Introduction
Climate change is expected to alter the hydrological cycle and thus the
amount and timing of groundwater recharge, storage and discharge. The future
is likely characterized by more extreme hydrological events such as droughts
and floods . Predicting the impact of future
climate change on groundwater resources therefore requires a sound
understanding of the propagation of extreme events from the atmosphere to the
groundwater.
One approach to understanding the variability of groundwater levels is the
analysis of the aquifer responses to extreme events in the past
. However, fluctuations of
groundwater levels may not only be driven by hydrologic events. In
particular, changes in land use or water management are known to be
additional important factors . Evaluating long-term trends
or short-term fluctuations in groundwater level data, therefore, requires
careful consideration of the factors potentially controlling the observed
changes.
To be able to compare hydrologic extremes between different sites and
different types of data various indices have been employed. For instance, the
Standardized Precipitation Index (SPI) has been used to
identify and analyze the occurrence of extreme events in precipitation. Only
recently a corresponding Standardized Groundwater level Index (SGI)
has been proposed. SGI values computed for
observation wells in the UK as well as in
Germany and the Netherlands show significant correlation
with SPI values. However, the maximum correlation and SPI accumulation period
are found to differ between the sites. Thus, as noted by the authors of both
studies, groundwater levels and SGI values are influenced by the local
hydrogeological conditions.
This work aims to identify factors controlling SGI values of alluvial
aquifers within a mountainous region and its foreland (Mur valley, Austria).
In this type of setting, groundwater levels measured in the vicinity of
rivers are expected to show correlations with the river stage. Therefore, going beyond
earlier work, variations of standardized river stages are
considered in addition to SPI and SGI. To decipher influences of the local as
well as the regional hydrogeological setting correlations between the
standardized hydrological time series within three subregions are evaluated
and compared with each other. In addition, distinct drought and flood periods
as well as a snow-rich and snow-poor year are analyzed separately, as
groundwater levels are known to respond in different ways to floods and
droughts . Similarly, one may expect that groundwater
levels respond in different ways to abundant and deficient snowfall. Finally,
the time series are split-up in several multi-year periods to identify
potential long-term changes in the correlations between groundwater levels,
precipitation and river stages.
For this purpose, a novel approach employing correlation matrices is
proposed. We visualize these subregions, showing how they differ from each
other, how the different bodies of water are related to one another, how they
respond to extreme events and how the dynamics in the systems changes over
time. We use this approach to select single wells and discuss the limitations
of this approach.
Results
Average Pearson correlation coefficients of a subregion for SGI time
series with each other (SGI with SGI),
SGI time series with single SRSI time series (SGI with SRSI 1, …) and SPI
averaging periods (SGI with SPI1, …), and the average correlation coefficient
for all SRSI time series of a subregion with the SPI averaging periods (SRSI with SPI1, …)
with their standard deviations for each subregion.
SGI with
Location
SGI
SRSI 1
SRSI 2
SRSI 3
Pöls
Mur up
Mur down
Aichfeld shallow
0.59 ± 0.15
0.55 ± 0.14
0.50 ± 0.16
0.52 ± 0.19
Aichfeld deep
0.96 ± 0.031
-0.13 ± 0.022
0.045 ± 0.017
0.24 ± 0.015
Mur down
Übelbach
Mur up
Murdurchbruchstal
0.55 ± 0.27
0.55 ± 0.21
0.51 ± 0.29
0.60 ± 0.11
Mur up
Mur down
Laßnitz
Leibnitzer Feld
0.73 ± 0.14
0.16 ± 0.15
0.38 ± 0.11
0.44 ± 0.14
SGI with
SPI1
SPI3
SPI6
SPI9
SPI12
Aichfeld shallow
0.15 ± 0.085
0.47 ± 0.13
0.57 ± 0.11
0.47 ± 0.12
0.38 ± 0.12
Aichfeld deep
-0.039 ± 0.013
0.0049 ± 0.023
0.19 ± 0.048
0.32 ± 0.077
0.38 ± 0.081
Murdurchbruchstal
0.16 ± 0.060
0.41 ± 0.092
0.51 ± 0.090
0.49 ± 0.094
0.47 ± 0.090
Leibnitzer Feld
0.21 ± 0.12
0.58 ± 0.10
0.72 ± 0.069
0.68 ± 0.10
0.61 ± 0.11
SRSI with
SPI1
SPI3
SPI6
SPI9
SPI12
Aichfeld
0.27 ± 0.083
0.45 ± 0.077
0.48 ± 0.088
0.42 ± 0.086
0.34 ± 0.074
Murdurchbruchstal
0.26 ± 0.044
0.38 ± 0.064
0.39 ± 0.095
0.35 ± 0.11
0.37 ± 0.080
Leibnitzer Feld
0.31 ± 0.19
0.41 ± 0.20
0.36 ± 0.12
0.26 ± 0.095
0.24 ± 0.068
Correlation matrices for the three subregions (left side) with selected time
series for SGI, SPI and SRSI (right side). The data for each subregion is
sorted in three groups, divided by blank columns and rows: 1 – Groundwater,
SGI, sorted by distance of the well to the stream, given in meters on the top
of the matrices; 2 – Precipitation, SPI1, 3, 6, 9 and 12; 3 – Surface water,
SRSI, for different Mur gauges or streams in the subregion (U: Mur upstream,
D: Mur downstream, T: Tributary stream). Deep wells in the Aichfeld are
marked with an asterisk ∗. The three letter markers on the left highlight
selected wells and river stages discussed in the text; see also Table . Also
shown are the time periods used in Sect. and Fig. , the
years 1985/86 and 1989/90 used in Sect. and Fig. , and the years 2003 and 2009 used in Sect.
and Fig. . For
further details on correlation matrices, please refer to
Appendix .
![]()
Observations within the subregions
Aichfeld
In the Aichfeld subregion two patterns emerge (Fig. ); a
large area in the plot shows SGI time series (standardized groundwater levels
measured at different wells) that are highly to very highly correlated with
each other and with the SRSI time series (standardized river water levels at
measured at different gauging stations) in the subregion. The SGI time series
are from the wells situated closest to the river Mur on both riverbanks
(represented by well AAr). Most wells outside of the core of this region show
a similar behavior, resulting in an average Pearson correlation coefficient
of all of these SGI time series with each other of 0.59. These SGI time
series show a low correlation with the SPI1 time series and moderate to high
correlations with the longer SPI averaging periods, as expected from the
previous literature . The average
Pearson correlation coefficient of all of these SGI time series with SPI1 is
0.15, which raises to a maximum with SPI6 of 0.57 and decreases to 0.38 with
SPI12. The average correlation of the SGI time series with the SRSI time
series in the subregion is similar for all river gauging stations, with an
average of 0.52 (see “Aichfeld shallow” in Table ).
