Evapotranspiration amplifies European summer drought


  • The copyright line for this article was changed on 13 APR 2015 after original online publication.


[1] Drought is typically associated with a lack of precipitation, whereas the contribution of evapotranspiration and runoff to drought evolution is not well understood. Here we use unique long-term observations made in four headwater catchments in central and western Europe to reconstruct storage anomalies and study the drivers of storage anomaly evolution during drought. We provide observational evidence for the “drought-paradox” in that region: a consistent and significant increase in evapotranspiration during drought episodes, which acts to amplify the storage anomalies. In contrast, decreases in runoff act to limit storage anomalies. Our findings stress the need for the correct representation of evapotranspiration and runoff processes in drought indices.

1 Introduction

[2] Recent summer droughts in central and western Europe have had widespread impacts on crop yield, ecosystems, and infrastructure [e.g., Ciais et al., 2005; Corti et al., 2011]. While drought characteristics over central and western Europe show no clear trend over the past century [Lloyd-Hughes and Saunders, 2002; Sheffield and Wood, 2008], climate model projections present a consistent increase in frequency and magnitude of droughts in that region [Seneviratne et al., 2012a]. Understanding the driving factors controlling the variability in water availability is key to correct simulation and early warning.

[3] While many studies have focused on the impacts of recent droughts [Ciais et al., 2005; Corti et al., 2011] or on the quantification of water availability [Li et al., 2012], few studies have focused on the terrestrial hydrological controls on drought evolution [e.g., Seneviratne et al., 2012b]. Specifically, the response of evaporation (hereafter evapotranspiration or ET) is uncertain, in particular due to a lack of long-term reference observations [Dolman and De Jeu, 2010; Mueller et al., 2011]. Lack of rainfall leads to drier soils, and it is often assumed that ET rates will decrease with decreasing soil moisture. There is, however, considerable evidence that ET is only limited at low values of available soil moisture [Teuling et al., 2010a; Seneviratne et al., 2012b]. As a result, over most of the range of available soil moisture, ET responds to variability in atmospheric conditions rather than variability in soil moisture. In humid climate regions, atmospheric conditions during drought are often more favorable for ET [De Boeck and Verbeeck, 2011] and can thus lead to increased rather than decreased ET [e.g., Seneviratne et al., 2012b]. Increases in ET were also found for warm conditions that often coincide with drought over most of Europe [Stegehuis et al., 2013], in particular over grasslands [Teuling et al., 2010a].

[4] Studies on the contribution of the different components of the terrestrial water cycle to drought evolution have been limited, mainly due to the lack of high-quality and long-term records of actual ET. Current gridded data sets of ET show considerable disagreement [Mueller et al., 2011], even on the sign of anomalies [Jiménez et al., 2011]. Therefore, we use unique long-term, high-quality, and independent in situ observations from four small headwater catchments in central and western Europe that cover a wide range of hydroclimatological conditions. To investigate how anomalies in the various water fluxes, in particular ET, impact the evolution of storage anomalies at the monthly time scale, we focus on examples of the 1976 and 2003 drought extremes, as well as on the average behavior in the complete data series.

2 Approach

[5] In general, drought is a well-below-average availability of water in the environment. Whereas meteorological drought expresses long-term deficiencies in precipitation only, agricultural and hydrological drought deal with deficiencies in soil moisture, groundwater, and surface water reservoirs. Hydrological drought is often diagnosed by streamflow drought. Here we define drought as anomalously (for the time of year) low total subsurface storage conditions at the catchment scale. This definition combines agricultural and hydrological droughts and only includes meteorological droughts as long as they result in a negative storage anomaly. The catchment water balance is a central element in our approach:

display math(1)

where S is catchment storage (L), P precipitation (L/T), Q runoff (L/T), and ET is evapotranspiration (L/T). Since the catchment water balance, when written in terms of anomalies, links changes in storage anomalies (S′) to flux anomalies (P′, ET′, Q′) [e.g., Seneviratne et al., 2012b], the temporal evolution of a storage anomaly can be reconstructed from:

