Climate change is one of if not the most pressing challenge modern society
faces. Increasing temperatures are observed all over the planet and the
impact of climate change on the hydrogeological cycle has long been shown.
However, so far we have insufficient knowledge on the influence of
atmospheric warming on shallow groundwater temperatures. While some studies
analyse the implication climate change has for selected wells, large-scale
studies are so far lacking. Here we focus on the combined impact of climate
change in the atmosphere and local hydrogeological conditions on groundwater
temperatures in 227 wells in Austria, which have in part been observed since
1964. A linear analysis finds a temperature change of
The thermal regime in the ground is coupled with the conditions in the atmosphere, and air temperature variations leave their traces in the ground. While already at a depth of a few metres, the amplitudes of periodic diurnal and seasonal temperature trends are strongly attenuated (Taylor and Stefan, 2009), long-term non-periodic changes of air temperature permanently influence the subsurface down to greater depths of several tens to hundreds of metres (Beltrami et al., 2005). Worldwide, borehole temperature profiles therefore show an increase in surface air temperature (SAT) due to recent climate (Huang et al., 2000; Harris and Chapman, 1997). In borehole climatology, the focus is set on “dry” boreholes in undisturbed natural areas, that is, boreholes with negligible influence of groundwater flow and no direct human impacts. Borehole temperatures logged in such boreholes can be used to invert vertical conductive heat transport models for deriving the corresponding trend of ground surface temperature (GST). By assuming that GST and SAT are directly coupled or similar, past climate can be reconstructed. Many boreholes, however, are located in urbanized areas and regions with past changes in land cover, where often accelerated ground heat flux and higher GST are observed (Bense and Beltrami, 2007; Menberg et al., 2013; Bayer et al., 2016; Cermak et al., 2017). Moreover, in humid climate regions boreholes are mostly not dry, but drilled for groundwater use or monitoring. When dynamic groundwater flow conditions exist, then advective heat transport can substantially affect the thermal regime in the subsurface (Ferguson et al., 2006; Kollet et al., 2009; Taylor and Stefan, 2009; Stauffer et al., 2017; Westaway and Younger, 2016; Uchida et al., 2003). Additionally, recharge processes, including snowmelt and rain-derived recharge, might impact the thermal regime of the shallow subsurface. Previous studies, however, indicate that in many cases their influence can be neglected. Ferguson and Woodbury (2005) and Bense and Kurylyk (2017) demonstrated that it is possible to estimate groundwater recharge by using temperature–depth profiles based on the common assumption that the mean annual groundwater recharge temperature is equal to the mean annual surface air temperature. Menberg et al. (2014) showed in their study that the contribution of snowmelt-induced recharge with low temperature is minor in comparison to the overall recharge. Finally, Molina-Giraldo et al. (2011) investigated the impact of seasonal temperature signals on an aquifer upon bank infiltration, also including varying groundwater recharge temperatures. They showed that the convective heat transfer by groundwater recharge compared to conduction through the unsaturated zone and convection within the aquifer is of minor impact. Still, the interplay of long-term climate variations, land use change and groundwater produces a complex transient system, which is difficult if not impossible to accurately understand based on a few borehole measurements (Irvine et al., 2016; Kupfersberger et al., 2017; Kurylyk et al., 2017, 2014, 2013; Taniguchi and Uemura, 2005; Taniguchi et al., 1999; Zhu et al., 2015).
The consequence of climate change for aquifers was illuminated with respect
to groundwater recharge and availability of freshwater resources (Moeck et
al., 2016; Scibek and Allen, 2006; Holman, 2006; Gunawardhana and Kazama,
2011; Loáiciga, 2003), groundwater quality impacts (Kolb et al., 2017)
and effects on groundwater(-dependent) ecosystems (Burns et al., 2017;
Jyväsjärvi et al., 2015; Kløve et al., 2014; Andrushchyshyn et
al., 2009; Hunt et al., 2013). Taylor et al. (2012) summarized various
connections and feedbacks between climate change and groundwater. A key
parameter is the temperature, which is expected to increase in shallow
groundwater globally following with some delay roughly the trends in the atmosphere. However,
long-term measurements of temperature evolution in groundwater are rare
(Watts et al., 2015; Figura et al., 2015). Instead, often well measurements
taken at a few different time points are compared to indicate elevated
temperatures, such as by Gunawardhana and Kazama (2011) for the Sendai Plain
in Japan, by Šafanda et al. (2007) for boreholes in the Czech Republic,
Slovenia, and Portugal, and Yamano
et al. (2009) and Menberg et al. (2013) for urban areas in eastern Asia and
central Europe. Others, such as Kupfersberger (2009) and Menberg et
al. (2014), examine repeated temperature records of single or a few selected
wells. The work by Lee et al. (2014) is one of the very few studies on
long-term groundwater temperature (GWT) time series recorded for a larger
area. They applied linear regression to hourly temperature data recorded from
2000 to 2010 at 78 South Korean national groundwater monitoring sites. They
found a mean increase of 0.1006 K year
In the presented study, GWTs of 227 wells in Austria, measured in part since 1966, are analysed and regional patterns and temperature anomalies are identified. In contrast to Blaschke et al. (2011), the focus here is not only set on linear trends, but also on detection of climate regime shifts in the measured GWT, following the suggestions by Figura et al. (2011) and Menberg et al. (2014). As a relevant mode of global climate variability, long-lived decadal patterns such as the Atlantic or Pacific decadal oscillation have been identified, e.g. Minobe (1997) and Rodionov (2004). These control atmospheric temperatures as well and are often described as sudden, step-wise temperature changes separating stable periods, called climate regimes. Even if these regime shifts arrive attenuated and delayed in shallow groundwater, they can be detected and thus can offer another hint at the influence of climatic variations. Aside from the statistical analysis of GWT time series, the influence of land cover as well as their correlation with surface air temperature are investigated to scrutinize potential local influences on the measured data.
