Introduction
Drought is a natural phenomenon occurring in all compartments of the
hydrological cycle. Accordingly, it is classified into meteorological
drought, hydrological drought, agricultural drought, socio-economic drought
and groundwater drought (Mishra and Singh, 2010). Due to the large number of
people affected by drought and the high economic loss related to drought
events (EC, 2007), it is important to enhance the understanding of drought
processes considering projected changes in drought hazard, most importantly
due to climate change. Empirical analysis of monitored hydrological time
series remains an important tool with which to validate theory-based or model-derived
hypotheses on these changes, since projected future changes are often
uncertain. Particularly for the groundwater compartment there is a high
diversity in response to climate input (Eltahir and Yeh, 1999; Green et al.,
2011), making predictions even more difficult. Most of the empirical studies
on long-term trends in drought hazard have focused on meteorological
drought (e.g. Sheffield et al., 2012; Spinoni et al., 2017) and on
hydrological drought (e.g. Stahl et al., 2010; Laaha et al., 2017). For
groundwater drought, empirical trend analysis is difficult for two reasons:
(i) groundwater time series are usually short or influenced by abstractions,
and (ii) where long and natural time series are available, they only give
point information.
Some countries and states now display groundwater level anomalies at observation wells
as part of their drought monitoring (e.g. Switzerland:
https://www.hydrodaten.admin.ch/de, last access: 28 November 2018;
Bavaria: https://www.nid.bayern.de/grundwasser, last access: 28 November 2018).
Studies on groundwater drought have mainly used borehole time series for index
generation and statistical analysis (e.g. Bloomfield and Marchant, 2013;
Bloomfield et al., 2015; Kumar et al., 2016; Van Loon et al., 2017;
Heudorfer and Stahl, 2017; Leelaruban et al., 2017). The groundwater response
to drought and its changes over time are typically very site specific, and
observations are therefore hardly scalable from point to catchment scale
(Kumar et al., 2016). However, larger regional-scale estimates of groundwater
resources are relevant for effective groundwater management and adaptation to climate change. Model-based
approaches for groundwater drought analysis and estimation (e.g. Peters et
al., 2003, 2006; Tallaksen et al., 2009; Li and Rodell, 2015; Apurv et al.,
2017) are one common way to get information on the catchment scale. There
are also a few studies that analyse spring data (Fiorillo and Guadagno,
2012) or baseflow time series (Brutsaert, 2008, 2010; Fendeková and
Fendek, 2012), two flux variables that provide a more integrative measure of
the groundwater situation during drought.
To overcome the difficulties related to borehole data, in this study we use a
baseflow approach to characterize groundwater drought on a catchment scale.
We analyse a large dataset of long baseflow time series in central Europe.
In this region groundwater is often used as drinking water, and aquifers act
as an important buffer to climatic variability. Most droughts start with a
deficit in precipitation, especially when precipitation falls as rain (Van
Loon and Van Lanen, 2012). For the propagation from a meteorological to a
groundwater drought different processes are relevant, i.e. attenuation,
delay and pooling (e.g. Peters et al., 2003; Tallaksen et al., 2009;
Heudorfer and Stahl, 2017). Therefore, the drought signal in groundwater
storage not only depends on current meteorological conditions, but also the
previous months are important. A catchment-specific timescale for this
dependence may be called the catchment's response time.
Response times have been analysed by correlations between groundwater depth
and time series of precipitation accumulated for different periods. Studies
have found that the response times for borehole water tables (Bloomfield et al.,
2015; Van Loon et al., 2017; Leelaruban et al., 2017) or spring
discharge (Fiorillo and Guadagno, 2012) vary strongly. Moreover, some
studies suggest time lags for the highest correlations between precipitation
and groundwater time series because of delayed groundwater response
(Bloomfield et al., 2015; Fiorillo and Guadagno, 2012). However, when
looking at monthly scales, this lag was always found to be quite small and
often non-existent (e.g. Haas and Birk, 2017).
There are two approaches to identifying drought periods in a time series. The
climatological approach is based on anomalies and often used also in
hydrology to track the propagation of relative seasonal water deficits
through the hydrological cycle (e.g. Barker et al., 2016; Kumar et al.,
2016). The traditional hydrological approach is the “threshold level
approach”, which defines streamflow droughts as events below a certain
fixed limit and is therefore focused on actual low water availability (e.g.
