Quantifying the impact of pumping on groundwater heads using observation data and advanced time series analysis

Abstract

The dynamics of groundwater levels within unconfined aquifers are often the result of numerous and interacting factors, such as land cover change, climate variability and groundwater pumping. For such unconfined aquifers, estimating the impact from pumping is highly significant for resource management but also very challenging. To date, in Australia, the HARTT multiple linear regression methodology (Ferdowsian et al. 2001) has been widely used to estimate the impact of climate variability on groundwater levels. Recently, transfer function noise (TFN) models have been developed to better link water table dynamics with different types of individual stresses, including pumping (von Asmuth et al. 2002, von Asmuth et al. 2008). Peterson & Western (2011) advanced the transfer function noise model of von Asmuth et al. (2008), which is hence referred to as SMS-TFN, to account for non-linear unsaturated zone processes by inclusion of a parsimonious vertically lumped soil moisture model. However, despite this model performing very well for non-pumped aquifers, there is little confidence in its ability, or any existing method, to predict the effect of human interventions, such as land use change and pumping, on water table dynamics. This paper proposes a new time series formulation for estimation of the impacts of groundwater pumping. It is based upon standard well hydraulics and is an extension to Peterson & Western (2011). Herein, the model is described and assessed against existing methods via use of a synthetic MODFLOW study. A MODFLOW model was constructed to derive groundwater hydrographs produced by known pumping rates in combination with varying climate forcing. The model has the following attributes: specified constant and general head boundaries, three layers, pasture landcover and climate data with different types of pumping wells. The model was used to simulate synthetic groundwater time series from observation bores near the pumping wells. Next, the HARTT model, standard TFN model and modified SMS-TFN model were applied to the synthetic groundwater hydrographs and the pumping impact was estimated. The performance of each model was than assessed by reviewing the modelled pumping contribution against the 'known' pumping. The predictive performance of models was also assessed by split sample calibration, evaluation and calculation of the coefficient of efficiency (COE) and the Akaike information criterion (AIC). The results of calibration suggested that the modified SMS-TFN model performed better than the other existing methods and produced estimates of groundwater impact very comparable to that estimated within MODFLOW. Moreover, the result of modelling the pumping contribution to head showed that only the modified SMS-TFN correctly modeled the behavior of pumping in groundwater time series. Further work is required to assess this model in more complex scenarios and on non synthetic cases. (Figure presented).