Post-drought increase in regional-scale groundwater nitrate in southwest Germany

Elevated nitrate concentrations in groundwater are a common challenge for water management. One important factor in this context is higher frequencies and intensities of wet-dry cycles that may cause increased nitrate concentrations in groundwater due to nitrate flushes after drought termination. Yet systematic studies on regional-scale impacts of droughts on groundwater nitrate concentrations are missing so far. Here we analyzed time series of 44 shallow groundwater wells and 41 springs all across the German Federal State Baden-Wuerttemberg from 2000 to 2018 to characterize patterns of post-drought nitrate increase in groundwater. In general, half of the exceptional nitrate concentrations, which exceeded the 80th percentile of long-term nitrate measurements, could be related to droughts in the research timeframe. The 2003 drought event stood out in terms of drought severity and post-drought nitrate concentration increases in our data. The great majority (91%) of all monitoring sites showed at least one exceptionally high nitrate concentration in the 4 years following the 2003 drought event. Springs were mainly located in forests of steep low mountain ranges and wells in cropland of flat river valleys. There-fore, delay times between drought intensity and nitrate concentration increases as well as magnitudes of nitrate concentration increase were diverse among wells and springs. We derived two distinct nitrate response patterns: (i) nitrate increases immediately following drought events (more common for springs and fractured rock aquifers) and (ii) delayed nitrate increases (more common for wells and porous aquifers). Springs generally showed quicker (median of 101 days) but weaker (median of + 1.3 mg/L) post-drought nitrate increases than wells (185 days, + 3.4 mg/L). Only few sites exhibited no post-drought nitrate increase and post-drought mean-nitrate concentrations of groundwater reservoirs were extraordinarily high in 2006. Overall, we demonstrate that post-drought nitrate increase in groundwater is omnipresent, while different landscapes and hydrogeological characteristics create a diverse regional pattern. As severe droughts become more frequent in a changing climate, post-drought nitrate increase may intensify problems regarding water quality and supply.


| INTRODUCTION
Throughout the world, elevated nitrate concentrations in groundwater violate drinking water standards and pose a risk to human health (Canter, 1997). In aquifers overlain by agricultural fields, nitrate inputs into groundwater largely originate from a nitrogen surplus of intense agriculture (Goulding, 2000;Hansen et al., 2011Hansen et al., , 2017Menci o et al., 2016). Sources of nitrate contamination are well known. They were identified by for example, statistical analysis of landuse types and vadose zone characteristics (Lockhart et al., 2013;Vazquez et al., 2005) or by stable isotope signatures, namely 15 N and 18 O of the nitrate molecule (Bourke et al., 2019;Savard et al., 2010). Stable isotopes may also characterize the extent of denitrification, the main reduction pathway of nitrate within the saturated zone (Webster & Dowdell, 1984;Wild et al., 2018).
During drought events, missing rainfall creates soil moisture deficits, which, if prolonged, lead to agricultural droughts (Tallaksen & van Lanen, 2004). Then nitrogen turnover in the unsaturated zone is affected, because mineralization of organic nitrogen and nitrification are reduced under dry conditions and less nitrate is formed. At the same time, reduced or absent percolation can cause an accumulation of inorganic and organic nitrogen in the unsaturated zone. When a drought is terminated by extensive rainfall, nitrogen turnover resumes in wetter soil and may generate a quick downward nitrate response with percolating soil water (Borken & Matzner, 2009). In agricultural land intensified post-drought nitrate leaching was exemplified in pasture soil (Shepherd et al., 2018), beneath fertilized grassland (Klaus et al., 2020), under wheat and maize (Zhu et al., 2009) or sugar beet (Groves & Bailey, 1997). In principle, similar nitrate dynamics can be expected in forests, although absolute loads are less (Di & Cameron, 2002). While Tietema et al. (1997) did not find clear postdrought nitrification pulses within the acidified soils of their experimental plots, various studies reported a reduction of nitrate leaching during droughts and an increase after drought termination (e.g., Bechtold et al., 2003;Leitner et al., 2020).