Yevjevich, 1967; Peters et al., 2006; Tallaksen et al., 2009). In this work
we use the term drought according to the threshold level approach; thus we
consider drought events as periods of low baseflow in absolute terms. If
there is a distinct seasonal regime, droughts mostly occur in the dry
season.
Recent work on central European low flows, i.e. periods when streamflow
mostly consists of baseflow, found that climate change is expected to alter
low flows (Marx et al., 2018; Forzieri et al., 2014; Van Vliet et al., 2015;
Gosling et al., 2017). However, the sign and magnitude of change in central
Europe are subject to model choice and emission scenario (Marx et al., 2018;
Forzieri et al., 2014). Those modelling studies were focused on large river
basins, and the change they predicted reflects strongly that of the
precipitation change. Marx et al. (2018) found a high correlation between
changes in annual precipitation sums and low flows. Stahl et al. (2012)
found that hindcasting summer low flow trends with large-scale models
suggests a too-homogenous spatial pattern of change compared to trends found
in observations. Together with difficulties of models to capture the
persistence of drought events in runoff generation found by Tallaksen and
Stahl (2014), it can be assumed that some large-scale models do not
necessarily resolve the heterogeneity of catchment storage and release for
the hydrological response on a headwater catchment scale. However, recent
drought events have demonstrated that especially headwaters are prone to
groundwater-related drought impacts like shortages in water supply (Van
Lanen et al., 2016). This coincides with findings that, independent from
elevation, groundwater is an important catchment storage (Staudinger et al.,
2017). Thus, predicting future changes in groundwater drought on a catchment
scale will be a prerequisite for effective drought management.
Depending on the projected climate change, different scenarios of the future
development of natural baseflow can be expected (Fig. 1). If there is an
increase (decrease) in precipitation projected for the entire year, flow
during the dry season is also expected to increase (decrease). However, if a
seasonal shift of precipitation is expected, the future development of flow
during the dry season is not that straightforward. It will depend on the
timing of seasonal shift and dry period and on the catchment's characteristic response time to precipitation. Stölzle et al. (2014)
found that for baseflow drought changes in precipitation are especially
relevant during the recharge period, which is dependent on the hydrogeology
of the catchment.
Schematic direction of change of natural baseflow in the dry
season under different climate change scenarios: (a) all seasons
becoming wetter, (b) all seasons becoming drier and (c) seasonal shift of precipitation.
![]()
For many parts of central Europe climate projections indicate a seasonal
shift of precipitation to wetter winters and drier summers rather than a
consistent increase/decrease (Jacob et al., 2014), urging for statistical
tools to assess the prospective changes in baseflow. Knowledge of the
seasonal to multi-annual scale of the baseflow response to climatic
variation and extremes is therefore particularly important in central Europe
under this seasonally diverging expected climate change.
This study aims to explain differences in drought trends by catchment
characteristics to allow for more accurate predictions under climate
projection uncertainty on a headwater scale. Firstly, past trends in baseflow
drought and catchment-relevant response times are analysed. Secondly, past
trends are attributed to climatic and catchment controls. Finally, based on
these statistics an estimate for future changes in baseflow drought valid
for all common emission scenarios and climate models is realized.
Study area and data
The dataset used in this study is the same set of headwater catchments (that
is, all catchment areas are below 200 km2) that were used in
Hellwig et al. (2018) to evaluate the representativeness of meteorological
grid data. It comprises streamflow data of daily resolution available from
the responsible environment agencies of the German federal states. The
lengths of available time series differ, so there is a trade-off between
selected period and the number of records with sufficient data. For this
study, we only used catchments with streamflow data of the period 1970 to
2009. Records with data gaps were not considered. Even though the dataset
consists of near-natural headwaters with minimal regulations, the
hydrographs and their double mass curves were visually screened for signs of
anthropogenic influence. In a natural catchment under relatively constant
conditions, the precipitation–streamflow relationship should be similar for
the entire observation period. Four of the gauges showed sudden changes in
the relationship and were subsequently removed. The final dataset consisted
of 338 unnested and independent gauges across Germany (Fig. 2).
Location of gauges in Germany, catchments' dominant type of porosity
(derived from the German hydrogeological map) and mean annual precipitation
sums (calculated from European Climate Assessment and Dataset E-OBS).
