Integrated recharge-groundwater flow model and evolutionary algorithms for the time and fractal domain assessment of groundwater levels

Abrar Habib, Adrian Butler, Imperial College London, London, UK, John Bloomfield and James Sorensen, British Geological Survey, Crowmarsh Gifford, Wallingford, Oxon, UK

ABSTRACT

1-minute groundwater levels and river stage and 15-minute rainfall (among other meteorological variables) have been monitored for a period of 4 years in Wallingford, Oxfordshire, United Kingdom. The site is formed of a highly responsive riparian aquifer formed of terrace gravels associated with the River Thames that traverses the site. The groundwater levels exhibit rapid response to individual rain events, hence, significant groundwater fluctuation is observed on a variety of time scales.


An integrated 1D recharge-groundwater flow model has been set-up to simulate the groundwater levels at 15-minute resolution. The recharge model is a simplified Richards’ Equation Emulator (Mathias, Skaggs et al. 2015) and the groundwater flow model is an implicit discretization of the 1D Boussinesq Equation. Out of a total of 15 parameters, a sensitivity analysis resulted in 7 sensitive parameters: The van Genuchten empirical parameters from the recharge model and aquifer parameters that are used for the groundwater flow model. The calibration and optimisation were performed using a multi-objective evolutionary (genetic) algorithm; In particular, a variant of the Elitist Non-Dominated Sorting Algorithm (NSGAII-ε) was used. Three objective functions were used to determine non-dominated parameter sets: The Nash-Sutcliff Efficiency (NSE), the cross-correlation, and the unbiased percentage error in monthly peaks.

Groundwater fluctuations are simulated using the non-dominated parameter sets. The groundwater series are assessed in both the time and fractal domains. In addition, advantages of the use of fractal behaviour in assessing model performance are investigated.

This was a poster presentation 


Permanent link: