PhD defence: Elisa Bjerre
Elisa Bjerre defends her thesis,
Bridging the Science-Policy Gap
An Interdisciplinary Study of Groundwater Modelling on the Interface of Decision Support and Stakeholder Involvement
Supervisors:
Professor Peter Engesgaard, IGN
Associate Professor Lone S. Kristensen, IGN
Chief consultant Anker L. Højberg, Ramboll
Associate Professor Søren Jessen, IGN
Assessment Committee:
Professor Okke Batelaan, Flinders University – Australia
Hydrologist Katherine Ransom, USGS – USA
Professor Karsten Høgh Jensen (chair), IG
Summary:
In a rapidly changing world, policy makers increasingly rely on scientific models to support knowledge-based decision making for dealing with arising water challenges. This Ph.D. study aims to bridge the science-policy gap to enable the implementation of science-based solutions for protecting groundwater quality in Denmark and abroad, in the context of agricultural nitrogen pollution. To this end, the thesis addresses two primary limitations of model-based decision support: i) model results are uncertain, and ii) models are time-consuming to run.
The interpretation of hydrogeological data into one model, the conceptual geological model, is a key uncertainty source in groundwater modelling. However, it is mostly overlooked in groundwater management for identifying groundwater protection zones. The thesis addresses conceptual geological uncertainty in a Danish case study by employing the multi-model approach, where multiple conceptual models are developed and evaluated in parallel. Drivers and barriers for taking account of geological uncertainty are identified amongst key stakeholders and decision makers. The drivers include long-term benefits of unpolluted drinking water and more robust management plans. Conversely, the barriers are immediate costs to stakeholders, complication of the local planning process and time limitations. The thesis presents recommendations to further integrate conceptual uncertainty analysis in the groundwater management framework.
Modelling groundwater for making forecasts to help water resource managers can be time consuming. Metamodeling applies machine learning methods to make fast and approximate models of a more complex hydrological model which enables applications in decision support. The thesis presents a Random Forest metamodel for predicting drainage fraction, an indicator for groundwater/surface water vulnerability to nitrogen pollution. Furthermore, the thesis proposes a new transferability assessment method, which is necessary to evaluate the model's usefulness when predicting into new geographical areas. The method combines a histogram-distance metric, quantifying differences in the covariates of the training and test datasets, weighted by covariate importance. Driven by the rise of machine learning applications in the hydrological sciences, it is anticipated that the need for such metrics will grow.
A digital version of the PhD thesis can be obtained from the PhD secretary Mikala Heckscher at mikala@ign.ku.dk