The second feature of the region are five wells (represented by well AAn in Fig. )
that show a very low to negative correlation of
their SGI time series with those of all other wells as well as with all SPI
and SRSI time series in the subregion, but are extremely highly correlated
with each other, with an average Pearson correlation coefficient of these SGI
time series with each other of 0.96, whereas the remaining wells have an
average correlation coefficient with each other of 0.59. This difference in
correlations is highly significant (p<0.01, t test). A look at the
underlying data reveals that the wells first discussed reach an end depth
significantly deeper (avg. 24.9 m b.g.l.) than that of the other wells in the
data set (avg. 9.7 m b.g.l.), so it is reasonable to assume that they show a
different, deeper aquifer system. This is also in accordance with
, who mentions that earlier wells of a similar depth for
the military airfield at this location encountered a conglomerate layer and
, who mention a significant groundwater inflow in this
area. The wells from the deeper aquifer also show a clear increase of
correlation of SGI time series with an increase in the length of the SPI
averaging periods, starting with an average correlation of the SGI time
series with the SPI1 of -0.04, reaching a maximum correlation of 0.38 with
the SPI12, which is significantly lower than the correlations seen in the
shallow wells SGI time series. The average correlations of the deeper wells
SGI with the SRSI time series range from -0.13 with the local Pöls to 0.24
with the downstream Mur. In the following, the focus is on the shallow
groundwater, but in some places we will consider the deep wells for
comparison.
The SRSI time series are correlated well with each other, indicating a
similar flow regime in the upstream and downstream Mur, as well as in the
tributary Pöls, but the correlations with the SPI time series are low to
moderate, ranging from an average of 0.27 with SPI1 to 0.48 with SPI6.
For further investigations, one of the wells from the shallow wells with the
highly correlated SGI time series and one well from the deeper aquifer have
been picked (see also Table ).
Well AAr, with a SGI time series highly correlated with most other shallow
wells SGI time series and closest to the river Mur, shows frequent changes
between wet and dry conditions of different lengths and magnitudes just as
the highly correlated AMr Mur gauge downstream of the subregion. Generally,
this fast changing well shows only moderate correlation with SPI time series,
no matter the averaging period. However, large events such as the 2002 and
2003 double drought are clearly visible.
Well AAn, situated in the deeper aquifer system and far away from the river
Mur, shows a much slower oscillation of the water levels, overlain by a
long-term trend from wet conditions into dry ones and then possibly back into
wet. Apart from large events, such as the double wet event in 1985 and 1986
and the double drought in 2002 and 2003, no similarities with the shallow
wells, the precipitation or the river gauging stations are obvious.
Murdurchbruchstal
In this subregion, the matrix visualization shows a picture noticeably
different from the upstream Aichfeld (see Fig. ).
A highly significant (p<0.01) difference is visible when comparing the SGI
correlation coefficients of the complete Aichfeld or the deep Aichfeld with
the Murdurchbruchstal, which is also reflected in the differences between the
average correlation coefficients of 0.39 for the complete Aichfeld, 0.96 for
the deep Aichfeld and 0.55 for the Murdurchbruchstal. The shallow Aichfeld
shows a groundwater signal similar to the Murdurchbruchstal and thus no
significant (p>0.05) change in the correlations. This is also shown by
similar average correlation coefficients of 0.55 for the Murdurchbruchstal
and 0.59 for the shallow Aichfeld (see also Table ).
As expected in a narrow valley with aquifers of small spatial extend, there
is a high correlation between SGI time series and SRSI time series. A cluster
of highly correlated SGI time series (represented by well MJc) is situated at
the furthermost distances to the river on its right bank, which are all –
except for one well – situated in the town Gratwein-Straßengel and are
also highly correlated with the SGI time series of the single well situated
in the neighboring town of Gratkorn on the opposite side of the Mur. This
cluster and the majority of the SGI time series in the subregion show high to
very high correlations with the SRSI time series. The average correlation for
the SGI with the SRSI time series is the highest for the upstream Mur gauge
with a Pearson correlation coefficient of 0.6 and the lowest for the local
tributary Übelbach with 0.5. Correlations with the precipitation are
generally the lowest with the SPI1 with an average of 0.16 and have the highest
correlations with the SPI6 and 9 with average Pearson correlation
coefficients of 0.51 and 0.49, respectively.
Surprisingly, some of the wells closest to the Mur on both sides of the river
are not very well correlated with each other and are also not among the wells
with the highest correlations between SGI and SRSI time series. In particular,
the matrix view shows three clear outliers (the well fourth closest and well closest
to the Mur on its left bank and second closest on its right, well MKr), whose
SGI time series are correlated very low or negative with the rest of the SGI
time series, but high to very high with each other. The pair of Mur-close
wells is situated in the same stretch of the river Mur opposite each other.
These wells are also the only wells that have SGI time series negatively
correlated with the SRSI time series in the system. For further
investigations, one of the wells from the cluster (MJc), the well closest to
the river Mur (MDp) and one well from the outliers also very close to the
river Mur (MKr) have been picked (see also Table ).
Well MJc is located centrally in the highly correlated cluster of wells and
shows a trend from mostly dry conditions to wetter conditions, which matches
the observation of the local tributary Übelbach (MUr). The SPIs for the
subregion show no such trends; however, the SPI6 and SPI9 show large dry
events in the period from 1980 to 1992, as well as the 2003 drought and 2009
flood. Some large events, such as the 2003 drought and 2009 flood are also
noticeable in well MJc, albeit not too significantly due to the underlying
trend from dry conditions to wetter conditions.
Well MKr is located very close to the river Mur, yet it shows no high
correlation with it. We observe wet conditions until 1999 and dry conditions
thereafter. Large events are also visible in this time series, albeit damped
or amplified by the change in conditions around 1999. Well MDp is located
very close to well MKr and very close to the river Mur and shows an opposite
change from dominant dry conditions until 1999 to wet conditions afterwards.
This phenomenon is discussed in detail in Sect. .
The river gauges SRSI time series are very highly correlated with each other,
but only show some minor correlations with SPI1 (average correlation
coefficient 0.26) and SPI3–9 (average correlation coefficients 0.35–0.39).
Leibnitzer Feld
In the Leibnitzer Feld, the situation is different again (see Fig. ).
Besides the fact that this region has a much higher
amount of groundwater wells, the matrix visualization again shows a very
different picture compared with the previous two subregions. These
differences in the correlations of the SGI time series in each subregion with
each other are highly significant (p<0.01) and are also reflected in their
average correlation coefficients of 0.59 for the shallow Aichfeld, 0.55 for
the Murdurchbruchstal and 0.73 for the Leibnitzer Feld.
Apart from a zone of wells with differing SGI time series on both benches of
the river (represented by well LUr) and some wells with moderately correlated
SGI time series on the left side, high to very high correlations of most SGIs
with each other prevail, resulting in an average Pearson correlation
coefficient of 0.73. Likewise high correlations of SGI time series with SPI
time series can be observed in almost all wells, with the highest
correlations found with the 6- and 9-month SPI, with average correlation
coefficients of 0.72 and 0.68, respectively. Unlike the other subregions, the
correlations of the SGI time series with the SRSI time series are generally
low to negative even for the SGI time series of wells very close to the Mur.
The lowest average correlation is seen at the upstream Mur with an average of
0.16 and the highest at the local river Laßnitz with 0.44.
It should be noted that part of this can be explained by the fact that the
Leibnitzer Feld is also a region where the Mur is heavily used for power
production, so the river levels and their fluctuations are not natural. Due
to the different times the dams have been built, it is also likely that
significant changes in the river regime have occurred during the life time of
the data set. In addition, both gauging stations for the Mur used for this
subregion are outside of the subregion and outside of the area of influence
of the power plants in the subregion, so they likely show a behavior
different from that of the river Mur within this subregion.
