Increasing influence of evapotranspiration on prolonged water storage recovery in Germany

Central Europe, including Germany, has faced exceptional multi-year terrestrial water storage (TWS) deficits since 2018, negatively impacting various sectors such as forestry, energy production, and drinking water supply. Currently, the understanding of the recovery dynamics behind such extreme events is limited, which hampers accurate water management decisions. We used a simulation of the mesoscale hydrological model (mHM) over the last 257 years (1766–2022) to provide the first long-term perspective on the dynamics of the TWS deficit recovery in Germany. The results show that severe TWS deficits surpassing a peak deficit of −42 mm (−15 km3) exhibit large variability in recovery times (3–31 months). The 2018–2021 TWS deficit period was unprecedented in terms of recovery time (31 months), mean intensity and the associated negative 30-year TWS trend. In recent decades, we identified increased evapotranspiration (E) fluxes that have impacted TWS dynamics in Germany. Increased E flux anomalies contributed to prolonged TWS recovery, given that the TWS deficit did not quickly recover through above-average precipitation (P). An extreme TWS deficit similar to that in 2018 was recovered by above-average P within three months in the winter of 1947–1948. Our research contributes to an improved understanding of the dynamics and drivers of TWS deficit recovery.


Introduction
Central Europe, including Germany, has been facing exceptional multi-year water deficits since 2018 (Boergens et al 2020, Hari et al 2020, Boeing et al 2022, Rakovec et al 2022).These prolonged water storage deficits negatively impacted various sectors, including forestry, energy production, and drinking water supply (Buras et al 2020, Madruga de Brito et al 2020, Schuldt et al 2020).Furthermore, terrestrial water storage (TWS) anomalies, as estimated by the gravity recovery and climate experiment (GRACE) satellite mission since 2002, have shown a substantial decline over the past 20 years in Germany (Güntner et al 2023).This has raised public concern regarding the beginning of a long-term negative TWS trend driven by climate change (BMUV 2023).However, the TWS time series of GRACE began with a very wet period, ended with an exceptional drought, and are still relatively short.Therefore, caution must be exercised when using these short time series to derive conclusions on the likely impacts of climate change on current and future TWS dynamics (Güntner et al 2023).
Although the onset and development of drought conditions have been extensively discussed, for example the phenomenon of flash droughts, the recovery of water storage deficits has received less attention (Van Loon 2015, Parry et al 2016).For instance, it is unclear how much time is typically needed for TWS deficits to recover, and how large the variability of TWS recovery times is.However, understanding the dynamics of storage deficit recovery is crucial for making informed water management decisions, particularly during prolonged deficit periods (Parry et al 2012(Parry et al , 2016)), in the context of drought termination forecasting (Pan et al 2013, DeChant andMoradkhani 2015), and under a changing climate.
TWS deficits occur when the TWS is lower than the monthly climatology over an extended period (typically a minimum of 3 months) (Thomas et al 2014).Since the TWS encompasses the integrated signal of all TWSs including snow, soil moisture, interception and groundwater (Famiglietti and Rodell 2013), a TWS deficit can combine different types of drought (Teuling et al 2013), involving water storages with varying response times to meteorological conditions (Buitink et al 2021).Typically, water deficits propagate through the hydrological system from fast to slow responding storages (Van Loon et al 2016, Tijdeman et al 2022).Among these, soil moisture storage exhibits the greatest variability (Rodell and Famiglietti 2001), responding more rapidly to changes in water fluxes, whereas groundwater responds more slowly and is also influenced by aquifer characteristics (Bloomfield and Marchant 2013, Stoelzle et al 2014, Kumar et al 2016).
During recovery from a TWS deficit, near-surface storages (e.g.soil moisture) are replenished first, while subsurface storages (e.g.groundwater) are likely to remain below normal levels (Thomas et al 2014).Ecosystem recovery, such as restoring plant functionality to pre-drought conditions, is usually delayed with the recovery of water storage deficits and can be further impaired by longer TWS recovery times (Schwalm et al 2017).Increased occurrences of extreme multi-year drought events may reduce the time between deficits, making ecosystem recovery less likely.The link between storage deficits and related ecosystem variables can be complex owing to legacy effects (Bastos et al 2020, van Hateren et al 2021, Pohl et al 2023).
In addition to precipitation (P) being the primary driver of changes in TWS (shortened to W), the progress of recovery from water storage deficits is controlled by the main water fluxes evapotranspiration (E) and runoff (Q), which represent the loss terms in the water balance equation (Orth and Destouni 2018): TWS recovery times are also likely to be affected by climate change impacts on the hydrological cycle.Similar P deficits, i.e. negative deviations from the mean climatology, could lead to longer and more intense droughts due to increased E under a warming climate (Teuling et al 2013, Samaniego et al 2018, Rakovec et al 2022).Climate modelling experiments have shown that the severity and frequency of the 2018 soil moisture drought would have been enormously enhanced under further warming (Aalbers et al 2023).However, increased water limitation in the soil storage in summer is expected to limit the increase of E during TWS deficit development (Denissen et al 2022, Aalbers et al 2023).On the other hand, during the recovery period of TWS, the upper soil storage may already be refilled, making E potentially more influential in prolonging recovery compared to deficit development.
Long-term analysis of the historic TWS deficit periods and recovery dynamics in Germany is lacking so far.To close this gap, our study examines the periods of TWS deficit and recovery in Germany from a water balance perspective, reconstructing historic data with the mesoscale Hydrological Model (mHM) covering the period from 1766-2022.Hydrologic simulations are suitable for analyzing long-term water storage dynamics, because of the availability of historical meteorological forcing data.In doing so, we place the recent negative trend and TWS deficits observed in Germany over the last 20 years within the context of a much longer historical time series.To understand the water balance fluxes driving TWS recovery, this study quantified the cumulative flux anomalies of P and E during the storage recovery periods relative to the mean monthly climatology.Our study aimed to address the following research questions: • What are the long-term dynamics of the recovery times of TWS deficits in Germany?• Which water fluxes influence TWS deficit recovery times and their variability in Germany?

