Modeling interbasin groundwater flow in karst areas: Model development, application, and calibration strategy

Highlights

• SWAT was modified to simulate interbasin groundwater flow in karst and non-karst area.

• Multi-site streamflow data and satellite-derived ETa were used for model calibration.

• Model performances for ETa and for streamflow are highly correlated.

• The use of satellite-derived ETa (MOD16 ETa) does not affect the model performance.

Abstract

Karstification is considered as one of the most common reasons for interbasin groundwater flow (IGF). IGF in some karst areas could be significant such that it must be accounted for in hydrologic modeling. In this study, the Soil and Water Assessment Tool (SWAT) was modified to explicitly account for IGF in karst areas. The modified model uses two conceptual models to simulate hydrologic processes in karst and non-karst regions. The modified model was applied in the karst-dominated region in the southwest Harz Mountains, Germany. Multisite streamflow data and satellite-derived actual evapotranspiration (ETa) were used for model calibration. Results show that (1) the modified model can be satisfactorily calibrated and validated for streamflow and ETa (2) the model performance for ETa and streamflow at some gauging stations are highly correlated, and (3) the use of satellite-derived ETa does not affect the model performance.

Introduction

The term “karst” refers to a region with distinct landscape features (e.g., sinking streams, sinkholes, and springs) and underground features (e.g., underground conduits and caves). In some karst regions, the karst landscape features could be absent or subtle, but their aquifers could be heavily karstified (Ford and Williams, 2007). Karst aquifers are developed as a result of dissolution of karstifiable rocks (e.g., limestone, dolomite, gypsum, and rock salt), the so-called karstification (Ford and Williams, 2007, Bögli, 1980, Howard, 1963). Karst aquifers account for about 10% to 15% of the continental area and karst groundwater is one of the sources of drinking water for approximately a quarter of the world’s population (Ford and Williams, 2007). However, karst groundwater is particularly vulnerable to contamination due to their distinct hydrogeologic characteristics (Goldscheider, 2005, Doerfliger et al., 1999, Drew and Hötzl, 1999). Therefore, understanding the hydrogeologic characteristics of karst aquifers plays an important role in water resources management in karst regions.

Hydrogeologic characteristics of karst aquifers are different from other aquifers (Bakalowicz, 2005). Karst aquifers often exhibit a duality of recharge, infiltration, porosity, flow and storage (Goldscheider and Drew, 2007, White, 2002, Gun, 1986). Karst aquifers also show a high degree of spatial heterogeneity in hydraulic properties (Bonacci et al., 2006). Especially, the surface drainage basin in karst aquifers usually do not coincide with the groundwater basin (Spangler, 2001, Dar et al., 2014). Karstification is considered as one of the most common causes of interbasin groundwater flow (IGF) (Le Moine et al., 2007). Water recharged to karst aquifers could flow through an underground conduit system spanning over several basins and emerge at springs located at distant sites (e.g., Anderson et al., 2006, Belcher et al., 2009, Le Moine et al., 2008). It should be noted that IGF could also occur in porous aquifer in form of regional groundwater flow (Tóth, 1963, Nguyen and Dietrich, 2018, Danapour et al., 2019), however, in this study we focus on IGF in karst areas. The term IGF in this study could be also understood as regional groundwater flow across surface topographic divides. IGF in karst areas could significantly alter the water budget of a basin (e.g., Anderson et al., 2006, Le Moine et al., 2008). Considering the aforementioned facts, IGF in karst areas should be accounted for in hydrological modeling, especially in the context of transboundary or interbasin groundwater management.

Various models have been used to simulate IGF in karst aquifers with varying model complexity, ranging from physically based distributed to conceptual lumped models. Physically based distributed models simulate groundwater flow based on hydraulic head gradient, therefore, groundwater could flow across topographic divide units, which are normally considered as isolated groundwater units in surface hydrology. Conceptual models can simulate IGF by allowing the simulation (or routing) of groundwater flow between topographical basins. Some models of these types are the Modular Three-Dimensional Finite-Difference Ground-Water Flow Model (MODFLOW, Scanlon et al., 2003), the modified WetSpa model (Liu et al., 2005), the modified Soil and Water Assessment Tool (SWAT, Arnold et al., 1998, Nerantzaki et al., 2015, Malagó et al., 2016, Palanisamy and Workman, 2014), modèle du Génie Rural à 4 paramètres Journalier (GR4J, Perrin et al., 2003, Le Moine et al., 2007, Le Moine et al., 2008), the tank model (Anaya and Wanakule, 1993), and the multi-cell aquifer model (Rozos and Koutsoyiannis, 2006, Barrett and Charbeneau, 1997). SWAT is one of the most widely-used models to simulate the effect of land use, agricultural management practices and climate change on water and chemical yields in non-karst areas (Arnold and Fohrer, 2005, Gassman et al., 2007, Krysanova and White, 2007, Molina-Navarro et al., 2017). Therefore, the modified SWAT versions which account for IGF in karst areas could potentially help to explore these effects in karst regions.