For further investigations, one of the wells from the highly correlated group
and one well close to the river have been picked (see also Table ).
Well LJc, whose SGI time series is highly correlated to most SGI values in
the subregion, shows frequent changes between dry and wet conditions.
Compared with the SPI1 or the river gauges LMr and LLr, it shows a smooth
signal visually similar to the highly correlated SPI6. Large events such as
the two droughts between 1976 and 1979 are also similar to the river Mur
(LMr) or river Laßnitz (LLr) in the case of the 2002 and 2003 droughts.
Average Pearson correlation coefficients of a subregion for SGI time series with each
other, SGI time series with the SPI6 and SGI time series with SRSI with their standard
deviations for each subregion for the complete time series, during drought (2003) and flood (2009) conditions.
Also shown is the p value indicating the statistical significance of the difference
between the correlation coefficients between the full time period, drought and flood conditions.
Location and
Avg. corr.
p value
Avg. corr.
p value
Avg. corr.
p value
type of data
coeff. ± SD
2003–
coeff. ± SD
2003–
coeff. ± SD
2009–
all time
all time
2003
2009
2009
all time
SGI, Aichfeld∗
0.59 ± 0.15
< 0.01
0.71 ± 0.21
< 0.01
0.13 ± 0.43
< 0.01
SGI, Murdurchbruchstal
0.55 ± 0.27
< 0.01
0.80 ± 0.23
< 0.01
0.50 ± 0.30
< 0.05
SGI, Leibnitzer Feld
0.73 ± 0.14
< 0.01
0.77 ± 0.19
< 0.01
0.70 ± 0.20
< 0.01
SGI-SPI6, Aichfeld∗
0.57 ± 0.11
< 0.01
0.83 ± 0.12
< 0.01
0.14 ± 0.30
< 0.01
SGI-SPI6, Murdurchbruchstal
0.51 ± 0.090
< 0.01
0.84 ± 0.19
< 0.01
0.53 ± 0.27
< 0.01
SGI-SPI6, Leibnitzer Feld
0.72 ± 0.069
< 0.01
0.83 ± 0.12
< 0.01
0.28 ± 0.15
< 0.01
SGI-SRSI, Aichfeld∗
0.52 ± 0.16
< 0.01
0.75 ± 0.19
< 0.01
0.20 ± 0.50
< 0.01
SGI-SRSI, Murdurchbruchstal
0.55 ± 0.22
< 0.01
0.81 ± 0.16
< 0.01
0.42 ± 0.30
< 0.01
SGI-SRSI, Leibnitzer Feld
0.32 ± 0.18
> 0.05
0.32 ± 0.31
> 0.05
0.27 ± 0.27
> 0.05
SRSI-SPI6, Aichfeld∗
0.48 ± 0.088
< 0.01
0.88 ± 0.051
> 0.05
0.27 ± 0.33
> 0.05
SRSI-SPI6, Murdurchbruchstal
0.39 ± 0.095
< 0.05
0.85 ± 0.020
> 0.05
0.34 ± 0.35
> 0.05
SRSI-SPI6, Leibnitzer Feld
0.36 ± 0.12
> 0.05
0.43 ± 0.24
> 0.05
0.41 ± 0.30
> 0.05
∗ Only shallow wells
The SGI time series of well LUr, situated right next to the river Mur, shows
only moderate correlations with most SGI time series in the subregion. Just
as well LJc, it shows frequent changes between dry and wet conditions. The
correlation is the highest with the SPI3 (not shown in Fig. ),
but despite a slightly lower correlation the SPI6 shows
a good visual fit with well LUr too. Large events such as the 1976–1979 and
2002–2003 droughts are visually similar to the river time series LMr and
LLr, but apart from that, the river gauging stations show a behavior
different from that of the nearby wells.
The abovementioned discrepancies in the water levels of the river Mur are also
visible in the correlations of the three river gauging stations SRSI time
series with each other. Here, unlike in the other regions, generally very low
correlations are seen not only when comparing the Mur with the Laßnitz –
which is expected due to their different catchments – but also when comparing
the two Mur stations, which would be expected to show a similar signal, if
they where behaving naturally. Only the local tributary Laßnitz shows a
moderate correlation with the 1- to 3-months SPI. For the average correlations
with the SRSI time series, the highest value is seen for the SPI3 with a
Pearson correlation coefficient of 0.41.
Selected flood and drought years
Correlation matrices for the three subregions, showing the effects
of the drought year 2003 and the flood year 2009. Legend for the colors
and description of the distances; see Fig.
![]()
Figure shows the correlation matrices for the
standardized time series for the well-known European drought year 2003 (see
for example , and
and BMLFUW, 2006 for Austria) and the local flood year
2009 (see BMLFUW, 2011 for Austria, and and for the Mur region).
In the 2003 drought year, the Mur catchment saw only 80 % of the 1961–90
average precipitation, 64 % of discharge at the Mur in Leoben (between the
Aichfeld and Murdurchbruchstal), 59 % of discharge at the Mur in Spielfeld
(downstream of the Leibnitzer Feld), compared with the 1991–2000 average
and a general reduction in groundwater levels (BMLFUW, 2006). In the
2009 flood year, the Mur catchment saw 123 % of the 1961–90 average
precipitation, 128 % of discharge at the Mur in Leoben, 135 % of discharge at
the Mur in Spielfeld, compared with the 1991–2000 average, and a general
increase in groundwater levels (BMLFUW, 2011).
Compared with the correlations over the total time period (see Fig.
and Table ), the drought year
generally, apart from the deep wells within the Aichfeld shows mostly highly
significant (p<0.01) higher correlations of the SGI time series with each
other, with the SPI6 time series and with the SRSI time series and higher
correlations between SRSI and SPI time series, albeit with differing
significance. The flood year shows mostly highly significant (p<0.01)
lower correlations than the drought year. Compared with the total time
period, the difference is not as visible as with the drought, which is also
visible in the somewhat reduced significance. The strongest difference
between flood and drought is visible in the Aichfeld, where negative
correlation prevails under flood conditions, going even lower than the -0.45 threshold chosen for the color scheme in the figures.
Another noticeable phenomenon is that certain wells can show a behavior that
strongly deviates from their average behavior and the general trends for a
given time span. For example well MFd in the Murdurchbruchstal and well ARf
in the Aichfeld are among the wells with highly correlated SGI time series in
their respective subregions for the complete time period (see Fig. )
but show low correlations under flood conditions (well
ARf, 2009, Fig. ) or drought conditions (well
MFd, 2003, Fig. ). Well MFd shows less wet
conditions in spring and is less affected by the 2003 summer drought than
most other wells in the subregion. Well ARf shows a drier spring and winter
than most other wells in the subregion during the wet year 2009.
Selected snow-rich and snow-poor years
Average Pearson correlation coefficients of a subregion for SGI time series with each
other, SGI time series with the SPI6 and SGI time series with SRSI with their standard
deviations for each subregion for the complete time series, during snow-rich (1985/86) and snow-poor (1989/90) conditions.
Also shown is the p value indicating the statistical significance of the difference
between the correlation coefficients between the full time period, snow-rich and snow-poor conditions.