display math(2)

where S0 is an arbitrary integration constant. Note that due to the integration of measurement errors, the accuracy of S′ will degrade with the length of the integration period. Alternatively, when multiyear records of profile soil moisture are available, absolute local storage anomalies can be obtained from vertical integration of volumetric soil moisture anomalies:

display math(3)

where θ is the volumetric soil moisture content. Note that this approach yields absolute values of S′ when soil moisture observations are made down to the groundwater level and compressibility effects are small.

3 Data

[6] Only few catchments exist with multiyear observations of at least three of the four most important components of the catchment water balance (equation ((1))) and including observations made during years of extreme drought. In this study, we use data from four such experimental catchments in central and western Europe (Figure 1 and Table 1). These nonurbanized catchments cover a wide range of topographic, geographic, and hydroclimatologic conditions found in Europe.

Figure 1.

Location of the study catchments (inverted open triangles) Hupsel Brook, (open triangles) Rietholzbach, (open squares) Salm, and (open circles) Wernersbach. Colors/contours indicate elevation.

Table 1. Data Used in This Study
SiteYearsClimatologySETSource P, QReference
Hupsel Brook1976–19841977–1984Neutron probe (grass-, cropland)N/AWageningen UniversityBrauer et al., 2011
Rietholzbach1976–20112004–2010N/ALysimeter (grassland)Seneviratne et al., 2012bSeneviratne et al., 2012b
Salm/Vielsalm1996–20102004–2010N/AEddy covariance (mixed forest)Wallone public service (Voies hydrauliques)Aubinet et al., 2001
Wernersbach/Tharandt2000–20102004–2010N/AEddy covariance (spruce forest)Technische Universität DresdenGrünwald and Bernhofer, 2007

[7] Hydrological observations in the Dutch Hupsel Brook catchment started in 1976 and cover the extremely dry summer of 1976. From 1976 to 1984, profile soil moisture observations were made using a neutron probe at six locations throughout the catchment at a 14-day resolution, providing a unique and high-quality observational data set. At the Swiss Rietholzbach catchment, observations started in 1976, including ET measured by a weighing lysimeter with ambient moisture supply and covering the extreme dry summers of 1976 and 2003. Continuous eddy covariance observations of latent heat fluxes are available at the Belgian and German Fluxnet sites of Vielsalm and Tharandt since the late 1990s. The flux tower at Tharandt is assumed to be representative for ET at the nearby Wernersbach catchment, where precipitation and discharge have been measured since 1967 [Spank and Bernhofer, 2012]. Land use in the catchments is mixed, with the Hupsel being least forested (3%) and the Wernersbach most (90%). Average runoff coefficients range widely from 0.23 (Wernersbach) to 0.64 (Rietholzbach). Since exact error structures for the measurements are unknown, we assume a constant and independent error of ±5% of the monthly flux for P, Q, and ET.

[8] Here we use available monthly data up to the end of 2011 (Table 1). Climatologies are derived for the common period 2004–2010 (i.e., without the dry extreme 2003) unless stated otherwise. Differences between climatologies for 2004–2010 and the full data record for the Rietholzbach were found to be small (1%–6%), suggesting a limited effect of the relatively short data record. Observed ET was rescaled in order to close the long-term water balance (note that this operation does not affect the sign of the anomalies). Catchment water balance data are supplemented by gridded satellite estimates of storage anomalies by the Gravity Recovery and Climate Experiment (GRACE) taken from GRACE Tellus (grace.jpl.nasa.gov).