Historical reconstruction of TWS deficit periods and storage recovery in Germany
The Meteorological forcing data for 1766-2022 were derived from a merged P and near-surface air temperature observation-based dataset from Casty-CRU (Casty et al 2007, Hanel et al 2018, Moravec et al 2019) for the period 1766-2015 and E-OBS v21 for the period 1950-2022 (Hofstra et al 2009).The overlapping period was used for bias-adjustment in the Casty-CRU data-set with respect to E-OBS.A detailed description of the dataset generation and validation can be found in Hanel et al (2018) and Moravec et al (2019).Hereafter, the long-term mHM simulation is referred to as Casty-CRU-EOBS.PET was estimated based on mean near-surface air temperature using the Oudin Formula (Oudin et al 2005).To derive the mHM-based TWS, all monthly averaged water storages in mHM, including soil moisture, saturated, unsaturated, snow, and interception storage, were considered.
The TWS simulations were compared with GRACE based satellite-derived monthly TWS anomalies from the combination service for time-variable gravity fields (COST-G) product (Jäggi et al 2020) for the period 2002-2022.The COST-G product combines all three official GRACE-TWS solutions and includes an uncertainty estimation.The anomalies were calculated based on the available data, considering the period of 2002-2022 as the monthly climatology.Since there are missing values in the GRACE data (e.g. during the change of GRACE to GRACE-Follow On in 2018), the simulation data was masked to match the available time steps of the GRACE data before calculating anomalies.Because of the low spatial resolution of the GRACE-based TWS, the analysis concentrated on the spatial mean for Germany to allow a comparison with GRACE and to reduce lower GRACE measurement and leakage uncertainties due to larger spatial aggregation (Save et al 2016, Scanlon et al 2016).Overall, we found a very good agreement in terms of correlation and a slightly higher deviation in the variability of the anomalies between the mHMand GRACE-based TWS (supplementary figure S2).The mHM Casty-CRU-EOBS based TWS shows a smaller variability in comparison to the GRACEbased TWS (70.4% explained variance), but still a high correlation of 0.87.Retaining seasonality in the TWS (removing only the long-term mean) resulted in a higher overall agreement between mHM-and GRACE-based simulations (r 0.91, expl.var.81%, supplementary figure S3).