Location and
Avg. corr.
p value
Avg. corr.
p value
Avg. corr.
p value
type of data
coeff. ± SD
1985/86–
coeff. ± SD
1985/86–
coeff. ± SD
1989/90–
all time
all time
1986/86
1989/90
1989/90
all time
SGI, Aichfeld∗
0.59 ± 0.15
< 0.01
0.79 ± 0.19
< 0.01
0.29 ± 0.46
< 0.01
SGI, Murdurchbruchstal
0.55 ± 0.27
< 0.01
0.79 ± 0.16
< 0.01
0.52 ± 0.31
> 0.05
SGI, Leibnitzer Feld
0.73 ± 0.14
> 0.05
0.73 ± 0.26
< 0.01
0.40 ± 0.44
< 0.01
SGI-SPI6, Aichfeld∗
0.57 ± 0.11
< 0.05
0.72 ± 0.21
< 0.01
0.22 ± 0.34
< 0.01
SGI-SPI6, Murdurchbruchstal
0.51 ± 0.090
< 0.01
0.77 ± 0.10
< 0.01
0.026 ± 0.31
< 0.01
SGI-SPI6, Leibnitzer Feld
0.72 ± 0.069
< 0.05
0.80 ± 0.21
< 0.01
0.45 ± 0.27
< 0.01
SGI-SRSI, Aichfeld∗
0.52 ± 0.16
< 0.01
0.74 ± 0.16
< 0.01
0.33 ± 0.50
< 0.05
SGI-SRSI, Murdurchbruchstal
0.55 ± 0.22
< 0.01
0.67 ± 0.18
< 0.01
0.44 ± 0.32
< 0.05
SGI-SRSI, Leibnitzer Feld
0.32 ± 0.18
< 0.01
0.60 ± 0.16
< 0.01
-0.055 ± 0.41
< 0.01
SRSI-SPI6, Aichfeld∗
0.48 ± 0.088
> 0.05
0.62 ± 0.14
> 0.05
-0.24 ± 0.23
< 0.05
SRSI-SPI6, Murdurchbruchstal
0.39 ± 0.095
< 0.05
0.72 ± 0.042
> 0.05
-0.18 ± 0.41
> 0.05
SRSI-SPI6, Leibnitzer Feld
0.36 ± 0.12
< 0.05
0.66 ± 0.099
> 0.05
-0.14 ± 0.44
> 0.05
∗ Only shallow wells
Correlation matrices for the three subregions, showing the effects
of the snow-rich winter of 1885/86 and the snow-poor winter of 1989/90. Legend
for the colors and description of the distances; see Fig.
![]()
In order to compare the effects of a snow-rich and a snow-poor year on the
groundwater system, we selected the winters of 1985/86 (snow rich) and
1989/90 (snow poor). In 1985/86 (November 1985–October 1986), the average snow
height (including the summer months) was 11.98 cm in the Aichfeld, 9.4 cm in
the Murdurchbruchstal and 6.2 cm in the Leibnitzer Feld, with cumulated fresh
snow of 390 cm in the Aichfeld, 274 cm in the Murdurchbruchstal and 193 cm in
the Leibnitzer Feld. In 1989/90 (November 1989–October 1990), the average snow
height was 0.32 cm in the Aichfeld, 0.11 cm in the Murdurchbruchstal and 0.04 cm in
the Leibnitzer Feld, with cumulated fresh snow of 55 cm in the
Aichfeld, 23 cm in the Murdurchbruchstal and 9.3 cm in the Leibnitzer Feld.
Compared with the correlations over the total time period (see Fig.
and Table ), the snow-rich year generally
shows higher correlations between the SGI time series with each other and the
SGI and SRSI time series, whereas the snow-poor year shows lower correlations.
Similar to the situation with flood and drought (see Sect. ) most of the differences are highly significant
(p< 0.01) or significant (p< 0.05), although there are some non-significant
differences (see Table ). Comparing the snow-rich year with
the snow-poor year, all of the differences, except for the correlations of
the SRSI with the SPI6 time series, are highly significant (p< 0.01).
In all cases, some patterns also visible in Figs. and remain. The set of five deeper wells in the
Aichfeld is almost always visible, but appears clearest for the years
1985/86, with a sixth well showing a similar behavior under these
conditions. The highly correlated clusters close to the river in the Aichfeld
and the Murdurchbruchstal also prevail, as do the two clusters in the top
left and the bottom right of the Leibnitzer Feld.
Development over time
Correlation matrices for the three subregions split into
three time periods. Note that the first period for the Murdurchbruchstal
is from 1980 to 1986 due to lack of data before 1980.
Legend for the colors and description of the distances; see Fig.
![]()
Figure and Table show the development of the three
subregions when split-up into time periods of 12 years (1975–1986, 1987–1998,
1999–2010). It should be noted that the Murdurchbruchstal only got a
significant number of groundwater wells after 1980, so the first time period
differs for this region, and is only 7-years long, from 1980 to 1986.
In the Aichfeld, there is no noticeable trend over time, besides the
deviating behavior of the deep wells in the last period. As mentioned in
Sect. , we are going to focus on analyzing the
shallow wells in the Aichfeld. From the first to the second period, we see an
increase in SGI correlations for a cluster of wells around the river and thus
an increase of correlation of those wells SGI time series with the SRSI time
series. In the last period, these correlations decrease. These small changes
in the correlations of the SGI time series are also reflected by their
average correlation coefficients (see Table ). However, the averages do not
necessarily reflect the significance of the change. While the second and last
time period have similar averages, the change in the underlying set of
SGI-SGI correlations is still significant (p< 0.05), as it is the case with
the SGI-SPI6 correlations, which also show a significant (p< 0.05) change
from the second to the last period. On the other hand, the average
correlation coefficients can show noticeable changes between the time
periods, but the changes are not significant (p> 0.05), as it is the case
for the SGI-SRSI correlations.
The Murdurchbruchstal shows similar behavior in the first and second period,
with some slightly different clusters. In the first period, the upstream and
downstream Mur gauges show SRSI time series highly correlated with each
other. In the last period, we see higher correlations of all SGI time series
with each other, the SPI time series and the SRSI time series, with only the
1-month SPI and the downstream Mur gauge showing some low correlations.
These visible changes are also reflected in the average correlation
coefficients for the SGI time series within the subregion (see Table ).
Highly significant (p< 0.01)
changes occur in the SGI-SGI and SGI-SPI6 correlations between the second and
last period, as well as for all periods for the SGI-SRSI correlations.
The Leibnitzer Feld also shows a slight decrease in correlations in the
middle period, followed by a strong increase in the last time period.
Compared with the complete time period shown in Fig. ,
the Leibnitzer Feld shows higher correlations of SGI time series with the
SRSI time series for the shorter time periods, but wells close to the river
show a comparably lower correlated SGI time series. The mentioned decrease
followed by an increase is reflected by highly significant (p< 0.01) changes
of the correlation coefficients for the SGI time series within the subregions
for the first, second and third time period. The correlations of SGI and
SPI6 also seem to follow the decrease–increase pattern, with highly
significant (p< 0.01) changes between all three periods. Only the SGI-SRSI
correlations deviate from the general pattern and show no significant (p> 0.05) change between the first and the second period.
Average Pearson correlation coefficients of a subregion for SGI time series
with each other, SGI time series with the SPI6 and SGI time series with
SRSI with their standard deviations for each subregion and each time period.
Also shown is the p value indicating the statistical significance of
the difference between the correlation coefficients of the two time periods.
