Challenges in developing a global gradient-based groundwater model (G<sup>3</sup>M v1.0) for the integration into a global hydrological model

Abstract

Abstract. In global hydrological models, groundwater (GW) is typically represented by a bucket-like linear groundwater reservoir. Reservoir models, however, (1) can only simulate GW discharge to surface water (SW) bodies but not recharge from SW to GW, (2) provide no information on the location of the GW table, and (3) assume that there is no GW flow among grid cells. This may lead, for example, to an underestimation of groundwater resources in semiarid areas where GW is often replenished by SW or to an underestimation of evapotranspiration where the GW table is close to the land surface. To overcome these limitations, it is necessary to replace the reservoir model in global hydrological models with a hydraulic head gradient-based GW flow model. We present G3M, a new global gradient-based GW model with a spatial resolution of 5′ (arcminutes), which is to be integrated into the 0.5∘ WaterGAP Global Hydrology Model (WGHM). The newly developed model framework enables in-memory coupling to WGHM while keeping overall runtime relatively low, which allows sensitivity analyses, calibration, and data assimilation. This paper presents the G3M concept and model design decisions that are specific to the large grid size required for a global-scale model. Model results under steady-state naturalized conditions, i.e., neglecting GW abstractions, are shown. Simulated hydraulic heads show better agreement to observations around the world compared to the model output of de Graaf et al. (2015). Locations of simulated SW recharge to GW are found, as is expected, in dry and mountainous regions but areal extent of SW recharge may be underestimated. Globally, GW discharge to rivers is by far the dominant flow component such that lateral GW flows only become a large fraction of total diffuse and focused recharge in the case of losing rivers, some mountainous areas, and some areas with very low GW recharge. A strong sensitivity of simulated hydraulic heads to the spatial resolution of the model and the related choice of the water table elevation of surface water bodies was found. We suggest to investigate how global-scale groundwater modeling at 5′ spatial resolution can benefit from more highly resolved land surface elevation data.

Importance of Spatial Resolution in Global Groundwater Modeling

“…The assessment of global groundwater resources has been the subject of multiple studies recently, including impacts of groundwater abstractions (Wada et al ) and climate change (Portmann et al ; Cuthbert et al ). Furthermore, global hydrological models (GHMs) started to replace their bucket‐like linear groundwater storage models with (hydraulic) head gradient‐based models to better represent the interaction between surface water and groundwater as well as lateral and vertical flows, including capillary rise (de Graaf et al , , ; Reinecke et al , ). Compared to established regional models, global groundwater models are a new endeavor for groundwater hydrologists that is mostly unexplored.…”

Spatially distributed sensitivity of simulated global groundwater heads and flows to hydraulic conductivity, groundwater recharge, and surface water body parameterization

“…Range of parameter multipliers used in the Morris experiments. Each parameter multiplier is sampled in log space (log 10 (Multiplier)) with sampling based on Campolongo et al (2007) and optimized with Ruano et al (2012).…”

“…For 7 parameters (without the ocean boundary) and 6 GHRUs we get a total number of parameters k = 42 + 1, where +1 stands for the ocean boundary, which is not varied by GHRU, resulting in 1848 simulations. Elementary effects are based on an initial random sampling of 10 000 trajectories using Campolongo et al (2007) and then reduced by assuming 42 (number of parameters times GHRUs without ocean boundary) so-called optimized trajectories following Ruano et al (2012). Only random sampling might result in nonoptimal coverage of the input space; thus the initial random trajectories are used to select only those that maximize the dispersion in the input space.…”

“…Only random sampling might result in nonoptimal coverage of the input space; thus the initial random trajectories are used to select only those that maximize the dispersion in the input space. This optimal set of trajectories is approximated with a reasonable computational demand using the methodology developed by Ruano et al (2012). The experiment resulted in 1848 simulations with an overall runtime of 2 months on a machine with 20 computational cores (enabled hyper-threading) and 188 GB RAM.…”

Development of the Community Water Model (CWatM v1.04) A high-resolution hydrological model for global and regional assessment of integrated water resources management

“…In particular, simulations of human water demands from different sectors such as agriculture, industry, and households could have a large impact on estimated hydrological storage (e.g., groundwater) and fluxes (e.g., discharge) (Alcamo et al, 2007;Wada et al, 2016). More efforts have gone into better groundwater representation in large-scale hydrological models to realistically simulate groundwater levels and surface-35 groundwater interactions (Pokhrel et al, 2015;Wada, 2016;Reinecke et al, 2019;de Graaf et al, 2015de Graaf et al, , 2017Decharme et al, 2019).…”