Estimation of deficit periods and storage recovery times
TWS follows a strong seasonal cycle in temperate climates such as Germany (Humphrey et al 2016), driven by E and P seasonality.E and P show larger fluxes in summer than in winter, especially for E, which drives seasonal TWS reduction during the summer half year.Therefore, the seasonal cycle must be removed before estimating storage deficits.TWS anomalies are then defined as the deviation of the monthly TWS from their respective climatological values (Thomas et al 2014).The monthly TWS climatology is calculated as monthly averages over the entire period from 1766 to 2022.Following the approach of Parry et al (2016), the time during a storage deficit can be separated into three distinct phases: development, peak and recovery (supplementary figure S1).The deficit onset marks the time when the TWS negatively departs from the climatological average conditions.The peak of the deficit occurs when TWS reaches its minimum during the deficit period.The month of recovery is set to the month before the TWS surpassed the climatological average conditions again.The recovery time of the storage deficit is then estimated as the time from the month after the peak to the month of recovery.The total deficit length is determined from the onset to recovery.To exclude very short deficit periods, the analysis considers only those periods lasting for at least three months (Thomas et al 2014).The drivers of storage recovery are analyzed by calculating the cumulative anomalies for P and E during the recovery time of the deficit periods using the same climatology as for the TWS deficit estimation.To assess the trends in the TWS, P and E anomalies, we performed the Mann-Kendall test accounting for serial auto-correlation (Yue and Wang 2004) on the anomaly time series for the entire period and 30-year moving sections.

Reconstructed TWS deficit periods and recovery times 1766-2022
The first objective of our research was to analyze the long-term dynamics of TWS deficit recovery times in Germany.The time series of the reconstructed monthly TWS anomalies by the Casty-CRU-EOBS mHM simulation between 1776-2022 are shown in figure 1.The derived characteristics of the 10 most severe TWS deficit periods, namely: peak deficit, intensity, total duration, recovery time, and development time, are shown in table 1.In general, the derived characteristics of large TWS deficit periods are highly variable.The 1947-1948 event was estimated to have the strongest peak deficit of −101 mm (−36.2 km 3 ), followed by the ones noticed for the 2018-2021 event and the 1857-1860 event.The mean deficit ('deficit intensity') for the 2018-2021 event was significantly larger than that of other historical events with a magnitude of almost −55.1 mm (−19.7 km 3 ) mean water deficit.Over the last 257 years, the second most severe event was the period 1920-1922 with a mean deficit of around −46.7 mm (−16.7 km 3 ).The 2018-2021 event is well captured in the simulations compared to the GRACE TWS data

(figures S1 and S2
).There is some uncertainty regarding the TWS deficit recovery in 2021, as the combined GRACE COST-G solution did not show recovery for the mean estimate, but the recovery is still in the uncertainty range (figure S2).In general, the identified TWS deficit periods correspond well with other drought reconstructions in Germany (Glaser and Kahle 2020).
Figure 2 demonstrates that events with larger TWS deficits have high variability in their recovery times.The recent exceptional deficit period in 2018-2021 was characterized by the longest recovery time of 31 months.Furthermore, we identified that severe events with a peak water deficit of −42 mm (−15 km 3 ) or more, required a minimum recovery time of at least 3 months; and the events below this peak deficit could be recovered in 1 month.Overall, the mean recovery from TWS deficits with a peak deficit greater than −42 mm was approximately 11 months.While events with the most extreme deficits evolved relatively quickly (within 4-13 months), the 1862-1866 event showed prolonged development (41 months) and comparatively rapid recovery (11 months).A similar large variability in recovery times has been reported for river flows ranging from one season to multiple years (Parry et al 2016).Mo (2011) found that soil moisture and meteorological droughts typically took longer to develop (5-8 months) and recovery was less predictable, with the possibility of fast recovery within 2-3 months.Overall, the lengths of the investigated time series in these studies were much shorter compared to our study, which allowed for an improved representation of extremes using a very long time series (257 years).