Location and
Avg. corr.
p value
Avg. corr.
p value
Avg. corr.
type of data
coeff. ± SD
coeff. ± SD
coeff. ± SD
1975a–1986
1987–1998
1999–2010
SGI, Aichfeldb
0.63 ± 0.16
> 0.05
0.62 ± 0.15
< 0.05
0.66 ± 0.11
SGI, Murdurchbruchstal
0.64 ± 0.19
> 0.05
0.62 ± 0.20
< 0.01
0.75 ± 0.10
SGI, Leibnitzer Feld
0.77 ± 0.12
< 0.01
0.71 ± 0.15
< 0.01
0.80 ± 0.14
SGI-SPI6, Aichfeldb
0.62 ± 0.14
> 0.05
0.55 ± 0.075
< 0.05
0.67 ± 0.072
SGI-SPI6, Murdurchbruchstal
0.61 ± 0.12
> 0.05
0.48 ± 0.13
< 0.01
0.67 ± 0.057
SGI-SPI6, Leibnitzer Feld
0.76 ± 0.069
< 0.01
0.69 ± 0.082
< 0.01
0.74 ± 0.060
SGI-SRSI, Aichfeldb
0.58 ± 0.13
> 0.05
0.55 ± 0.19
> 0.05
0.58 ± 0.17
SGI-SRSI, Murdurchbruchstal
0.44 ± 0.26
< 0.01
0.56 ± 0.17
< 0.01
0.65 ± 0.11
SGI-SRSI, Leibnitzer Feld
0.41 ± 0.15
> 0.05
0.38 ± 0.21
< 0.01
0.46 ± 0.15
a For the Murdurchbruchstal this period is from 1980 to 1986
due to lack of data before 1980; b Only shallow wells
Discussion
Spatial variability
As already shown in Sect. , a large number of
groundwater wells in each subregion shows SGI time series highly correlated
with each other. Some of those wells are also in close vicinity to each other
(e.g., the cluster of highly correlated wells in the Murdurchbruchstal
subregion, all located in the town Gratwein-Straßengel), to the river Mur
(e.g., most of the shallow wells in the Aichfeld subregion) or located in a
similar geologic setting (e.g., the deep wells in the Aichfeld subregion or
almost all the shallow wells in the Leibnitzer Feld subregion).
As a result of the different behavior of the groundwater wells in the
different subregions, the correlations of SGI time series with the SPI time
series also differ between the subregions (see Table ).
While the SPI1 still shows similar, low average correlations, the longer SPI
averaging periods show a different behavior in the subregions. Hence, we are
only discussing the higher averaging periods, except for parts of the
Aichfeld.
Since there are two distinct aquifer bodies in the Aichfeld, the groundwater
data was split-up into a shallow (average depth of the wells: 9.7 m) and a
deep (average depth of the wells: 24.9 m) part. The deep wells SGI time
series are only lowly correlated with SPI time series, with a minimum for the
SPI1 of -0.04 and a maximum of 0.38 for the SPI12. The average SPI-SGI
correlations for the shallow wells range from 0.38 for the SPI12 to a
maximum of 0.57 for the SPI6.
In the Murdurchbruchstal, where all of the wells have similar depths
(average: 10.7 m), the average correlations between SGI and SPI time series
range from a minimum of 0.41 for the SPI3 to a maximum of 0.51 for the SPI6.
The wells in the Leibnitzer Feld also have similar depths (average: 6.4 m).
Here, the average correlations between SGI and SPI time series range from
0.58 for the SPI3 to 0.72 for the SPI6.
All subregions (or the shallow part of the subregion in the case of the
Aichfeld) have the highest correlation with the SPI time series for an
averaging period of 6 months. Only the deep part of the Aichfeld has its
maximum correlation with the 12-month SPI, which fits the findings of
, who found that deeper wells correlate better with longer
SPI averaging periods.
The SPI6–SGI correlations follow the average depths of the wells, with the
highest correlation found in the most shallow Leibnitzer Feld, and the lowest
correlation found in the deep part of the Aichfeld, a pattern that is also
repeated for all other averaging periods. The shallow part of the Aichfeld
and the Murdurchbruchstal have very similar average depths (9.7 and 10.7 m,
compared to 24.9 m for the deep Aichfeld and 6.4 m for the Leibnitzer Feld),
so that they show similar correlations, ranging between those of the deep
wells in the Aichfeld and the shallow wells in the Leibnitzer Feld.
In all regions, there is a low correlation between standardized river stages
and standardized precipitation, with an average correlation coefficient for
SPI with SRSI ranging from 0.47 in the Aichfeld to 0.37 in the Leibnitzer
Feld (see also Table ), with the highest correlations
between river and precipitation generally found for the 3- and 6-month SPI.
This suggests that, in addition to the transformation of the rainfall signal
due to the runoff processes within the subregion, the rivers can transport a
precipitation signal from a region upstream of the subregion in question,
which can have a different precipitation signal from the local precipitation.
This is also supported by the fact that the differences between the
correlations of the SRSI time series from each subregion with the SRSI time
series from the different subregions appear not to be significant (p> 0.05).
Also, this upstream signal can in itself be a “collection” of many
different regional precipitation patterns. This suggests that the correlation
of the SRSI time series with the 3- and 6-month SPI results from the influence
of the large, general “climate” in the region.
Another factor affecting the rivers are the numerous run-of-the-river power
plants, which alter the natural course and timing of the rivers and remove
their natural short-term precipitation signal. For the Aichfeld, where there
are only five small-scale power plants in its upstream part, this does not
affect the river Mur too much, shown by the high average correlation of the
river gauging stations SRSI time series with each other of 0.65. A similar
value of 0.61 is observed in the Murdurchbruchstal, even though there are eight hydro
power plants in the subregion. In the Leibnitzer Feld, however, the
combination of five power plants, and the fact that the gauging stations are
located outside of the subregion results in an average correlation of the
SRSI time series with each other of only 0.17. However, as mentioned above,
the differences in SRSI time series between the subregions are still not
significant (p> 0.05).
Thus, in small systems such as the Aichfeld and the Murdurchbruchstal – and to
some extent probably also the Leibnitzer Feld – the river and the
groundwater will be closely related to each other. At high water levels, the
river feeds the groundwater, thus superpositioning its signal onto the
groundwater, whereas the groundwater provides the river baseflow in low water
conditions, thus controlling river flow and river stage at low water levels
(see also Sect. ).
In summary, the most obvious differences between the subregions are the low
correlation of the river gauge SRSI time series with the groundwaters SGI
time series in the Leibnitzer Feld, described in detail in Sect. , and the differences between SGI-SPI
correlations, where Aichfeld and Murdurchbruchstal show generally low to
moderate correlations, and the Leibnitzer Feld shows generally high to very
high correlations, following the thickness of the aquifers in the subregions.
Selected flood and drought years
As shown in Fig. and Sect. , the drought and flood years of 2003 and 2009
show a very different behavior in the regions investigated herein. As
mentioned in Sect. , we are mainly going to discuss
the shallow Aichfeld, since the shallow aquifer is directly affected by this
relatively short-term events.