![]()
The selected catchments cover the flat lowland regions in the north of
Germany, the low mountain ranges in south-central Germany, and the
Alpine foothills and non-glacierized front range in the south. Precipitation
varies with highest annual precipitation sums in the Alpine south
(> 2000 mm) and lowest sums in the northeast (< 500 mm).
Climate in Germany is humid with slightly higher precipitation sums in
summer than in winter for most regions. Precipitation was analysed in the
form of monthly precipitation sums taken from the European Climate
Assessment and Dataset (Haylock et al., 2008), Version 13.1. According to
the procedure described in Hellwig et al. (2018), catchment-specific
precipitation was calculated as an area-weighted mean of the intersecting
grid cells. Hellwig et al. (2018) found that due to the low spatial
resolution of the meteorological dataset compared to catchment size there
are some biases towards products of higher resolution; however, correlations
between products were found to be very high.
Information on the hydrogeology of the catchments was taken from the digital
German hydrogeological map (Hydrogeologische Übersichtskarte von Deutschland 1:200.000, Oberer Grundwasserleiter; BGR and SGD, 2016). The catchments were
classified according to the main type of porosity found in the underlying
geology, either “porous” for porous aquifers such as unconsolidated
alluvial and glacial fillings or “fractured” for fractured bedrock. Where
less than two-thirds of the catchment's area is covered by one of these types, the
class “mixed” was assigned, including catchments which are dominated by
other types of porosity like karst. Additional analysis revealed that other
reasonable thresholds for classifying the catchments do not change the results.
According to this classification, 80 headwater catchments (23.7 %) are
mainly “porous”, 170 catchments (50.3 %) are mainly “fractured” and 88 catchments (26.0 %) have “mixed” porosity (Fig. 2).
The dominant land use was derived from CORINE Land Cover 2006 data
(available online from the German Environment Agency (https://www.umweltbundesamt.de,
last access: 28 November 2018). Each catchment is classified as agricultural
(53.6 %), urban (1.2 %) or forested (45.2 %).
To differentiate the catchments regarding topography, the average height
above the catchment's outlet was calculated. This metric describes the
potential groundwater waterbody that can contribute to the flow at the
gauge and thus characterizes the potential storage. Elevation data (1 arcsec digital elevation model over Europe) were obtained from the European
Environment Agency (2013).
Methods
Characterizing baseflow drought
Baseflow as a measure of groundwater storage at the catchment
scale
Baseflow Qb is defined as the delayed component of streamflow Q, i.e.
flow originating from stored sources as opposed to flow originating from
event water (WMO, 2008). In groundwater-dominated catchments, where
groundwater is the main water storage, Qb is directly related to
groundwater outflow (see Appendix A for a detailed rationale). Therefore,
Qb can be taken as an integrated measure of groundwater storage S as it
was done in the study of Fendeková and Fendek (2012) and others. It is
important to be aware that the unknown functional relationship between
Qb and S strongly differs between catchments and might be non-linear,
e.g. due to temporal changes in connectivity of groundwater and surface
water (Elthair and Yeh, 1999; Brunner et al., 2011). Hence, without further
knowledge about this functional relationship, the observation of Qb
allows for conclusions about S solely on an ordinal scale.
Despite the clear concept of baseflow, there is no universally valid way to
separate Qb from time series of Q. Instead, a number of methods exist
(e.g. WMO, 2008; Lyne and Hollick, 1979; Nathan and McMahon, 1990; Eckhardt,
2005; Mei and Anagnostou, 2015) that differ especially in baseflow
estimation during high flow, when Qb is relatively uncertain. For the
type of extreme low-flow conditions considered in this study, most methods
consistently assume that Q is almost completely comprised of Qb. The
World Meteorological Organization recommends a baseflow separation method
(WMO, 2008) that was applied in this work using the package “lfstat”
(Koffler et al., 2016) in the programming language R (R Core Team, 2016). The
separation procedure consists of five steps: (i) divide the time series into
non-overlapping blocks of a certain length N, where N is recommended to be
5 days for rainfall regimes which have a fast streamflow response; (ii) select the minimum value for every block; (iii) compare each minimum value
with the adjacent ones (if the central value is a times smaller than the
adjacent values, then it becomes a turning point; a is recommended to be 0.9 for rainfall
regimes which have fast streamflow response); (iv) join the turning points by
straight lines; and (v) compute the baseflow for every day by linearly
interpolating between the turning points (if the computed value is higher
than observed flow, use the observation instead).