4 Results

[9] The 1976 drought struck large parts of western Europe. The Hupsel Brook soil moisture data set reveals the temporal evolution of the storage anomaly (Figure 2a) and how the drought propagated through the subsurface (Figure 2b). In absence of ET measurements, we reconstructed storage anomalies S′ − S0 from P′ − Q′ assuming that ET′ = 0. Comparison of these reconstructed storage anomalies with absolute storage anomalies derived from the soil moisture profile observations shows that while both agree on the magnitude of the storage anomaly (order −120 mm), the latter shows a larger amplitude. This difference can be explained by a possible positive ET′, which would lead to larger negative S′. The timing and depth of the storage anomaly support the role of ET: strong increases in storage anomalies in April and July (Figure 2a) deviate from the P′ − Q′ estimate and occur at relatively shallow depths (< 50 cm; Figure 2b) at which water is taken up by plants. The downward propagation of the negative storage anomaly is consistent with the preference of plants to first utilize soil moisture in the upper layers, as well as with the influence of the shallow groundwater table, which prevents early development of deeper soil moisture anomalies.

Figure 2.

Evolution and propagation of the 1976 drought in the Hupsel Brook catchment. (a) Storage anomaly evolution from (orange) ∫θ′dz and (blue) ∫(P′ − Q′)dt. Error bars represent propagated 5% random error in monthly P and Q, and 1 vol% in θ. (b) Profile soil moisture monthly anomaly evolution and depth of groundwater table. Dashed line: 1976, solid line: climatology with error bars indicating interannual standard deviation. Data are averages over four profiles, and climatology is calculated over the period 1977–1984.

[10] In many parts of central and western Europe, the intensity of the 2003 summer drought exceeded that of the 1976 drought. Observations in the Swiss Rietholzbach catchment cover both drought years and provide an excellent example of how the interplay between the different water balance components drives drought evolution [e.g., Seneviratne et al., 2012b]. In Figure 3, the storage anomaly evolution for 2003 shows a large negative anomaly at the end of the summer of nearly −200 mm (assuming S0 = 0 on 1 January). Data from GRACE confirm that the storage anomaly evolution can be reconstructed correctly from the catchment water balance. The error bars of both time series largely overlap. Similar to the 1976 drought example for the Hupsel Brook catchment (Figure 2), the 2003 drought in the Rietholzbach did not evolve gradually: most of the storage anomaly increase took place in March and June [see also Seneviratne et al., 2012b]. In both months, ET' contributed significantly to dS′/dt, with monthly anomalies of up to +38 mm. Without this contribution, i.e., by assuming ET′ = 0, the amplitude of the reconstructed storage anomaly is much smaller (−121 versus −203 mm) and no longer consistent with GRACE estimates.

Figure 3.

Evolution of the 2003 drought in the Rietholzbach catchment and attribution of storage anomaly changes. (orange) GRACE storage anomalies are plotted for reference. Error bars represent (propagated) 5% random error in monthly P, Q, and ET and GRACE errors. (bottom) Observed flux anomalies combine to (middle) dS′/dt, which is integrated (top) from S0 to yield S′. The axis have the same scale for comparison. All climatologies are calculated for the period 2004–2010. Note that the errors for S′ with ET′ = 0 (not shown) are shared with S′.

[11] Besides ET, it is also relevant to look at the role of precipitation and runoff. Obviously, precipitation anomalies are generally the main drivers of drought evolution. However, much of this effect can be counteracted by negative runoff anomalies caused by low storage conditions. In humid catchments, such as the Rietholzbach, runoff closely follows storage levels [Teuling et al., 2010b]. This has some interesting implications: whereas below-average runoff is one of the main problems during drought, it can limit the effects of the precipitation anomalies on the storage anomaly budget.

