Quantifying the impact of climate drivers on groundwater dynamics in diverse environmental settings and heterogeneous time series data – a method comparison

While critical to humans and ecosystems, groundwater accessibility is threatened by climate change, which alters groundwater recharge and can lead to water table decline. However, quantifying groundwater response to climatic changes remains highly uncertain. Modeling efforts are constrained by the oversimplification of processes, a lack of data for calibration and validation, and an uncertainty regarding processes and state variables. Improving our understanding of hydrological systems requires better quantification of the relationship between individual system inputs (climate signals) and outputs (groundwater table). However, this task is challenging, as groundwater is an integral part of the Earth system, where complex feedback mechanisms and multiple interacting factors and drivers (confounders) complicate the investigation of single processes. Topography, for example, is a core driver of groundwater flow and determines water table depth as well as where recharge and discharge areas develop. Geological properties like permeability and porosity strongly influence groundwater flow, response time, and storage capacity. Infiltration from surface waters is often the main source of groundwater recharge in drylands and may blur the direct influence of precipitation. Consequently, the nature and strength of the climate-groundwater connection are likely to vary across different environmental contexts. In particular, the subsurface's damping effect complicates a climate-groundwater analysis. Damping is described as the delay and smoothing of an output signal (i.e., water table) in a system compared to the input signal (i.e., precipitation). This effect is not only dependent on static subsurface characteristics but also nonstationary and nonlinear. Finding a relationship between precipitation and groundwater time series with classical correlation analysis remains, therefore, often unsuccessful. Here, we propose analyzing in situ groundwater time series and other groundwater-associated variables using statistical methods that account for confounders and damping. We compare the performance and feasibility of methods like (1) partial cross-correlation, (2) deconvolution, (3) clustering of pulse-response functions (a byproduct from deconvolution), (4) clustering functional relationships between climate variables and groundwater, and (5) causal interference methods (i.e., PCMCI–CMI). We give an overview of the advantages and disadvantages of every tested method. We aim to provide clarity in a landscape of numerous available methods and to offer practical guidance for holistic analyses that encounter similar challenges.