Post-drought nitrate leaching has negative effects both on surface and on sub-surface water resources. After an early summer drought in 2005, Lange and Hänsler (2012) measured elevated nitrate concentrations in springs and inside a small first-order stream draining a forested headwater catchment. Opsahl et al. (2017) found increasing post-drought nitrate concentrations in wells of a karst aquifer using high-frequency in situ optical sensor measurements. During droughts, river flow is largely sustained by baseflow from groundwater reservoirs, and the river water resembles the water quality of discharging aquifers (Smakhtin, 2001). Due to the strong relationship between water quality and below average streamflow (Hellwig et al., 2017), exceptional nitrate responses in springs and groundwater are directly relevant for the ecological status of rivers. Morecroft et al. (2000) found evidence for a post-drought stimulation of mineralization and nitrification that was more pronounced at woodland compared to agricultural river sites. In the same area, Outram et al. (2014) identified a large transfer of nitrate following a prolonged dry period in one out of three investigated catchments. As a response to a severe drought in 2012, Van Metre et al. (2016) found anomalously high nitrate levels in 17 streams of midwestern United States (US) with concentrations outside the 95% confidence interval of the regression-predicted mean. Recently, Lisboa et al. (2020) detected post-drought peaks in phosphorous but none in nitrogen in streamflow samples of 12 tributaries in the Owasco Lake watershed, New York. In groundwater, large scale studies mainly concentrated on general factors influencing groundwater nitrate contamination (e.g., Wick et al., 2012) or revealed rising tendencies of average nitrate concentrations following drought years (e.g., LUBW, 2018). However, we are not aware of a systematic study on regional-scale impacts of droughts on nitrate concentrations in groundwater that included wells and springs of diverse aquifer types. This is also important in the context of climate change, since potential impacts of climate change on nitrate concentrations in groundwater are complex and not yet well enough understood (Stuart et al., 2011). Droughts are one factor in this relationship, and the long-term effects of extreme drought events for groundwater processes like recharge are still unclear (Riedel & Weber, 2020). They have become more frequent in the last years in Europe (Jacob et al., 2012) and climate change projections predict a higher frequency of extreme events in future (Xu et al., 2019).
Here we used data from 19 years of a regional-scale groundwater monitoring network in southwest Germany and related site-specific nitrate concentrations to drought events. We included both well and spring data and aimed to contribute to the answers of the following The LUBW groundwater monitoring network contains both water quality and quantity data. Nitrate sampling and analysis is based on LUBW quality guidelines for groundwater sampling (LUBW, 2013), which set strict rules to provide a representative sample of groundwater at a given well screening depth. Stagnant water must be removed prior to sampling and a drawdown of the water table kept as low as possible. Springs are directly sampled at their source. All samples are collected in glass or polyethylene bottles and stored at 2-5 C until analysis the following day. Analysis is performed in laboratories that are accredited following DIN EN ISO/IEC 17025 standards and use liquid chromatography (DIN EN ISO 10304-1) or spectrophotometry . Analytical errors vary among the laboratories but are typically below 5%. While in our data nitrate sampling was irregular at monthly or seasonal intervals, spring discharges and groundwater levels were measured at daily or weekly intervals. We Since we were interested in event-specific nitrate dynamics and we assume that long-term changes in nitrate concentration lead to relevant biases in the analysis of post-drought nitrate dynamics, we first F I G U R E 1 (a) Location of monitoring sites in the study area Baden-Wuertemberg. (b) Proportion of groundwater wells and springs per aquifer type where β and y 0 are regression coefficients. Since 47 sites showed significant trends (p < 0.05) in nitrate concentrations and 64 sites in spring discharge or groundwater level (trends are reported in Data S1), we subtracted the linear trend from the parameter values X: where X detrended are the detrended parameter values with a mean of zero, that is, values indicate the deviation from the expected nitrate concentration. The detrended values were further analyzed regarding their relation to drought events. Corresponding to previous studies (e.g., Andreadis et al., 2005;Tallaksen et al., 2009), we used the 20th percentile as a fixed threshold value to define hydrological drought events. The use of percentiles ensured comparability among different sites (springs/wells) and data types (discharge/water level/nitrate concentration). We considered all days with groundwater level/spring discharge below the threshold a drought day, all other days were nondrought days. A drought event started with the first day below threshold and persisted until the last day below the threshold. To reduce the influence of small peaks on drought formation in groundwater levels and spring discharges, a pooling procedure removed insignificant and depended droughts (Fleig et al., 2006). Corresponding to Heudorfer and Stahl (2017) we excluded minor droughts shorter than 4 days and merged events with an inter-event time of 5 or less days. The day of maximum deviation from the threshold value during a drought event marked the end of drought development (Parry et al., 2016). In case of a multi-year drought, this day assigned the drought event to a specific hydrological year. Similar to the drought definition we assumed nitrate concentrations to be exceptional in 20% of the measurements, that is, the 80th percentile was used as a threshold value to classify nitrate samples as exceptional. For exceptional nitrate samples we then determined the last proceeding drought event. If several exceptional nitrate concentrations were related to the same drought event, only the first one was taken as the post-drought exceptional nitrate concentration.

| Nitrate responses to the 2003 drought
To investigate the site-specific nitrate response to drought, we focused on the severe 2003 Central European summer drought, which was one of the major drought events of the last decades in Central Europe (e.g., Fink et al., 2004;Van Lanen et al., 2016). We analyzed We applied further criteria to disentangle characteristic temporal patterns of nitrate responses. Sites with a post-drought exceptional nitrate concentration were classified as "direct response", if only a single drought event occurred before the nitrate concentration rose to exceptional levels. If two or more separate droughts occurred, the site was classified as "delayed response". Sites without post-drought exceptional nitrate increase were categorized as "no response" or "no recovery" depending on the occurrence of a recovery from the 2003 drought (i.e., spring discharge/water level above the 60th percentile before the end of 2007).