[12] A more general picture of the role of different flux anomalies on drought development can be obtained by looking at the data for all warm season months in the available time series. In this way, smaller droughts are also included. Figure 4 shows the relation between dS′/dt and all contributing fluxes for the three catchments with flux observations. They confirm the role of P′ as the main driver of dS′/dt. Moreover, they show that typically P′ and −ET′ contribute similarly to the anomaly water budget; i.e. months with below-average rainfall tend to have above-average ET, which acts to amplify dS′/dt with respect to the effect of P′ alone. The control of the different flux anomalies on dS'/dt is surprisingly similar for the three catchments with long-term observations of all components. Precipitation makes up between 81% and 109% of dS′/dt. Runoff has a smaller effect with −6% to −31% of dS′/dt. The contribution of ET shows a striking similarity (between +19% and +27% and larger than 0% at the 95% level) between the three catchments. The ET contribution also does not differ significantly between the catchments, despite their different hydroclimatological conditions and the fact that the measurements were made independently and by different methods (lysimeter and eddy covariance).

Figure 4.

Contribution of monthly flux anomalies to warm season (April–September) storage anomaly changes. Positive slopes indicate positive contribution. Regression results have been obtained for the model Y = αX, i.e., by assuming E(Y|X = 0) = 0. Envelopes (upper panels) and error bars (bottom panel) indicate 95% confidence interval for α. Bottom panel summarizes the relative contribution of all flux anomalies to dS′/dt. All climatologies are calculated for the period 2004–2010. Storage anomaly changes are calculated from dS′/dt = P′ − Q′ − ET′.

5 Discussion and Conclusions

[13] Long-term observations of water balance components at four central and western European headwater catchments reveal a statistically significant and consistent positive contribution of ET to storage anomalies during summer drought. This can be explained by the strong increase in net radiation during drought conditions, since a lack of rainfall generally coincides with decreased cloudiness associated with anticyclonic circulation patterns. This acceleration of part of the terrestrial water cycle under drought conditions appears paradoxical since the terrestrial hydrological cycle is typically assumed to be driven by precipitation. The “drought paradox” shows that reduced precipitation can coincide with increased evaporation, effectively amplifying storage anomalies.

[14] The amplifying role of ET on storage anomaly evolution will reduce with drier climate conditions. During the 2003 drought in the Rietholzbach, the wettest catchment under study, ET was generally above average, whereas the evaporative fraction dropped to below-average values over the course of the summer [Seneviratne et al., 2012b]. In the driest catchment studied here (Wernersbach), strong negative ET anomalies occurred at the end of the 2003 summer, reflecting severe soil moisture stress. In other drier regimes, negative ET anomalies during drought will be common, leading to reduced evaporative cooling and strong coupling to temperature dynamics and extremes [Mueller and Seneviratne 2012; Miralles et al., 2012].

[15] Long-term water balance studies require observations of at least three components of the water balance; and only few catchments can be found where such dedicated monitoring has been undertaken. At those catchments, a nonclosure is generally observed, suggesting an underestimation of land surface fluxes. The nonclosure is worst for the two catchments with eddy covariance observations of ET (Salm and Wernersbach) and virtually absent for the climatology period used for the Rietholzbach with ET derived from lysimeter observations and is thus consistent with previous studies showing considerable underestimation of turbulent heat fluxes at FLUXNET sites [e.g., Wilson et al., 2002].

[16] In spite of the recent progress in land surface monitoring, current drought estimation in widely adopted operational products still largely relies on poorly parameterized potential ET in combination with bucket models (e.g., the Palmer Drought Severity Index or PDSI in the US Drought Monitor). This study confirms the need for realistic ET estimation that was also highlighted in recent studies on PDSI trends [Hobbins et al., 2008; Sheffield et al., 2012] that have shown that failure to represent realistic ET dynamics can lead to questionable results.


[17] We thank the Wallone public service (Voies hydrauliques) for sharing their data. AJT acknowledges financial support from The Netherlands Organisation for Scientific Research through Veni grant 016.111.002 and from the Swiss National Foundation through the NFP61 DROUGHT-CH project. AFVL acknowledges support from the EU FP-7 project DROUGHT-R&SPI (282769), and CB from the German Science Foundation through grant BE1721/13.

[18] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.