Generally, we see an increase in correlations under drought conditions and a
decrease under flood conditions, which is not only reflected by the color
coded, single correlations coefficients shown in Fig. but also by most average correlations
coefficients shown in Table . Apart from the
correlations between the SGI and SRSI time series in the Leibnitzer Feld and
all of the correlations between the SRSI and SPI6 time series which do not
show a significant change (p> 0.05), all of the differences between the 2003
drought year and the 2009 flood year are highly significant (p< 0.01).
In order to interpret these differences, it is important to look at the
differences in the underlying drought and flood. As shown in
Sect. , the 2003 drought was a long-term and
large-scale event, affecting all of Europe for most of the year (e.g.,
and
BMLFUW, 2006). The 2009 flood on the other hand, was a more small-scale
event, split-up into multiple flood peaks (e.g., and BMLFUW, 2011).
The 2003 deficit of only 59 % of discharge at the Mur gauge in Spielfeld
(BMLFUW, 2006) was the result of long-term and country-wide dry
conditions, whereas the 2009 excess discharge of 135 % in Spielfeld
(BMLFUW, 2011) is the result of multiple flood events, often very
localized in the small tributaries to the Mur , partly also resulting in considerable, localized overbank
flow. While the 2003 drought showed a slow decrease in water levels in the
aquifer and the rivers, the 2009 flood showed fast increases in water levels,
which in case of the rivers get transported downstream to an area that might
not be affected by a localized precipitation maximum.
The observation that a long-term drought affects the whole aquifer and that a
short-term flood only affects parts of the aquifer fits the idea of aquifer
response times . The aquifer response time
T∗ is a function of storativity (S), “some characteristic length” of
the aquifer (L) and the transmissivity (T): T∗=S×L2T. We approximate S for our unconfined case by the specific yield
(Sy) and T by multiplying the average K of a subregion with its average
saturated aquifer thickness (see Sects. ,
and ). For
Sy we use a value close to the average porosity of 22 % compiled by
. For L we are using the average distance perpendicular to
the river Mur within which most wells of a particular subregion are situated.
With these values and assumptions, we obtain values for T∗ ranging from
over 1 month to over 1 year. Thus, a short event such as a flood will not
affect the whole aquifer, whereas a long-term event such as the 2003 drought
affects the whole area or at least most parts of it. The aquifer response
time also offers a possible interpretation of the deeper aquifer in the
Aichfeld, which generally shows high correlations of SGI time series with
each other, irrespective of conditions, but especially so under flood
conditions (see Fig. ). The deep aquifer
is likely confined or semi-confined so that the storativity S is orders of
magnitude lower than the Sy of the shallow unconfined aquifers, and thus
results in response times from hours to days. This allows for all of the
wells in question to react to a perturbation, such as a short flood, well
within the 1-month timescale shown in the correlation matrices.
The phenomena discussed above also match the findings of
, who stated that droughts have a much more
“persistent signature on groundwater hydrology, in comparison to […]
floods”. They suggest that floods – increases in groundwater levels – can
dissipate very quickly by groundwater discharge,
whereas there is no dissipation mechanism available for low groundwater levels.
Following this interpretation, argued that this
explains the asymmetry of the water levels response to a flood or drought
event and suggest that this mechanism deserves further investigation. We
argue that this asymmetry is seen not only in a single hydrograph but also
in the whole area, resulting in the different pictures shown in Fig. , where only the SPI1 shows similar correlations
under flood and drought conditions.
Looking at the parts of the aquifers not influenced by rivers, an increase in
precipitation will increase infiltration and thus simply increase the water
levels, keeping the general flow direction and thus correlations between
neighboring wells time series intact, shown by the areas of high correlations
in Fig. . However, looking at the parts of the
aquifer close to the river – which includes many wells that are close to
small creeks and streams that are not considered for the general discussion
in this paper – a multitude of possible phenomena is seen. As a direct
pathway, bedload during floods can erode the clogging layer in the river bed
and thus provide a significant short-time improvement in infiltration
. showed that river floods can
transport pressure pulses in highly conducting channels, as described in
. A similar phenomenon, is shown by
following floods through the aquifer for distances of over 2 km.
described wells at similar distances that show a strong
and fast reaction to a river flood within 1.5 to 6 days, both with inundation
and without. In a further paper, argued that “overbank
flood recharge is not an insignificant volume”. As discussed in
, flood events – with overbank flow – can make up
significant parts of the recharge in river-close parts of an aquifer.
The mechanisms described above can result in two phenomena besides the still existing baseflow:
a pressure pulse propagating through the aquifer or a real and rapid infiltration,
both being oriented against the usually dominating flow towards the river, and a potential for local backwaters
where the inflow from the river and the baseflow towards the river meet. This
results in similar changes in all of the aquifer under normal and drought
conditions, resulting in high correlations, whereas flood conditions can
cause differing changes in the aquifer, resulting in low correlations.
Selected snow-rich and snow-poor years
As shown in Fig. and Sect. , the snow-rich and snow-poor
years of 1985/86 and 1989/90 show a very different
behavior in the regions investigated herein. As mentioned in Sect. ,
we are mainly going to discuss the shallow Aichfeld,
since the shallow aquifer is directly affected by this relatively short-term
events.
Average values (dotted lines) for the SGI (blue), SPI1 (yellow)
and SRSI (red) and their 5-year-running means (solid lines) for the Aichfeld subregion.
![]()
Generally, we see an increase in correlations under conditions with a lot of
snow and a decrease under conditions lacking snow, which is
reflected not only by the color coded single correlation coefficients shown in Fig.
but also by most average correlation coefficients shown in
Table . The differences between the snow-rich year 1985/86
and the snow-poor year 1989/90 are all highly significant (p< 0.01), except
for the differences between the correlations of the SRSI time series with the
SPI6 time series in all subregions (p> 0.05). As with drought and flood, we
have again singled out the SPI6 for detailed investigation, since this
appears to be the highest correlated SPI averaging period. Unlike drought and
flood, however, the SPI6 is the only SPI averaging period that shows
consistently high correlations (cf. Figs.
and ) under snow-rich conditions. In contrast to SPI1 and
SPI3, SPI6 is highly correlated with SGI in the snow-rich year, suggesting
that an aggregation period of 6 months is sufficient to account for the
effect of the snow accumulation, which prohibits most groundwater recharge,
just as a lack of precipitation under drought conditions does. However, while
drought conditions still allow for a connection of precipitation and
groundwater, a closed snow cover essentially breaks this connection, with
subsequent precipitation just adding to the existing snow cover.
It is noteworthy that the correlations are also much weaker though for the longer
aggregation periods (SPI9 and SPI12) just as generally observed in the entire time series of all three subregions (see Sect. ).
The observation that a snow-rich year affects the whole aquifer, whereas a
snow-poor year affects parts of the aquifer, also fits the idea of aquifer
response times as discussed in Sect. . Since
the aquifer response time T∗ ranges from over 1 month to over 1 year, a
lack of snow will enable the aquifer to react to short-term and localized
events, such as precipitation, melt or flood events, whereas the delayed
groundwater recharge and runoff under snow-rich conditions is a long-term
event that will be able to affect the whole aquifer.
Average values (dotted lines) for the SGI (blue), SPI1 (yellow)
and SRSI (red) and their 5-year-running means (solid lines) for the Murdurchbruchstal subregion.
