The influence of the length of the calibration period and observation frequency on predictive uncertainty in time series modeling of groundwater dynamics

Abstract

The influence of the length of the calibration period and observation frequency on the predictive uncertainty in time series modeling of groundwater dynamics is investigated. Studied series are from deltaic regions with predominantly shallow groundwater tables in a temperate maritime climate where heads vary due to precipitation and evaporation. Response times vary over a wide range from ∼60 to ∼1200 days. A Transfer Function‐Noise model is calibrated with the Markov Chain Monte Carlo method to both synthetic series and measured series of heads. The model fit and uncertainty are evaluated for various calibration periods and observation frequencies. It is often assumed that the required length of the calibration period is related to the response time of the system. In this study, no strong relationship was observed. Results indicate, however, that the required length of the calibration period is related to the decay time of the noise. Furthermore, the length of the calibration period was much more important than the total number of observations. For the measured series, the credible intervals could commonly be reduced to ∼10% of the measured head range and the prediction intervals to ∼50% of the measured head range with calibration periods of 20 years with approximately two observations per month.

Estimation of groundwater recharge from groundwater levels using nonlinear transfer function noise models and comparison to lysimeter data

“…Fank (1999) used the water table fluctuation method to estimate groundwater recharge from observed groundwater levels and computed an average recharge of 393 mm yr −1 over the period 1992-1996. This estimate is comparable to the 296 and 396 mm yr −1 reported by Stumpp et al (2009) for the two lysimeters that were operated at the site in the period 1992-2001. Stumpp et al (2009) also applied a HYDRUS-1D model to simulate unsaturated zone flow.…”

“…This estimate is comparable to the 296 and 396 mm yr −1 reported by Stumpp et al (2009) for the two lysimeters that were operated at the site in the period 1992-2001. Stumpp et al (2009) also applied a HYDRUS-1D model to simulate unsaturated zone flow. Using stable isotope δ 18 O measurements, it was shown that lysimeter recharge could be adequately simulated with this physically based model, although recharge peaks were generally underestimated.…”

“…In this data-driven method, impulse response functions are used to describe how groundwater levels react to different drivers such as precipitation, evaporation, and pumping. The method has been successfully applied to characterize and analyze groundwater systems around the world, for example in Brazil (Manzione et al, 2010), Italy (Fabbri et al, 2011), the United Kingdom (Ascott et al, 2017), and India (van Dijk et al, 2019). The advantages of datadriven models compared to numerical groundwater models are faster model development and a lower number of calibration parameters (e.g., Bakker and Schaars, 2019).…”

Groundwater Hydrograph Decomposition With the HydroSight Model

“…The lower AICc and BIC indicates the better model. Moreover, the Nash-Sutcliffe efficiency (NSE) has been widely used to assess the goodness fit of a hydrograph (Ritter and Muñoz-Carpena, 2013;van der Spek and Bakker, 2017). Therefore, to assess the TFN model performance, the NSE (Nash and Sutcliffe, 1970) and unbiased NSE (Shapoori et al, 2015a) were adopted in the calibration and evaluation periods, respectively.…”

“…Results indicated that sustained groundwater pumping over time has a significant influence on groundwater level decline. Depletion of groundwater levels is a global phenomenon and is defined as long term water level declination caused by sustained groundwater pumping over time (Tularam and Krishna, 2009). More attention should be paid to manage the groundwater pumping activities to ensure the sustainable use of groundwater.…”