Danger of groundwater contamination widely underestimated because of shortcuts for aquifer recharge

Authors: Andreas Hartmann, Scott Jasechko, Tom Gleeson, Yoshihide Wada, Bartolomé Andreo, Juan Antonio Barberá, Heike Brielmann, Lhoussaine Bouchaou, Jean-Baptiste Charlier, W George Darling, Maria Filippini, Jakob Garvelmann, Nico Goldscheider, Martin Kralik, Harald Kunstmann, Bernard Ladouche, Jens Lange, Giorgia Lucianetti, José Francisco Martín, Matías Mudarra, Damián Sanchez, Christine Stumpp, Eleni Zagana, Thorsten Wagener


Main text
Groundwater pollution threatens human and ecosystem health in many areas around the globe. Shortcuts to the groundwater through concentrated recharge are known to transmit short-lived pollutants into carbonate aquifers endangering water quality of around a quarter of the world population. However, the large-scale impact of such concentrated recharge on water quality remains poorly understood. Here we apply a continental-scale model to quantify the danger of groundwater contamination by degradable pollutants through concentrated recharge in carbonate rock regions. We show that in regions where concentrated recharge takes place, the percentage of non-degraded pollutants in groundwater recharge increases from <1% to around 10-50%. In those regions, pollutants like Glyphosate can exceed their permissible concentrations by up to 19 times when reasonable application rates are assumed. Our results imply that in times of continuing industrial agricultural productivity, shortcuts to the groundwater may result in a widespread and substantial reduction of usable groundwater volumes.
Clean water is essential for nature and society 1 but pollution may result in a widespread reduction of available drinking water and in a threat to ecosystem services 2 . Large-scale studies on water security so far have mainly focused on water quantity rather than water quality 3 . Local studies have shown that concentrated recharge is a dominant driver for the pollution of groundwater resources in carbonate rock regions 4,5 . Fast flow through fractures and macropores in the soil can substantially mobilize pollutants even though some have been considered 'nonleachable' owing to their strong adsorption to colloids and sediment surfaces in the soil or due to fast degradation times 6 . Consequently, in addition to the dangers of more persistent pollutants like nitrate 7 , concentrated recharge may cause unexpected groundwater quality deterioration in regions where agriculture relies on degradable fertilizers and pesticides 8,9 . Although agriculture occupies around 40% of ice-free lands 10 , as of yet there are no continental-scale assessments of the impact of concentrated recharge and degradable pollutants on groundwater quality. This work estimates the danger of groundwater contamination via shortcuts for recharge to the groundwater by enlarged cracks and fissures, often referred to as concentrated recharge 11 . We do this by contrasting the travel times of water from the surface to the subsurface with the degradation times of typical agricultural pollutants. Our research domain is the carbonate rock regions of Europe, Northern Africa and the Middle East, collectively home to around half a billion people and provider of up to half of national water supplies in these regions 12 . Globally, 10-25% of the world population are estimated to largely or entirely depend on groundwater from carbonate rock aquifers 13,14 . Chemical weathering (in this context often referred to as karstification) increases the abundance of concentrated recharge, resulting in shortcuts between the surface and the subsurface 15 . During rainfall events, this characteristic of karstified carbonate rock allows large volumes of water to enter the subsurface 12,16 and can transport surface-borne pollutantsdissolved or attached to suspended sedimentsto the groundwater at short time scales, i.e. within days or weeks 17,18 . We derive the fractions of groundwater recharge that correspond to these short time scales, here referred to as rapid recharge fractions, by simulating transit time distributions with a state-of-the-art continental model 16 that accounts for concentrated recharge processes in carbonate rock regions. For analysis, our simulation domain is divided into four regions: humid, mountains, Mediterranean, and deserts (see model description in methods section).