The Austrian Alps as the main part of the European Eastern Alps are characterized by a complex geology with various lithologies and were built up during multiple tectonic phases striking now in a west–east direction. The complexity of the tectonic and geologic settings of the European Alps and in particular of the European Eastern Alps is described and discussed by numerous authors (e.g. Schmid et al., 2004; Linzer et al., 2002). Active tectonic evolution resulting in high topography and uplift rates coincided largely with high stream power (Robl et al., 2017, 2008) and thus had an impact on the drainage system of the Alps. During the Pleistocene the Alps were affected by glaciations with a strong impact on the morphology, in particular on the inner Alpine valleys and the foreland. Due to sedimentation during the Holocene these areas now contain Quaternary porous aquifers. The herein analysed wells are located in shallow aquifers representing these Quaternary sediments in the inner Alpine valleys and foreland basin. Based on a compiled geology a hydrogeological overview as a hydrogeological map of Austria is provided by Schubert et al. (2003).
Climate and climate trends during the last 2 centuries (1800–2000) of the Great Alpine Region (European Alps and their surrounding foreland, GAR) were intensively investigated during the last few decades, yielding the HISTALP data set (Auer et al., 2007). This data set left its mark on the regional classification of climate zones by Köppen–Geiger, where Austria is mainly divided into three climate zones, warm temperate, boreal, and Alpine.
In Austria, GWTs up to December 2013 are provided by the Austrian Federal
Ministry of Sustainability and Tourism Directorate-General IV. – Water
Management (BMNT, 2016), the former Federal Ministry of Agriculture,
Forestry, Environment and Water Management (BMLFUW) in 1138 wells. Here, we
focus on all wells with a measurement depth of less than 30 m, a record of
at least 20 years and no major breaks (
The average measurement depth in the wells is
Following the CORINE Land Cover (CLC) data from 2012 (Fig. S2a), 45 % of
all wells are under artificial surfaces, 46 % under agricultural areas,
and 9 % under forest following the 100 m
Surface air temperatures (SATs) within Austria are monitored by the Central Institution for Meteorology and Geodynamics (ZAMG), Austria. In this study data from 12 individual weather stations are being analysed, each one located within 5 km of at least one analysed well and in the same climate zone (Cfb). Their location is displayed in Fig. 1a. Again, annual mean data were available for a time period of 1966–2013 (Fig. 1b). As expected and as previously shown in Benz et al. (2017b) for SAT and Benz et al. (2017a) for land surface temperatures, above-ground temperatures are generally lower than GWTs. All 12 analysed weather stations are located in areas classified as artificial surface and experienced no land cover changes.
Within this study, the Spearman correlation coefficient was determined, as it is especially robust to outliers caused for example by heat waves, which impact air temperatures but have only minor effects on groundwater temperatures. When determining the correlation between two time series, missing years were ignored. Next to the correlation between GWT and SAT, correlations between all individual wells and weather stations were determined in order to create a plot similar to a (semi)variogram that shows the correlation between two measurement stations, depending on their distance to each other.
Equivalent to the work by Lee et al. (2014), a linear temperature change was determined for all 227 wells. For this, a linear regression model of the annual mean temperature data was determined in Matlab (Version 2016b). Because all wells in our data set were continuously monitored between 1994 and 2013, only this timeframe was analysed.