The aforementioned modified SWAT models, the so-calledKarstSWAT (Palanisamy and Workman, 2014) and KSWAT (Nerantzaki et al., 2015, Malagó et al., 2016), simulate IGF in karst regions. The KarstSWAT model was specifically developed for watersheds dominated by sinkholes and springflow, which is mainly fed by the water from sinkholes (Palanisamy and Workman, 2014). The KSWAT model combines the adapted SWAT model (Fig. 3, Malagó et al., 2016) and the karst-flow model (Nikolaidis et al., 2013). The adapted SWAT model assumes that all water entering the soil profile is karst groundwater recharge (Fig. 3, Malagó et al., 2016). However, part of the infiltrated water could contribute to the streamflow as lateral flow and baseflow if the underlying aquifer of a subbasin is not entirely a karst aquifer (e.g., Palanisamy and Workman, 2014). The adapted SWAT model does not differentiate between concentrated recharge and diffuse recharge. The karst-flow model is the two-linear-storage reservoir model, which receives the recharge simulated from the adapted SWAT model (or from the original SWAT model, Nikolaidis et al., 2013) and routes it to spring. Outflows from the two reservoirs of the karst-flow model represent flow from wide conduits and narrow fractures (Kourgialas et al., 2010, Malagó et al., 2016). Because of the lumped feature of deep recharge from the adapted SWAT model, the KSWAT model does not explicitly differentiate between (1) the diffuse recharge and concentrated recharge, (2) between matrix storage and conduit storage. This is important because these recharges and storages are different in term of travel time and storage. In addition to the aforementioned disadvantages, the recharge area of the karst aquifer in the KarstSWAT and KSWAT models follows the subbasin delineation of SWAT.

In addition to the model development, parameter identification in karst regions is also subject to higher uncertainty compared to other regions (Brenner et al., 2018, Hartmann et al., 2017, Hartmann et al., 2013). This is because the karst aquifer is highly heterogeneous and the upper flux (actual evapotranspiration, ETa) and the lower flux (karst groundwater recharge) are usually unknown. In order to develop a robust model and to minimize the parameter uncertainty, especially in karst regions, multi-variable calibration is suggested. ETa is one of the main components of the hydrologic cycle. About 60% of the annual precipitation on the global land surface returns to the atmosphere as evapotranspiration (Jung et al., 2010, Oki and Kanae, 2006). Considering the aforementioned facts, observed ETa should be used for calibrating the model. However, a direct observation of ETa is very scarce.

In non-karst areas, many studies have used satellite-derived ETa for model calibration (e.g., Rajib et al., 2018, Franco and Bonumá, 2017, Vervoort et al., 2014, Rientjes et al., 2013, Droogers et al., 2010, Zhang et al., 2009, Muthuwatta et al., 2009, Immerzeel and Droogers, 2008). In these studies, satellite-derived ETa was either used as an independent calibration data set or as input data. Results showed that the model performance for streamflow could decrease when constraining model calibration with satellite-derived ETa as an additional variable (Vervoort et al., 2014). However, the above-mentioned studies showed that using satellite-derived ETa in combination with observed streamflow for calibrating a hydrologic model could (1) better reproduce the catchment’s water balance, (2) reduce the parameter uncertainty, (3) increase the model robustness, and (4) detect the structural model issues. In karst areas, the use of satellite-derived ETa as an additional calibration variable has not been given enough attention.

In this study, we developed a conceptual model which is able to (1) simulate surface and subsurface flows in both karst and non-karst areas, (2) apply for a region where the karst aquifer boundaries do not coincide with the surface subbasin boundaries, and (3) represent different recharges (diffuse recharge and concentrated recharge) and storages (matrix storage and conduit storage) in karst areas. The proposed concept was implemented into the SWAT model. The modified SWAT model was tested in the karst-dominated area in Lower Saxony, Germany. The effects of using satellite-derived ETa for model calibration on the model performance was examined in detail. The Moderate Resolution Imaging Spectroradiometer (MOD16 ETa, Mu et al., 2013) was used for the model calibration.