Continental-scale estimation of rapid recharge fractions
We derive the rapid recharge fractions using the half-life times and survival times of three example pollutants, veterinary pharmaceuticals in manure (Salinomycin, 10-day half-life time 8 ), degradable pesticides (Glyphosate, 25-day half-life time 19 ), and pathogens (E. coli, 60-day survival time 9 ). We limit our simulated rapid recharge times of pollutant transfer to the groundwater to 5 days as immediate transfer after rainfall events, and to 90 days as transfer within the same season. We quantify the influence of concentrated recharge processes on fast transit to the groundwater by repeating the same procedure with concentrated recharge turned off in our model (see model description in methods section). That way, only diffuse recharge is simulated similarly to presently available large-scale hydrological models 12 . To explore the impact of climate, we compare simulated rapid recharge fractions with climate descriptors such as mean annual precipitation and temperature, aridity index (defined as long-term ratio of precipitation to potential evapotranspiration), mean annual number of rainfall events, and mean annual duration of snow cover. We evaluate the consistency of our model with independently derived Young Water Fractionsthe fraction of water less than 60-90 days old 20,21 -corrected for precipitation seasonality 22 at 119 carbonate rock springs across our simulation domain (see simulation of transit times in methods section) and with a dataset of >2,500 groundwater samples that were analysed for glyphosate abundance and concentration inside and outside the karst regions of the United States 23,24 .
Our simulations indicate that in the Mediterranean, up to 77±14% of concentrated recharge transits to the groundwater within one season (90 days, Fig. 1a Factors controlling rapid transit to the groundwater Concentrated recharge is the most important driver for the rapid transit of water from the land surface to groundwater. We use our simulation model to quantify the impact of concentrated recharge and climate on the abundance and strength of rapid transport of pollutants to the groundwater. When not considering concentrated recharge in our model, we find substantially reduced rapid recharge fractions (Mediterranean: 1±2%, mountain: 4±9%, desert: 0±0.2%, humid: 0±0.7% of rapid recharge compared to total recharge), indicating that concentrated recharge is the dominant mechanism for rapid transit to groundwater (Fig. 2). Including concentrated recharge, we find that rapid recharge fractions most strongly correlate with the aridity index for both Mediterranean and desert regions. The second strongest correlation is between the rapid recharge fractions and the average number of rainfall events per year as well as mean annual precipitation, both closely correlated with the aridity index (r=0.65 and r=0.94, both p≤0.001, for the Mediterranean and desert regions, respectively). In humid and mountain regions, we find weaker yet still significant correlations of rapid recharge fractions with mean annual temperature and average months with snow cover (0.44 ≤ r ≤ 0.51 and -0.3 9≤ r ≤ -0.35, respectively).
[ Fig. 2 about here] Our simulations indicate that wetness, expressed by increasing values of the aridity index, mean annual precipitation, and mean annual number of rainfall events, are important secondary controls for the strength of rapid recharge fractions in the carbonate rock areas of the Mediterranean and the desert regions. This agrees with local studies already showed that limited soil storage capacities 25 and the formation of desiccation cracks and stormy periods favour the fast transit of water to the subsurface 26,27 . Although weak, the positive correlation with temperature and with the average number of snow months for the humid and mountain regions indicate a reduction of fast recharge through snow storage that may cause a longer delay to the precipitation signal until transmitted to the hydrological system 28 . Similar results are obtained when using the 60-day threshold to define the rapid recharge fractions ( Fig. S5 and Fig. S6).

Quantification of the danger of groundwater contamination
Shortcuts into the groundwater increase the danger of groundwater contamination from pollutants of varying half-life times and survival times, particularly in the Mediterranean region.
Our model calculates the rapid recharge fractions corresponding to varying pollutants with thresholds from 5 to 90 days (Fig. 3). In order to quantify the influence of concentrated recharge, we repeat the same procedure without considering concentrated recharge (see model description in methods section). We find that among the four regions rapid recharge fractions increase from 5.1-15.2% for the 5-day-threshold to 36.3-77.3% for the 90-day-threshold with the lowest rapid recharge fraction constantly found in the humid region, while the largest rapid recharge fractions are always produced by the Mediterranean regions. Averaging over all threshold times, concentrated recharge increases the rapid recharge fractions by 20.4±10.8% (humid region), 24.7±12.7% (mountain region), 27.7±10.9% (desert region), and 49.5±20.5% (Mediterranean), compared to averages of 0.01-0.76% when concentrated recharge is not considered (Fig. 3).
Regarding our three example pollutants, we find that 9.9±8.8% of Salinomycin, 15.5±13.0% of Glyphosate, and 33.1%±21.5% of E. coli remain in groundwater recharge over all simulated carbonate regions when concentrated recharge is considered (Fig. 3). All three example pollutants show their largest rapid recharge fractions in the Mediterranean region where thin soils favour rapid fast transit of pollutants to the groundwater. Although no comprehensive datasets exist in Europe to evaluate these modelled values, our results correspond well with a national survey conducted in the contiguous United States 23,24 that found concentrations of 1.82±1.74 µg l -1 in 93 glyphosate detections out of 751 groundwater samples collected from carbonate aquifers (Fig. S 7). Glyphosate was detected ~5.3 times more often within carbonate rock regions, and with concentrations ~4.1 times higher compared to non-carbonate rocks regions (0.45±0.66 µg l -1 in 42 detections out of 1805 groundwater samples), which supports our finding that concentrated recharge increases the danger of groundwater contamination at larger scales (Fig. 3).