Trends in TWS and water balance flux anomalies
Figure 3 depicts the annual variations and 30-year running mean for the yearly anomalies of TWS, P, and E with respect to the mean climatology of 1766-2022.Over the total period from 1766-2022, E and P show very modest but significant increases.Significant decreasing and increasing trends can be detected in the 30-years moving sections of annual TWS anomalies (represented by the dots in figure 3), with the strongest negative trend in the last 30 years (−1.8 mm y −1 /−0.64 km 3 y −1 ).The simulations suggest that E significantly increased after the 1990 s and was above the average for the past two decades.However, E did not increase significantly after 2010, restricted by water limitation, whereas PET increased continuously over the last decade (supplementary

Cumulative anomalies of water balance fluxes during TWS recovery periods
The second research question was to identify the influence of the water balance flux anomalies on TWS recovery in Germany and analyze its change over  3).The month of peak deficit and recovery are denoted for those deficit periods.Note that the axes have different limits to enhance the visibility.
time. Figure 4 shows the cumulative E and P anomalies during the recovery time of TWS deficits surpassing a peak TWS deficit of −42 mm.Furthermore, it shows that P is the main driver of TWS recovery duration.High positive P anomalies can terminate even extreme TWS deficits within a few months.The 1947 event represents such an ideal case of recovery from an extreme TWS deficit.However, for the 1947 event, there was only a 0.38% chance of reaching such a P surplus of +163.7 mm (+58 km 3 ) within three months in winter (supplementary figure S7).Therefore, a winter with a high P surplus may be sufficient to end extreme TWS deficits such as in 2018, but it is very unlikely to occur.Fast recovery times of extreme water deficits within one season with low climatic probabilities have been estimated for other regions and different drought types (Karl et al 1987, Mo 2011, Pan et al 2013, Thomas et al 2014, Parry et al 2018).It should be noted that during the progression of a TWS deficit, slower responding storages, such as groundwater, become increasingly depleted; hence, the potential recovery may be much slower when the storage deficit already prevailed for a longer time (Orth and Destouni 2018).In addition, further TWS recovery depends on conditions in the months following the recovery.
Generally, Q is typically strongly reduced during a deficit period, limiting the effect of P anomalies on storage deficits (supplementary figure S6), which corroborates findings of Teuling et al (2013) and Orth and Destouni (2018).During TWS deficit developmnet, cumulative Q flux anomalies are always negative.Yet, during the recovery phase cumulative Q flux anomalies can be positive for events with fast recovery (1947)(1948), which demonstrates that larger parts of excess P do not contribute to refill water storages.We also found that more recent TWS deficit periods with long recovery times (1972-1974 and 2018-2021) showed less pronounced Q reductions during the recovery phase in comparison to TWS deficits in the 18th century (1857-1860 and 1814-1815) (supplementary figure S6).
For E anomalies, the influence on TWS recovery is less clear.Long recovery periods may be associated with below-normal E, as in 1972-1974, due to belownormal PET.Instead, the 2018-2021 event was characterized by much higher than average E compared to other deficit periods with long recovery times, so the increased E may have played a critical role in extending the recovery time during this event.P during the recovery period was much higher than that during other periods with longer recovery times.Assuming that E during the 2018-2021 recovery period followed the long-term climatology instead of being higher than the average, similar to the theoretical experiment in Teuling et al (2013), the deficit would have recovered by winter 2020 due to a period of wet months (supplementary figure S7).It is important to note that this is only a thought experiment that demonstrates the potential impact of increased E on TWS recovery.The increased warming that drives above-average E also affects P intensity, and higher saturated soil moisture would have resulted in higher Q (which in turn would have reduced TWS recovery).One explanation for the increased E during recovery phase is that since the TWS is the integrated signal of different water storages, the root zone soil moisture is first again saturated, driving again E (in contrast during the development phase of the TWS deficit root zone soil moisture is reduced directly).Not considering the enhanced E during the TWS recovery period would lead to an underestimation of the recent TWS deficits due to the long-term overall increasing P sums.
Future water balance components will be affected by climate change, which, in turn, could impact future TWS dynamics and recoveries.Therefore, the combined seasonal interplay of the water balance components P and E (Mishra and Kumar 2020) as well as Q must be taken into account.Since P is the main driver of TWS dynamics, P trends are expected to largely determine occurrence of future droughts (Bevacqua et al 2022, Wang et al 2022).Changes in the seasonality of P (decrease in summer, increase in winter) or increasing P intensities (Pfeifer et al 2015) can affect future TWS deficit recoveries.Rising temperatures will lead to increases in PET.Furthermore, it is projected that water limitation will increase as a consequence of increases in net radiation and decreases in soil moisture (Denissen et al 2022).During TWS recovery (peak deficit to deficit end), potentially much more P is needed to recover from TWS deficits under future warming, as increased E from already recovering surface soil storages is driving TWS depletion.Global TWS simulations by Pokhrel et al (2021) using coupled global climate models (GCM) and various hydrological models indicate an increase in the frequency of extreme TWS droughts in (southern) Germany under climate change.Similar study setups using climatehydrology-ensembles could be used to further explore TWS deficit dynamics under climate change.

