwater models. The study showed that although the choice of interpolation method and sample size affected the results of the modelled groundwater flow rate, none of the deviations between the model results were larger than what would be considered as acceptable in a hydrogeological context. The models based on at least 90 data points showed good precision while the precision of modelled groundwater flow decreased when sample size decreased. Thus, when interpolating data for groundwater modelling, the sample size should be determined by the required precision of the study. For the modelling of the 1% data, only one of the six models was possible to simulate. This since at least three data points are required for each geological layer in order to perform interpolations, and this was only obtained in one of six randomly selected data sets. The study indicates that around 10-20 sample points is a minimum for a study area with similar conditions as the reference project in order to have a high probability of obtaining enough data of all geological layers of interest. Finally, the study indicates that there are other parameters in groundwater modelling, for example hydraulic conductivity and boundary conditions, that might have an equal of higher impact on the model results. These parameters should therefore also be determined with a high precision in order to gain accurate modelling results.