Climate data are often thought not to change linearly, but in the form of a step function, dividing a time series into individual climate regimes of a constant mean (Andrushchyshyn et al., 2009; Minobe, 1997). These regimes are changed when so-called climate regime shifts (CRS) occur and long-term mean values change. While several methods to model these shifts have been in use (Easterling and Peterson, 1995), in recent years the method by Rodionov (2004) has become standard. It identifies the significance of each possible shift by calculating the so-called Regime Shift Index (RSI): the cumulative sum of the normalized differences between the observed values and the long-term mean of the assumed regime. Only shifts with a positive RSI are considered significant, and a higher value of RSI denotes more pronounced CRS. The entire algorithm is described in detail by Rodionov (2004). This sequential analysis is data driven and requires no prior knowledge of the timing of possible shifts. It was updated to further include prewithening in order to reduce background noise (Rodionov, 2006) and is available online as a Microsoft Excel add-in (NOAA, 2017). In this study we applied the method to the complete time series of all 227 wells and 12 weather stations. Because the algorithm cannot handle gaps within the analysed series, gaps in our data were filled using a linear fit. All parameters were set to the same values as in the work by Menberg et al. (2014), who applied the method to four GWT time series in Germany. A target significance level of 0.15 was used by Menberg et al. (2014), and in our analysis, the cut-off length was set to 10 years and the Huber weight parameter was set to 1.
Influence of distance on the correlation between the annual means of
two measurement points.
Change from 1994 in surface air temperature (SAT) and groundwater
temperatures (GWTs) of all wells within 5 km of the analysed weather
station. See Fig. S3 for an overview of the locations. Minimum and maximum
correlations and
Figure 2a displays the correlation between different wells or rather
different weather stations in relation to their distance to each other. Shown
is the distance between two wells/weather stations on the
As expected the moving average correlation decreases with distance. This decrease is more extreme in GWTs than in SATs and GWTs correlate less than SATs overall. This agrees with the observations in Benz et al. (2017b), who showed that annual mean GWTs show greater variations than SAT over the same distances.
Additionally, the correlation between two wells seems to be anisotropic: correlation coefficients between two wells decrease faster with north–south distance than with west–east distance (Fig. 2b), which can be explained by the dominant striking direction of the geology and the resulting topography in Austria, where valleys generally run from west to east. Hence, larger rivers typically follow this direction and wells at the same latitude experience similar temperature signals.
Correlation coefficient and corresponding
In a next step, correlations between GWT and SAT are determined. On a
country-wide scale the Spearman correlation coefficient between spatial
median GWT and SAT (Fig. 1b) is 0.83. In comparison, the correlations between
individual weather stations and wells are shown in Fig. 3; locations are
displayed in detail in Fig. S3. Here correlations vary greatly and Spearman
correlation coefficients are
The best correlations between individual pairs of a well and a weather station can be observed in the southern part of the city of Graz, where all wells and the weather station are located close to or within Graz airport, respectively. The well with the highest correlation of 0.80 with SAT is located less than 1 km from the weather station close to the airport parking lot next to suburban housing. It has been continuously monitored since 1970 and is the longest time series in the area. The well with the lowest correlation (0.45) with the weather station here is located slightly to the east near a dog park and suburban housing. Here observations started in 1994: it is the shortest time series in this area. At all other wells, measurements began in 1986 and show correlations between 0.6 and 0.7 with SAT, indicating that the duration of the measurements plays a significant role for local comparisons. In contrast, duration of the time series appears to be of minor importance on a country-wide scale. For example, the long time series in Wiener Neustadt (Fig. 3), which started measurements in 1970 and is located near a mineral extraction site, has a correlation of 0.48 and is therefore comparable to the short time series in Graz, starting in 1994 and located in a suburban area.
Additionally, measurement depths of GWT can have an impact on the correlation between SAT and GWT. While it is generally assumed that a measurement depth closer to the surface results in a better correlation with SAT as there is less of a shift between both data sets, this is only the case for some of the locations analysed here, such as Villach (Fig. S4a). In contrast, correlation increases with GWT measurement depth for other locations, such as the one in Graz. This might be related to local underground heat sources such as sewage systems impacting GWT near the surface more than temperatures at greater depth. However, as the depth of the wells analysed here varies only slightly, no definite conclusions can be drawn without further inspection of specific cases.
Table 1 displays the correlations between spatial median GWT and spatial median SAT for each of the SAT locations in Figs. 3 and S3. For all locations with at least two wells besides Zeltweg and Graz correlation does improve when spatial median GWT is analysed instead of the individual locations. In all likelihood the spatial median GWT provides a more general temperature trend that is not influenced by local influences on temperatures such as construction work, plant development and shading, and is therefore more closely related to surface air temperatures.