Limitations of the study
The TWS simulations are subject to different sources of uncertainty.Overall good agreement was found between mHM and GRACE-based TWS anomalies on the scale of Germany for the period 2002-2022 (supplementary figures S2 and S3).However, hydrological models tend to underestimate the TWS trends estimated by GRACE (Scanlon et al 2018).Güntner et al (2023) pointed out that signal leakage into Germany from glacier loss in the Alps overestimates the negative trends of GRACE-based TWS in Germany, which is also supported by this analysis as the simulations show smaller trend values and glacier loss is not captured by the mHM model.Furthermore, it is important to note that land use was kept constant in the Casty-CRU-EOBS simulations because sufficient land use maps for the complete time period were not available for Germany.This allows us to focus on the climatological effects on TWS dynamics; however, not considering land use change in the analysis introduces uncertainty in the simulation results, as land use and vegetation changes can exert a similar influence on changes in E as climate (Teuling et al 2019).Previous evaluations of the Casty-CRU-EOBS data-set demonstrated that the choice of the PET formula did not significantly modify the hydrological simulation results since 1900 (Hanel et al 2018, Moravec et al 2019).This is supported by the finding of Denissen et al (2022) that air temperature, used here for PET estimation, can be used as a simple and widely available proxy for energy availability that reproduces water limitation trends based on surface net radiation because of the high correlation between surface net radiation and air temperature on a monthly time scale.
Anthropogenic alterations of the water balance (groundwater abstractions, reservoirs, and water transfers), which were not included in the mHM simulations, can influence the TWS storage trends and storage deficit termination (Margariti 2019).On a regional scale, this can make a substantial difference, but for the spatial mean of Germany the influence of climatic factors is expected to dominate recent trends, such as estimated for groundwater storage (Xanke and Liesch 2022).Globally, climate contribution to land water storage has exceeded human intervention by about a factor of two over the past decade (Scanlon et al 2018).However, under continued warming, increases in water demand due to heat waves and crop irrigation may play a larger role in the TWS dynamics in Germany.Additionally, Alam et al (2021) highlighted the importance of considering water management implications for future deficit recovery in highly managed basins within the context of climate change scenarios (e.g.implementation of pumping restrictions can decrease recovery times).

