Prediction Based Groundwater Abstraction and Models for Habitat Assessment

Student thesis: Master thesis (including HD thesis)

  • Kirsten Lyngholm Lindbjerg
4. term, Water and Environment, Master (Master Programme)
Today Danish groundwater abstraction licenses are given as a license to abstract a certain amount of groundwater every year over a period of years. Due to the natural variations in the size of the groundwater reservoir, this is a rather conservative approach to giving groundwater abstraction licenses, as the availability of the sustainable resource is sometimes higher than what is being abstracted. Thus, it is the aim for this project to investigate a strategy for groundwater abstraction in which it is possible to abstract a maximum amount of groundwater whilst taking the natural variations in the available resource into account. Furthermore, the possibility to perform habitat modelling with simpler habitat models is investigated.

The investigation of this strategy commences with a study of the possibility to develop models for prediction of the yearly minimum flow in streams and assessing the equivalent habitat area with habitat modelling. Two methods for prediction of the yearly minimum flow in streams are investigated: Linear models and trained neural network models. Both prediction model concepts use measurements of groundwater level from late winter or early spring as input data. Based on the prediction of the minimum flow, the goal is to assess the corresponding habitat area in order to evaluate the possibility to abstract more or less the following year.

In this project, the above mentioned is investigated through a case study of Binderup Å situated in North Jutland, Denmark. For this purpose, the two model concepts, linear modelling and neural network models, have been used to produce models for prediction of the yearly minimum flow in Binderup Å based on measurements of groundwater level in three wells. The linear regression model with the highest Nash-Sutcliffe Effeiciency amongst the developed models has a 95% confidence interval of 95 l/s and is based on groundwater level measured in February in well no. 26.1943. The neural network model with the highest NSE is based on measurements of groundwater level in February, January and December in the three wells 34.492, 26.1943 and 26.536. This model has a 95% confidence interval of 61 l/s.

The study of a simple habitat modelling concept is based on a comparison between a habitat model with cross sections and modelled water levels from the National Water Resources Model for Denmark and a habitat model with cross sections and water levels measured for this project. This study concludes that with the available data, the simple habitat model is adequate for further modelling in this project. The simple habitat model concept is then used to produce habitat models for three life stages of trout for sections of Binderup Å. These models can be used to evaluate the habitat area at different flows, including the effect of flow reductions from increased groundwater abstraction.
Publication dateJun 2018
Number of pages77
ID: 280488509