Implications of potentially underestimated contamination
Our carbonate rock recharge estimates only consider vertical infiltration and percolation fluxes.
We are aware that infiltrating pollutants may still experience attenuation in transit to the water supply system due to: (1) long travel times, as well as dispersion and diffusion processes, especially towards deep groundwater systems (our model accounts for depths up to ~30 m), (2) mixing with infiltrating waters from non-agricultural areas from which no contaminants originate, (3) mixing with less polluted groundwater that was recharged before application of the pollutant, (4) sorption to immobile colloids or sediment surfaces in the soil, or (5) removal or retardation of the pollutant during lateral groundwater flow towards a well. While toxicity disappears for most pathogens and pharmaceuticals after removal 8,9 , pesticides can transform into metabolites that can be toxic 31 . Furthermore, the half-life times or survival times of all pollutants may strongly vary depending on moisture, temperature, and redox conditions in the unsaturated zone. For all these reasons, and despite their good agreement with observed Young Water Fractions and Glyphosate concentrations, our results should be seen as a first-order estimate and worst-case scenario of the potential danger of contamination through degradable pollutants in regions with significant subsurface heterogeneity.
Overall, our continental study clearly elaborates that the danger of contamination through concentrated recharge is not limited to individual sites but relevant across a larger scales.
Especially in regions like the Mediterranean, travel times of recharge to the subsurface can be short, and decrease with increasing degree of aridity. In these regions the fast transit of agricultural pollutants to the groundwater poses a significant challenge for water and land use management. Rising human population and the increased consumption of water, food and energy will increase pressure on agriculture and natural resources 10,32 . While carbonate rock regions can be a valuable source of drinking water 12 , our results show that increasing agricultural production with the help of synthetic chemical fertilizers, pesticides and/or veterinary pharmaceuticals may cause substantial and pervasive groundwater pollution. This can result in a widespread reduction of available drinking water quality and harm ecosystem services more intensely than previously available large-scale models. While local approaches exist to map and protect areas with increased concentrated recharge 4 , large-scale water quality models are urgently needed to identify regions of increased danger of drinking water contamination over large scales relevant for water governance 3 . Our approach is the first to quantify the danger of groundwater contamination over an entire continent and therefore supports water governance to ensure future water security and ecosystem services.

Methods
The carbonate rock recharge model The model was developed to simulate carbonate rock groundwater recharge over Europe, Northern Africa and the Middle East (VarKarst-R 12,16 ). It simulates terrestrial hydrological processes on a 0.25° x 0.25° grid and at a daily temporal resolution for a 10-year period from 2002 to 2012. Its conceptualization is derived from previous applications at the aquifer scale where high observation densities allowed thorough evaluation 33,34 . It includes the spatiotemporal variability of carbonate rock infiltration due to rainfall and snowmelt, evapotranspiration, downward percolation from the upper soil layer to a lower soil epikarst layer and vertical percolation from the epikarst layer towards the groundwater. The latter corresponds to depths up to ~30 m 33 , considered in this study as the definitional groundwater depth to which recharge is added. Previous work shows that the model's representation of concentrated recharge provides more realistic estimates of groundwater recharge in carbonate rock regions compared to other large-scale hydrological models 12,16 because it assumes that even within the same hydrological landscape type there is a distribution of subsurface properties.
This variability is considered through distribution functions that allow for variable subsurface storage capacities, as well as of vertical hydraulic properties, over N horizontally parallel model compartments: where   (2) discard those with a coefficient of determination of fitted sine curve and observations ≤ 0, and (3) remove those with more than 2 months of continuous snow cover, creating substantial delay before infiltration.
A preliminary comparison of observed Young Water Fractions of the non-discarded springs (78 in total, see Table S 3 and Table S Fig. S 3b).

Derivation of climatic controls
To explore the impact of climate on rapid recharge fractions, we derive different climate descriptors from the daily 10-year input data of the continental carbonate rock recharge model 16 : mean annual precipitation, mean annual temperature, aridity index (defined as mean annual precipitation over potential evapotranspiration), mean annual number of rainfall events, mean annual number of months with snow cover, high intensity rainfall events defined by the mean intensity of the upper quartile of rainfall events. We explore the linkage between these climatic controls and rapid recharge fractions using the seasonal threshold (90 days) because the model evaluation was performed using the same threshold (Table S1, Fig. 2 and Fig. S 4). The same procedure is repeated using the alternate threshold for the model evolution (60-day threshold, see

Author Information
The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to A.H. (andreas.hartmann@hydmod.uni-freiburg.de).