Conclusions
Our study provides the first long-term perspective on the dynamics of TWS deficit recovery in Germany over the last 257 years (1766-2022) by reconstructing monthly TWS anomalies using mesoscale Hydrological Model (mHM) simulations.Our findings revealed substantial variability in the recovery times of severe TWS deficits, ranging from 3 to 31 months.The TWS deficit during the period from 2018 to 2021 was identified as highly exceptional with a mean water deficit of −55.1 mm (−19.7 km 3 ).It exhibited an unprecedentedly long recovery time of 31 months, with a notable mean intensity.Since the early 21st century, we observed an increase in evapotranspiration (E) fluxes, which had a pronounced effect on the TWS dynamics.These elevated E flux anomalies contributed to the extended recovery period of TWS, as the deficit did not replenish fast enough despite the higher precipitation (P) during the recovery period compared to other longlasting TWS deficit periods.In the historical context of Germany, an extreme TWS deficit comparable to that of 2018 was able to recover within just three months due to above-average P during the winter of 1947-1948.Our research is significant for adapting to current and future changes in the water cycle.Long-lasting TWS deficits can have negative implications on ecosystems and the economy.For instance, forests in Germany were severely impacted by the multi-year water deficits during 2018-2021 with large-scale canopy cover loss that accounts for about 5% of the forested area (5010 km 2 ) (Thonfeld et al 2022).While short-term droughts at the beginning of the vegetation season do not necessarily cause water stress if the soil is well saturated from the dormant season, and may coincide with high vegetation productivity in the case of high temperatures, long-term droughts are usually considered to reduce vegetation productivity (Pohl et al 2023).Under future warming, compounding heatwave and TWS deficits that are linked via soil moisture-atmosphere coupling will further increase the impacts on ecosystems (Chiang et al 2018, Zscheischler andFischer 2020).
Placing current TWS trends and shortages from a long-term perspective is crucial for understanding current and future changes in the water balance.The research also motivates other studies to consider more extended time series, which allow for a much larger sample of extreme events than using higher-resolution hydrologic simulation runs restricted to more recent time periods.So far, many essential aspects related to TWS dynamics need further investigation.Separation into different storages needs to be conducted to better understand the interplay of the components of TWS during deficit recovery.Spatially differentiated analysis can be performed to gain insight into the spatial patterns of TWS recovery.

Figure 1 .
Figure 1.Timeseries of the mHM-based monthly TWS anomalies for 1766-2022 from the Casty-CRU-EOBS run averaged over Germany.The TWS anomalies were derived by removing the mean monthly climatology of the respective time periods.TWS deficits (marked as red) occur when the TWS is below the monthly climatological mean.A TWS surplus (marked as blue) exceeds the climatological mean.

Figure 2 .
Figure 2. The figure displays the recovery time in months against peak TWS deficit in mm.The color denotes the mean deficit during total TWS deficit time and the size denotes the mean area in Germany under deficit conditions.The blue boxplot shows the distribution of recovery times for the TWS deficit periods with peak deficit larger than −42 mm (highlighted by the blue rectangle).The month of peak deficit and recovery time are denoted for the 10 largest deficit periods with respect to the peak deficit.

Figure 3 .
Figure 3. (A) yearly TWS anomalies 1766-2022, colored bars denote the length of the recovery periods.(B) and (C) depict yearly P and E flux anomalies.The thick line in all subplots is a 30 year running mean to depict long-term trends.The points denote significant positive or negative 30-year trends including the year where the point is marked and 29 preceding years.The mean climatology for the anomalies in (A)-(C) is calculated based on the 1766-2022 period.Note that the y-axes have different limits to enhance the visibility.

Figure 4 .
Figure 4. Cumulative flux anomalies (deviation from mean monthly climatology) of P and E during the recovery time for deficit periods with peak TWS deficits larger than −42 mm (see blue rectangle in figure3).The month of peak deficit and recovery are denoted for those deficit periods.Note that the axes have different limits to enhance the visibility.

Table 1 .
Characteristics of the 10 largest TWS deficit periods, ranked by peak deficit.