In addition, the data indicate that city size, or rather population of a
city, do not necessarily influence the correlation between GWT and SAT
(Table 1). For example, both locations, Graz (population of more than
250 000) and Eisenstadt (population of 13 000), have similar correlation
coefficients despite their different populations. Meanwhile, Bregenz and
Feldkirch have a similar population (
During the time between 1994 and 2013, GWTs changed on average by
To evaluate the goodness of this linear approach when representing climate
change, RMSE of the fit was determined for each well for 1994 to 2013. We
found an average RMSE of 0.4
Increase in temperature for all individual measurement points for
the 20-year timeframe January 1994 to December 2013. The mean change in
groundwater temperature is
Annual mean time series and linear fit (in grey) of the wells with
the lowest (
When looking at the individual wells, no obvious spatial pattern for temperature changes is visible (Fig. 4). However, most wells with temperature changes lower than the 5th percentile are located close to the Drava River in Ferlach, Villach, and Kleblach-Lind in the very south of Austria (Figs. 5 and S5). Although they are up to 80 km away from each other, all of these wells show a sudden drop in temperatures in the year 2007 (wells Ia, Ib, IIa, IIb, Va, and Vb marked in blue in Fig. 5). This temperature reduction can be seen in most of the 27 wells that are less than 1 km from the Drava (Fig. S6); for 24 of these wells, temperatures in 2006 are more than 0.6 K warmer than temperatures in 2008. However, temperatures (as well as additional parameters such as water level) within the river do not indicate any connection between this sudden temperature reduction and the Drava River (Fig. S6). Either way, further research is necessary to identify the cause of this temperature anomaly.
Additionally, three other wells in the lowest 5 % of temperature change
are all located less than 10 km from each other near the village of Kappel
am Krappfeld (wells IVa, b and c marked in orange in Fig. 5). They and also
additional surrounding wells show a steep decline in temperatures in 2006
before temperatures start to increase steadily again. These wells seem to be
affected by the new drinking water supply (four wells with a total pumping
rate of about 100 L s
All detected CRS of the spatial median temperatures time series are shown in Fig. 6a. Overall GWTs increased by 1.2 K between the first and last CRS and SAT increased by 1.5 K.
Global CRS in air and also groundwater were detected for the late 70s, the late 80s and the late 90s by Menberg et al. (2014). Using the same algorithm spatial median annual mean GWT and SAT in Austria show shifts in the late 80s and 90s (Fig. 6a). GWTs show additional shifts in 1981 and 2007. While the shift in the late 80s is observed during the same year (1988) in GWT and SAT, the shift in the late 90s appears earlier and is more significant in GWTs. However, because SATs are the drivers of GWTs and not vice versa, the fact that the GWT change precedes the SAT change suggests that this method does not have the necessary resolution to determine short time lags between SATs and GWTs. Accordingly the detected time shifts in wells within 5 km of a weather station generally do not indicate the same CRS as the weather station: of 56 CRS observed in at least one well only, 12 are also observed in a nearby weather station no more than 1 year before (Fig. S8). However, it is also important to note that some of the analysed time series only span a 20-year time period and are thus on the shorter end for a statistically relevant analysis of climate regime shifts (Rodionov, 2006).
Like with the linear approach, the goodness of the CRS and corresponding
statistical step model was evaluated by determining the RMSE for the time
period 1994–2013. We determined a mean RMSE value of 0.3
Temperatures in 227 shallow wells and 12 weather stations in
Austria, monitored in part since 1966, were analysed in this study. Linear
temperature change was determined and revealed a general increase in
temperature between the years 1994 and 2013 of approximately
The GWT data used in the study were provided by the Austrian
Federal Ministry of Sustainability and Tourism Directorate-General IV. –
Water Management (BMNT, 2016) and are available to the public under
The SAT data used in this study were acquired through the Central Institution for Meteorology and Geodynamics (ZAMG), Austria, in 2017.
The supplement related to this article is available online at:
The authors declare that they have no conflict of interest.
We would like to thank Erich Fischer (BMNT, formerly of BMLFUW) for information and data regarding groundwater temperatures, and Alexander Orlik (ZAMG) for information and data regarding surface air temperatures of Austria.
Furthermore, we would like to acknowledge the financial support for the first author through portfolio project “Geoenergy” of the Helmholtz Association of German Research Centres (HGF) and the support by the Deutsche Forschungsgemeinschaft and Open Access Publishing Fund of the Karlsruhe Institute of Technology. A big thank you also to one anonymous reviewer and to Randy Hunt for their helpful and very constructive comments. The article processing charges for this open-access publication were covered by a Research Centre of the Helmholtz Association. Edited by: Brian Berkowitz Reviewed by: Randy Hunt and one anonymous referee