![]()
Development over time
As shown in Fig. and Sect. , the Murdurchbruchstal and Leibnitzer
Feld subregions show an increase of correlations with time within the aquifer
and between the aquifer and the rivers and the precipitation time series. In
contrast, the Aichfeld shows no clear trend over time. As shown in
Table , this is also in part reflected by
the significance of the changes between the periods. While the
Murdurchbruchstal and the Leibnitzer Feld show highly significant changes
from the second to the last period, the shallow Aichfeld does not. Also the
changes from the first to the second period are in part significant or highly
significant for the Murdurchbruchstal and the Leibnitzer Feld, whereas the
Aichfeld only shows insignificant changes for these periods.
Compared with the increased correlations under drought conditions (Sects.
and ), one simply
could assume that the split-up time series show a development towards dryer
conditions, which is in line with the general assumption of an already
warming and drying climate for Austria . Another
assumption could be that the split-up time series show a development towards
increasing amounts of snow, as the comparison of snow-rich and snow-poor
years has shown that the correlations between the SGI time series of the
wells and those between the SGI and the SPI tends to be stronger in the snow-rich years.
However, looking at the underlying means (see Figs. , ,
and ), a different picture manifests itself. The average
unstandardized snow levels and fresh snow amounts shown in Fig.
show an increase in snowfall and heights in the first
period, with a sharp drop, followed by a strong increase again in the second
period and a drop and an unsteady development in the third period. Thus, the
most recent time period, which exhibits the highest correlations, is clearly
less affected by snow than the preceding time periods. This is contrary to
the observation made when comparing snow-rich and snow-poor years (Sect. ).
Thus, the following discussion focuses on a tendency
towards drier conditions as potential explanation for the observed increase
in correlations.
Average values (dotted lines) for the SGI (blue), SPI1 (yellow)
and SRSI (red) and their 5-year-running means (solid lines) for the Leibnitzer Feld subregion.
![]()
Average values (dotted lines) and their 5-year-running means (solid lines)
for the fresh snow (pale colors) and the snow height for the Aichfeld subregion (red),
the Murdurchbruchstal subregion (purple) and the Leibnitzer Feld subregion (green).
![]()
While the average SPI in all regions remains more or less stable, there are
some noticeable changes in SGI and SRSI. As shown in Fig. ,
the Aichfeld shows a slight increase in groundwater
levels for the first half of the time, followed by a decrease, whereas the
Murdurchbruchstal (see Fig. ) shows an increase in
groundwater and river water levels in all time periods. Contrary to those two
regions, the Leibnitzer Feld, shown in Fig. , shows an
incoherent signal.
When analyzing the occurrence of extreme events (SGI, SPI and SRSI
below/above -2/+2), we observe the following:
For values below -2, the SPI1 has the largest count in 1987–1998 in the
Aichfeld and the Murdurchbruchstal and in 1987–1998 and 1999–2010 in the
Leibnitzer Feld. This is only reflected in the groundwater in the Leibnitzer
Feld, where the largest count of below -2 events is seen in the 1999–2010
SGI.
The SGI does not follow this pattern for the Aichfeld and the
Murdurchbruchstal, where the highest count is observed in the 1975
(1980)–1986 period, medium count is observed in 1999–2010 and lowest count is
observed in 1987–1998. As shown in Sects.
and , the SPI6 is the highest correlated to the SGI. For this
SPI averaging period, the highest count of below -2 events is observed in the
1999–2010 period in all subregions.
Only the Leibnitzer Feld shows the highest number of below -2 events in the
same (1999–2010) period in the SGI, the SPI1 and SPI6, which is another
indicator for the dominant role of precipitation in this subregion.
The most extreme values below -2.5 only occur in SPI, most prominently in the
Murdurchbruchstal, where we are observing an increase from 0 in the 1980–1986
period to 2 events (one each in SPI1 and SPI3) in 1987–1998 to 17
(SPI6: 3; SPI9: 6; SPI12: 8) in 1999–2010. The other subregions show
smaller counts, with most of the below -2.5 events being observed in the
higher SPI averaging periods and the 1999–2010 period.
The SRSI behaves inconclusive. For the Aichfeld it shows the same pattern of
events with values below/above -2/+2 as the SPI6, indicating a delayed
precipitation controlled river system. In the Murdurchbruchstal, it follows
the same pattern as the groundwater, which fits the interpretation of the
rivers being the driver of the groundwater dynamics.
The SRSI pattern of the Murdurchbruchstal (highest counts of negative events
in 1975/1980–1986, lowest in 1987–1998) is also seen in the Leibnitzer
Feld, but here it fits neither the behavior of the SPI nor that of the SGI.
This is in accordance with our other observations that the river in this
subregion is intensively human influenced and that both, the upstream and the
downstream gauging stations are outside of the subregion.
For extreme flood values above +2.5, it is again only the SPI where those
occur, but with a lower count of only 1 in SPI1 and SPI3 each in the Aichfeld
in the 1975–1986 period and 11 (SPI6: 1; SPI9: 4; SPI12: 6) in the
Murdurchbruchstal in the 1999–2010 period.
SPI1 and SPI6 values above +2 show the same patterns as SPI1 and SPI6 values
below -2, with the largest counts mostly occurring in the 1999–2010 period.
In contrast, SGI above +2 shows the 1975–1986 period as the wettest in the
Aichfeld and the Leibnitzer Feld. Only in the Murdurchbruchstal, the highest
count of +2 SGI events occurs in the time period from 1999 to 2010.
The SRSI also shows inconsistent patterns for positive events, where only the
Murdurchbruchstal has the same behavior in the +2 SRSI as it does in the +2
SGI, confirming again the influence of the river on the groundwater.
This different patterns follow our previous interpretation of river dominated
upstream subregions and a precipitation dominated Leibnitzer Feld. It thus
appears that the influence of precipitation is sufficient to cause a similar
behavior in groundwater levels within the shallow aquifer of the Leibnitzer
Feld, while it is overruled by direct human impacts in the upstream part of
the catchment. When looking at detailed time series (e.g., well MJc or river
gauge MUr in Fig. ), it becomes obvious that many events
above the ±2 threshold are not flood or drought events, but result from an
overlying trend or are the result of direct human activities. The only
exception from this is the SPI since there is no direct human influence on
precipitation. This poses the question of the feasibility of the indices,
which is going to be discussed in the following section.
Feasibility of the indices and synthesis
As already discussed in Sect. , SPI and – to a
smaller amount – SGI have seen considerable use. However, the shallow aspect
of most of our region presents a challenge to the SGI – or similar indices
such as the SRSI; it has been suggested that the assumption of stationarity
underlying many hydrogeological and hydrological assessments and the
engineering decisions based upon them is inadequate in view of the ongoing
hydroclimatic change . In fact, some of the time series singled out in this
investigation show a behavior that is non-stationary within the observed
period of time. Besides the looming threat of climate change, as for example
mentioned by , various events that cause a deviation from a
stationary trajectory (see also Sects.
and ) can be observed. As shown
in Fig. , the wells MDp, MKr and MJc and the river
gauges MUr and LLr exhibit a split pattern, where at a certain point in time,
the standardized values
change from a wet-dominated to a dry-dominated regime or from a dry-dominated to a wet-dominated regime.