Catchment-specific response times to precipitation
To quantify the catchments' baseflow response times to
precipitation (TR), time series of very different characteristics need
to be related to each other. To enhance comparability over space, time and
different parts of the hydrological cycle, it is common to standardize time
series. Recent studies on drought have used either parametric (e.g. Fiorillo and
Guadagno, 2012; Barker et al., 2016) or non-parametric approaches (e.g.
Bloomfield and Marchant, 2013; Kumar et al., 2016) for standardization.
Parametric approaches like the Standardized Precipitation Index (SPI),
introduced by McKee et al. (1993), rely on the fitting of a theoretical
distribution. Since time series of precipitation (P) and Qb have distinct
characteristics, entirely different distributions would have to be selected.
To ensure consistency in the way of standardization, in this study a
non-parametric approach was applied (Bloomfield and Marchant, 2013; Kumar et
al., 2016). All time series (P in accumulation periods from 1 to 36 months,
Qb) were standardized according to the procedure of the Standardized
Groundwater Index (SGI; Bloomfield and Marchant, 2013).
P (or Qb) was transformed by using the inverse normal
cumulative distribution function to calculate n equally spaced pi values
between 1/(2n) and 1-1/(2n), where n is the number of observations. These
values were afterwards assigned to the time series by reordering them
according to the ranks of the original time series. Hence, the ranks of the
original time series and the new ones coincide, but the newly generated time
series are normally distributed. Standardization was done separately for all
12 months of the year to account for the regime.
For all 338 catchments, correlation coefficients R between P and Qb were
calculated:
R=cov(x,y)σxσy=cov(xy),
where x and y are P and Qb, respectively, with cov() the covariance and σ the
standard deviation of each time series, which equals one for standardized
time series. To derive the catchments' response times to precipitation,
precipitation accumulation periods from 1 to 36 months were tested as well
as time lags of these indices to baseflow from 0 to 10 months. For each
catchment, the combination of precipitation accumulation period and time
lag that has the highest correlation coefficient was selected as the
optimal accumulation period and the optimal time lag to describe the
Qb response. Longer optimal accumulation periods indicate higher
groundwater attenuation of precipitation input, whereas higher time lags
represent a more delayed groundwater response to precipitation. To account
for seasonal variations in the response times, correlations were calculated
separately for all seasons (winter: DJF; spring: MAM; summer: JJA;
autumn: SON).
Catchment-relevant recharge period
Catchments analysed in this study follow a distinct regime in Qb. Thus,
droughts in Qb are most likely to occur in a certain period of the year.
To derive the catchment-characteristic drought season, annual minimal
7-day Qb (Qb7) were calculated. The most frequent month of
Qb7 was assigned as the month of highest baseflow drought hazard
(MQb7). Subsequently, the derived seasonal TR was used to determine
the months influencing Qb in MQb7. For example, if MQb7 in a
catchment was October and TR for that catchment in autumn was 6
months, then the catchment-relevant recharge period would be May through
October.
Detection of past baseflow drought trends
Time series of Qb7 for the time period of 1970–2009 were used to analyse
past trends in minimal baseflow. The analysis was carried out with the
non-parametric Mann–Kendall (MK) test, which is a common tool used to detect
monotonic increases or decreases in hydrological time series (e.g. used by
Douglas et al., 2000; Lins and Slack, 1999; Lorenzo-Lacruz et al., 2012;
Rennermalm et al., 2010; Asarian et al., 2016) and is robust to outliers. The
test was performed after pre-whitening the time series to account for the
influence of serial correlation (see Appendix B for calculation steps). A
significance level of α=0.05 was used.
Trends only give information on the period they are calculated for. Many
studies found that trends of a certain period are not part of a trend on
another timescale (e.g. Stahl et al., 2010; Hannaford and Buys, 2012;
Giuntoli et al., 2013; Hannaford et al., 2013). The analysis of trends for
multiple periods may help to assess whether observed trends are steady or
rather fluctuating. To evaluate the trends found for the period 1970–2009,
we additionally calculated trends over multiple periods for the five gauges
with longest continuous records. Three of these are in the porosity class
“fractured”, and one each in “porous” and “mixed”.