Differences in T and ΔC were further investigated using the analysis of variance (ANOVA). In an ANOVA, a sample is split for levels of a predictor and explained residual variances are compared. We used drainage type (well/spring), aquifer type (porous/fractured/porous and fractured/karst), land use (cropland/pasture/forest) and response type (direct response/delayed response) as predictors. If the ANOVA indicated that the split into predictor levels significantly reduced the residual variance, we applied the Tukey's honest significant difference test (TukeyHSD) as a post-hoc test to determine which predictor levels differed significantly from each other (Abdi & Williams, 2010).
For all tests we used a significance level of α = 0.05. Delay times T exhibited a large variability ranging from a few days up to more than a year (Figure 4). In accordance with the different maxima of nitrate responses in the annual time series (Figure 2(c)), T was shorter for springs (median = 101 days) than for wells (median = 185 days, Figure 4), though not statistically significant (TukeyHSD,p = 0.252). This corresponds to a faster drought propagation from precipitation to base flow than from precipitation to groundwater (Hellwig et al., 2020). Furthermore, the porosity (i.e., aquifer type) theoretically dictates the water volume and hence transit times in a groundwater system: the higher the water volume, the longer the mean transit time (if water fluxes are the same). Nevertheless, T did not significantly differ between aquifer types in our data (ANOVA, p = 0.13). However, four springs stood out and responded with T > 2 years. These long T might have been caused by missing responses due to a coarse seasonal sampling resolution.
Excluding these four outliers, the difference of T between the aquifer types became significant (ANOVA, p < 0.01). Then Drought induced increases of nitrate concentrations ΔC were significantly (TukeyHSD, p < 0.01) smaller for springs (median 1.3 mg/L) than for wells (median 3.4 mg/L, Figure 4). The surrounding land use types of wells were cropland and pasture, whereas springs were mostly located in forests. For these land use types ΔC differed significantly (ANOVA, p < 0.01) with ΔC in cropland being on average 4.6 mg/L higher than in forests (TukeyHSD, p < 0.01). This reflects the higher nitrogen input in cropland, which has been reported for other regions as well, for example, for the Central Valley, California (Ransom et al., 2017). However, the influence of aquifer type on ΔC was not significant (ANOVA, p = 0.09), because sample sizes were too different (Figure 1(b)). Lange and Hänsler (2012) reported similar ΔC between 0.2 and 4.0 mg/L in springs draining a crystalline fractured aquifer in a forested headwater and also the post-drought nitrate peaks of 0.9 mg/L measured in a well by optical sensors (Opsahl et al., 2017) fell in these ranges.
As a next step, we classified water quantity and quality dynamics and defined four characteristic response types ( sites with direct response were directly influenced by precipitation, whereas both the hydrograph and the nitrate concentration were smoother at sites with delayed response and responded only after a long period of strong precipitation ( Figure 5). In three wells, the water tables did not fully recover after the drought and nitrate concentrations remained below the 80th percentile until 2007 (type C, 4%). A group of three wells and two springs did not show any nitrate response (type D, 6%), despite a recovery from the 2003 drought.
Here, missing nitrate responses might have been caused by a too coarse sampling interval that excluded short-term exceptional nitrate peaks. According to the definition of the response types, T largely differed between response types A (direct) and B (delayed). However, the difference in ΔC was only 9% lower for type A compared to type B (Table 1)  Overall, our analysis underlined that regional patterns of quantitative groundwater responses to drought are variable in space and time, because they reflect the travel times of water from the land surface to the sampling location and thus include, apart from meteorological forcing, both unsaturated zone variability of recharge areas and different aquifer characteristics. Therefore, it is also not suprising that indices such as SPI or SPEI failed to predict the severity of European groundwater droughts (Kumar et al., 2016), because these indices are pure surface water measures. As up to 75% of the public water supply in our study region originates from groundwater wells or springs (LUBW, 2020) this analysis has a direct bearing on drinking water supply. Since droughts and extreme weather events are projected to occur more often in a changing climate (Xu et al., 2019), higher intensities and frequencies of dry-wet-cycles may be expected. Then considerable nitrate flushes after drought termination will occure more often. Further studies are needed to investigate potential consequences. These may include long-term increases of nitrate concentrations with negative effects on drinking water quality and on the ecological status of rivers, like for example, found in the upper Mississippi agricultural basin (Loecke et al., 2017). Extraordinary high post-drought mean nitrate concentrations found in this study for groundwater reservoirs of southwest Germany in 2006 (Figure 2(d)) should be regarded as a warning sign. when mean nitrate concentration in groundwater was highest. In general, it is concluded that the termination of a severe drought with following nitrate flushes may affect groundwater quality for several years, particularly in large, porous aquifers. As droughts and extreme weather events are projected to occur more often in a changing climate, higher intensities and frequencies of dry-wetcycles may be expected. Then nitrate flushes after drought termination could intensify problems regarding water quality and supply.

| CONCLUSIONS
Research and the Arts of the Land of Baden-Württemberg). Inga