For some of the time series in question the reason can be easily found. In
the case of MDp and MKr, it is the construction of the power plant
“Friesach” in 1998 with a pondage of approx. 7 m upstream and a decrease in
tailwater of approx. 1 m . Well MDp is situated
approx. 200 m upstream of the weir, and thus shows “dry” conditions before
the construction and “wet” ones afterward. MKr is just located 1.1 km
downstream of MDp and 1 km downstream of the weir, and thus shows “wet”
conditions before the construction and “dry” ones afterward.
Other time series also seem to be linked to a certain event, such as the case
with MJc and MUr, where a change from a wet to a dry regime happens around
1990. However, in this case, both points are situated 9 km apart from each
other, and none of the power plants that could affect them have been built at
the time in question.
It is interesting to note that those time series discussed above are very
similar to the synthetic time series discussed in ,
most notably Fig. 3 in the cited paper. discussed
a synthetic time series that is running for 1000 terms, which has the
following properties; when looking at the first 50 terms, it appears very
irregular but can be assumed to have a constant mean over time. We argue that
this phenomenon is also visible in our time series MDp and MKr, where the
period from 1975 to 1998 shows a similar behavior to Koutsoyiannis synthetic
series. Zooming out, showed the first 100 terms of
the synthetic time series, which now show two distinctive periods with
different averages and an apparent “shift” or “change” between those
periods. This phenomenon is also visible in our time series MDp and MKr,
where this apparent “shift” or “change” occurs around 1998. Following
this, zoomed out further to 1000 terms of the
synthetic time series, to show that it still is stationary in the long term.
This latter step cannot be seen in our data, but we would expect a similar
picture, when looking at the time series MDp and MKr 1000 years from now.
This effect of apparent stationarity when zooming in can also be seen in
Sects. and , where it becomes apparent that the
split-up time series are generally showing higher correlations than the full
time series, since only comparably smaller parts of the time periods are
affected by a large change.
A quantification and counting of extreme events for the full time, as
attempted in Sect. , is thus
problematic. Calling, e.g., an index value of -1 to -1.49 “moderate drought”
can be misleading when assessing a non-stationary time
series, such as well MDp (Fig. ). Here the first approx.
18 years would be interpreted as a period of multiple and persistent moderate
to severe droughts, followed by a period of multiple and persistent moderate
to severe floods. What the time series really shows is an aquifer in
equilibrium with its surroundings before and after the construction of a
run-of-the-river power plant and the associated change in groundwater level.
To enable a quantification of the negative (and positive) events, the time
series in question could be split-up at the time of the change, standardized
independently and put back together. However, this requires knowledge of the
nature and the timing of the underlying events, which in our case was not
always available.
For systems understanding and correlation, however, these jumps in time series
are not an issue. As shown with wells MDp and MKr, the construction of a
run-of-the-river power plant not only changes the water levels of the
river in question, but also affects the groundwater up- and downstream of
it. With our matrix view (Fig. ), it can be shown that
this change not only affects the two wells singled out, but also at least
one other well downstream in the case of MKr, where the first “blue
outlier” above it is situated directly across the Mur from MKr. The second
one, however, is upstream of MKr and its power plant but in a similar
downstream distance from another power plant . With
well MDp, there are at least two other wells upstream that show very high
correlations with MDp.
This shows that large events or human-induced changes in the river, such as
the construction of a run-of-the-river power plant, can affect not only its
direct vicinity, but also large portions of the surroundings. This is a
further important factor besides other human-induced changes, such as change
in land use (surface sealing, afforestation, deforestation, etc.) and pumping
activities as for example mentioned by . In small, and
heavily human-impacted systems, such as in the Mur valley described herein,
those human-induced changes can be among the most important influences,
rendering the concept of “natural conditions” almost impossible in shallows
wells. Short-term disruptions, on the other hand (as demonstrated by well MFd
in the Murdurchbruchstal in 2003), do not affect the long-term correlations.
Conclusions
Three subregions of the Austrian Mur catchment were analyzed. Long-term time
series (1975/1980–2010) of 75 groundwater monitoring wells, 9 river gauging
stations and 3 regional average precipitation time series have been
standardized and correlated in order to gain insight into the controlling
factors for groundwater in alluvial aquifers, the effects of extreme events,
the impacts of human activities and the development over time. It was shown
that the correlation matrix approach enables a quick visualization and
comparison of different locations and time spans and that standardized
indices, such as the SPI, the SGI and the SRSI (SGI applied to river levels),
allow for a thorough comparison of groundwater wells, rivers and
precipitation.
With the help of these tools, it was shown that subregions in a catchment can
show very different behavior, stemming from their different climatic and
geologic conditions as well as human impacts. In general, in small subregions
and shallow alluvial aquifers as shown here, the river is always an important
driver in the system. As a consequence, (human) impacts on the river (e.g.,
construction of a run-of-the-river power plant) propagate into the aquifer
system. When assessing shallow groundwater basins in a densely populated
area, human impacts must be taken into account. Without this context, many
phenomena observed in the system can easily be misinterpreted.
The correlation of standardized groundwater levels with standardized
precipitation is more significant in the foreland than in the upstream,
Alpine part of the catchment. This corresponds to a tendency towards more
shallow water tables in the foreland, and the existence of a second, deeper
aquifer in the upstream basin. The shallow wells show time series that are
highest correlated with the SPI6, whereas the deep wells show the highest
correlation with the SPI12. This highest precipitation–groundwater
correlation of the deep wells is still considerably lower than the highest
precipitation–groundwater correlation of the shallow wells. Besides being
only lowly correlated with precipitation, the deep wells also appear to be
unaffected by river stage fluctuations.
Extreme events, exemplified by the 2003 drought, the 2009 floods, the 1985/86
snow-rich winter and the 1989/90 snow-poor winter, significantly impact the
correlations between the standardized time series, but differ in their
effects. Drought and snow-rich conditions show a tendency towards higher
correlations and thus uniform behavior of precipitation, surface water and
groundwater, whereas flood and snow-poor conditions result in lower
correlations and thus irregular behavior. A possible explanation for this
observation is the fact that the unconfined aquifers in our subregions have
response times of at least 1 month; therefore, short-term events,
such as floods, will not affect the whole aquifer, whereas events of long duration,
such as a drought will propagate through the whole subregion, which will be reflected in
the mentioned high correlations.
In contrast, the aquifer represented by the deep wells in the Aichfeld subregion are
likely confined or semi-confined, which results in much lower response times explaining
the consistently high correlations among those wells even under flood conditions.
When assessing the development over time, the most recent time period from
1999 to 2010 shows significant changes and a trend towards higher
correlations. This corresponds to an increase of the number of events in
precipitation with index values of SPI6 below -2 in all subregions and in the
groundwater of the foreland subregion, suggesting the increased number of
drought events as a possible cause of the observed trend towards higher
correlations. The investigated Alpine aquifers, however, exhibit a
contrasting behavior with the highest number of negative events in the time
before 1986. This suggests that the groundwater levels within these
subregions are more strongly influenced by direct human impacts, e.g., on the
river, than by changes in precipitation. Thus, direct human impacts must not
be ignored when assessing climate change impacts on alluvial aquifers
situated in populated valleys. Accounting for human impacts within such
assessments remains a challenging task that requires further investigation
into the nature of the various impacts and the mechanisms of their
propagation through the hydrological system. Further work could address
different types of aquifers, including larger aquifer bodies or aquifers in
different climate zones.