Attribution of baseflow drought trends
Trends in Qb7 were assumed to be driven by both climatic (climate
control) and catchment-specific (catchment control) factors. Thus, the
results were related to a set of potential predictors: (i) trends in P for
the catchment-specific relevant recharge period, (ii) TR for the
catchment-specific MQb7, (iii) MQb7, (iv) dominant porosity of the
catchment, (v) catchment size (A), (vi) dominant land use and (vii) potential
storage. Statistical significance of the factors was determined using the
correlation coefficient R (Eq. 1) for the continuous variables (trends in
P, A, potential storage), Spearman's rank-based correlation coefficient for
the discrete variable (TR) and an analysis of variance (ANOVA) combined
with the post hoc Tukey test (for details see Appendix C) for the
nominally scaled ones (MQb7, dominant porosity, dominant land use).
Estimate of drought sensitivity to expected future precipitation
change
Since past trends derived from empirical trend analysis (e.g. MK) are solely
valid for the observation period, they cannot be extrapolated beyond the
period of data availability. Future drought predictions therefore mostly
rely on climate projections and process modelling. For Germany climate
projections indicate little to no changes of annual precipitation sums but
seasonal shifts to lower summer precipitation and higher winter
precipitation. However, the magnitude of the shifts differs considerably for
different projections (Zebisch et al., 2005; Jacob et al., 2012, 2014;
Hübener et al., 2017). A common approach to dealing with such uncertainties
is to use a range of possible trajectories to model hydrological change.
Instead of using uncertain quantitative inputs in forward modelling, here we
propose a more qualitative inverse approach. We assume that the general
direction of future development is the most important piece of information for future
groundwater management planning and formulate a qualitative test scenario of
the consistent direction of different projections of future precipitation
change. Thus, the approach is scenario-neutral regarding emission scenarios
and climate models.
Trends in future baseflow drought hazard were assumed to depend not only on
precipitation changes but also on TR and MQb7 calculated for every
catchment. A particular test scenario can still have three possible outcomes
regarding future changes in baseflow drought hazard (Fig. 3): (i) increased
baseflow – there are relevant months with increasing precipitation but not
with decreasing precipitation; (ii) decreased baseflow – there are relevant months with
decreasing precipitation but not with increasing precipitation; or (iii) no change – either
there are relevant months with both increasing and decreasing precipitation
or there are no relevant months with any change.
The test scenario applied in this study is a decrease of precipitation in
summer (JJA) and an increase of precipitation in winter (DJF) with no change
in the annual precipitation sum. To derive the future change in Qb7 for
a catchment, we analyse whether months with increasing or decreasing
precipitation belong to the catchment-relevant recharge period. For example,
for a catchment with a relevant recharge period from May to October we would
estimate an increased baseflow drought hazard, since there are months with
decreasing precipitation in the relevant recharge period (June, July and
August) but no months with increasing precipitation. This analysis was
carried out for all 338 catchments.
Potential directional changes in Qb7 under a constant
seasonally diverging precipitation change scenario, depending on TR and
MQb7: (a) increase, (b) decrease and (c) no change.
![]()
Discussion
The catchment baseflow response to precipitation
Catchments' response times to precipitation were found to be
highly diverse across Germany, ranging from 1 month to 3 years. In
general, baseflow response times determined as a proxy of groundwater
response are rather short compared to other studies. Fiorillo and Guadagno (2012) found response times of 12 to 24 months for a karst region in
southern Italy and highest correlations for shorter precipitation accumulation periods when adding a short time lag. Bloomfield and Marchant (2013)
also found in 3 out of 14 cases a time lag for highest correlations. We
found time lags to be an exception, supporting the results of Kumar et al. (2016),
Barker et al. (2016) and Haas and Birk (2017). This indicates that,
in the headwater catchments studied, the delay of the groundwater baseflow
response to meteorological conditions may be shorter than 1 month and
therefore not detectable on the monthly scale, whereas the attenuation of
meteorological variability is clearly attributable to characteristic
precipitation accumulation periods.
The large differences of baseflow response times for different porosity
classes match the theoretical assumptions that baseflow strongly depends on
hydrogeological conditions. For the entire streamflow, differences were
found to be much smaller (not shown; compare e.g. Haslinger et al., 2014),
since other processes like overland flow are also important. The patterns of
TR which are related to the catchments' hydrogeology support the
assumption that baseflow can be used as a proxy of the groundwater situation
on a catchment scale. More detailed information on hydrogeology has the
potential to further improve the understanding of differences in baseflow
response. However, this information is not yet detailed enough for the scale
and distribution of headwater catchments. For example, two large-scale
estimates on hydraulic conductivity available for Germany (GLHYMPS from
Gleeson et al., 2014, and HÜK200 from BGR and SGD, 2016) were
considered and found to differ up to 6 orders of magnitude. Moreover,
information on superficial deposits that might be particularly important for
streamflow generation could enhance results but is not available
in detail yet.
Consistent with the work by Bloomfield et al. (2015), we found that
hydrogeology is a highly relevant factor for the explanation of different
groundwater baseflow response times. Kumar et al. (2016) did not find a
relationship between the hydraulic conductivity and groundwater response
time for boreholes. A possible reason for this different finding is that for
point data even small local influences, which are hard to determine, are
quite relevant (e.g. human influences), whereas baseflow reflects more the
overall situation within the catchment. Small influences may be negligible
at this scale, and the underlying influence of hydrogeology may be easier to detect.
In general, the results indicate that groundwater storage – represented by
baseflow – is vitally driven by precipitation on a catchment scale in the
relevant recharge period. However, the season of low flow is also expected
to have an influence: regimes with winter low flows in central Europe are
governed by snow storage during that season. Thus, not only precipitation
but also temperature is a major factor for that catchments. Moreover, it is
impossible to distinguish snowmelt from groundwater outflow during baseflow
separation. Therefore, a baseflow approach does not allow for conclusions on
groundwater storage in snow-dominated catchments.
Past and potential future changes in low baseflow
The trend analysis revealed that the trend in baseflow minima is highly
dependent on the period it is calculated for. The observation of a trend in
Qb7 thus is a poor indicator of future developments. Despite the
relatively narrow seasonality of low-baseflow timing (except for the alpine
areas), trends are highly variable across the region – a result of the
likewise variable response times. However, the correlation tests for
attribution of the trends revealed that precipitation during the
catchment-relevant recharge period is an important factor, which confirms
previous model-based findings for part of the region (Stölzle et al.,
2014). Thus, this catchment-relevant recharge period should be considered
for projections.
The method employed in this study provides a straightforward projection of
the probable future directions of changes in baseflow or
groundwater drought under climate change. We accounted for the uncertainty
in future climate projections by taking only concordant directions of
precipitation change in our approach. Contrary to future climate
projections, past trends in precipitation for the catchment-relevant
recharge period were found to be small and mostly positive. Similarly, Kopp
et al. (2018) found for southern Germany a high variability of annual
groundwater recharge without distinct trends. In the past, precipitation
trends have not been seasonally diverging, but climate models suggest that
changes will become more relevant in the second half of this century (e.g.
Jacob et al., 2012). As the magnitude of the trends differs for the climate
models, we did not quantify our scenario and thus did not quantify the
magnitude of future baseflow either.
Catchments' characteristic response times were assumed to remain constant.
Under a more extreme climate change however, changes in
catchments' responses (e.g. due to non-stationary response
times) cannot be excluded. Based on assumed precipitation change in the
catchments' respective recharge periods, decreasing Qb7 (i.e. increased
groundwater drought hazard) were estimated for the majority of the
catchments. This is because of the timing of MQb7, which occurs
predominantly in summer/autumn. Only those catchments with long response
times can compensate for the decreases in summer precipitation, and only in the
Alpine foothills, where MQb7 is in winter, is a decrease in drought hazard
predicted. However, in this region winter precipitation is mainly snow
and therefore does not immediately contribute to groundwater recharge.
Moreover, air temperature and its changes over time – which are not
considered in the test scenario – are especially important for winter
low flows. Hence, uncertainty is quite high for these projections. If
response times are not assumed constant, further conclusions may be drawn. A
decrease of response times might be caused by decreased storage, e.g. due to
urbanization, or accelerated drying and increased evapotranspiration in
spring. Such changes would increase the drought hazard for most parts of
Germany due to the timing of MQb7. An increase in response times by the
opposite processes is possibly less likely but could compensate for the reduced
precipitation during summer for the catchments that are currently estimated
to have increased drought hazard.