Drought monitoring and early warning (M & EW) systems are a crucial component of drought preparedness. M & EW systems typically make use of drought indicators such as the Standardised Precipitation Index (SPI), but such indicators are not widely used in the UK. More generally, such tools have not been well developed for hydrological (i.e. streamflow) drought. To fill these research gaps, this paper characterises meteorological and hydrological droughts, and the propagation from one to the other, using the SPI and the related Standardised Streamflow Index (SSI), with the objective of improving understanding of the drought hazard in the UK. SPI and SSI time series were calculated for 121 near-natural catchments in the UK for accumulation periods of 1–24 months. From these time series, drought events were identified and for each event, the duration and severity were calculated. The relationship between meteorological and hydrological drought was examined by cross-correlating the 1-month SSI with various SPI accumulation periods. Finally, the influence of climate and catchment properties on the hydrological drought characteristics and propagation was investigated. Results showed that at short accumulation periods meteorological drought characteristics showed little spatial variability, whilst hydrological drought characteristics showed fewer but longer and more severe droughts in the south and east than in the north and west of the UK. Propagation characteristics showed a similar spatial pattern with catchments underlain by productive aquifers, mostly in the south and east, having longer SPI accumulation periods strongly correlated with the 1-month SSI. For catchments in the north and west of the UK, which typically have little catchment storage, standard-period average annual rainfall was strongly correlated with hydrological drought and propagation characteristics. However, in the south and east, catchment properties describing storage (such as base flow index, the percentage of highly productive fractured rock and typical soil wetness) were more influential on hydrological drought characteristics. This knowledge forms a basis for more informed application of standardised indicators in the UK in the future, which could aid in the development of improved M & EW systems. Given the lack of studies applying standardised indicators to hydrological droughts, and the diversity of catchment types encompassed here, the findings could prove valuable for enhancing the hydrological aspects of drought M & EW systems in both the UK and elsewhere.
Drought is widely recognised as a complex, multifaceted phenomenon (e.g. Van Loon, 2015). Unlike many other natural hazards, drought develops slowly, making it difficult to pinpoint the onset and termination of an event. Fundamentally, a drought is a deficit in the expected available water in a given hydrological system (Sheffield and Wood, 2011). Since Wilhite and Glantz (1985), drought has popularly been classified into various types (e.g. meteorological, hydrological, agricultural, environmental and socio-economic). The drought type generally reflects the compartment of the hydrological cycle or sector of human activity that is affected; deficits typically propagate through the hydrological cycle, impacting different ecosystems and human activities accordingly.
The desire to quantitatively identify and analyse drought duration, severity, onset and termination has led to the development of drought indicators. Lloyd-Hughes (2014) counted over 100 drought indicators in the literature, this proliferation reflecting the complexity of the subject matter. It has been argued that indicators should be chosen according to the type of drought in question; for example, meteorological indicators should not be used in isolation to characterise hydrological drought due to the non-linear responses of terrestrial processes to climate inputs (Van Loon and Van Lanen 2012; Van Lanen et al., 2013).
One of the primary uses of drought indicators is in monitoring and early
warning (M & EW), a crucial part of drought preparedness (Bachmair et
al., 2016). Little can be done to prevent a meteorological drought from
occurring, but actions can be taken to prevent or mitigate the impact of a
hydrological drought. An effective drought M & EW system is the foundation
of a proactive management strategy, triggering planned actions and responses
(Wilhite et al., 2000). There are numerous examples of drought M & EW
systems globally, for example, the US Drought Monitor
(
The Standardised Precipitation Index (SPI; McKee et al., 1993) is one of the most widely used drought indicators. It allows consistent comparison across both time and space as well as providing the flexibility to assess precipitation deficits over user-defined accumulation periods. The SPI also gives an indication of the severity and probability of the occurrence of a drought, with increasingly negative values indicating a more severe, yet less likely, drought (Lloyd-Hughes and Saunders, 2002). Despite the advantages and flexibilities of the SPI, there are known deficiencies. The choice of an appropriate probability distribution is still under investigation in the literature (e.g. Stagge et al., 2015; Svensson et al., 2015b) and the fitting of a probability distribution function to data with a high proportion of zeros can be problematic (Wu et al., 2007). It has also been noted that as the SPI accumulation period increases, the spatial behaviour of the index becomes more fragmented, making it more difficult to identify regions with similar patterns of drought evolution (Vicente-Serrano, 2006). Notwithstanding these deficiencies, the relative simplicity of calculation, comparability and flexibility of the SPI have led to an endorsement by the World Meteorological Organization as the indicator of choice for monitoring meteorological drought (Hayes et al., 2011). The use of precipitation alone does not take evaporative demand into account, which may result in drought severity being underestimated in regions or seasons with high levels of evapotranspiration. This led to the development of the Standardised Evapotranspiration Index (SPEI; Vicente-Serrano et al., 2010). A growing trend in drought M & EW research is the application of the same standardisation principles to other hydrological data types (soil moisture, streamflow, groundwater etc.), producing a family of standardised indices for all compartments of the hydrological cycle (Bachmair et al., 2016).
In the UK, there is no nationwide, drought-orientated M & EW system in
place. Regular hydrological reporting, published by the National Hydrological
Monitoring Programme in monthly Hydrological Summaries
(
The absence of a coherent drought-focused M & EW system across the UK is, in part, due to the lack of consensus on appropriate drought indicators or drought definitions for the UK. A number of drought analyses have been applied using a range of non-standardised indicators (e.g. Marsh et al., 2007; Rahiz and New, 2012; Watts et al., 2012), but the SPI and other standardised indicators have only been used in a few research studies (e.g. Hannaford et al., 2011; Lennard et al., 2016; Folland et al., 2015). Such indicators are generally not used operationally, although the Scottish Environment Protection Agency use a variant of standardised indicators for drought M & EW (Gosling, 2014) and Southern Water use SPI in their drought plan (Southern Water, 2013).
Recently, there has been growing interest in applying the standardised family
of indicators at the national scale in the UK. A Drought Portal
(
Many studies investigating hydrological drought characterisation and drought propagation have done so at the national, continental or global scale using modelled data (e.g. Vidal et al., 2010; Van Lanen et al., 2013), or at a smaller scale using a limited number of sites and observed data (e.g. Fleig et al., 2011; López-Moreno et al., 2013; Lorenzo-Lacruz et al., 2013b; Haslinger et al., 2014). Furthermore, few studies have used standardised indicators for both meteorological and hydrological droughts, which enables consistent characterisation across components of the hydrological cycle (and thereby potentially forming the foundation of a more integrated drought M & EW system). Very few observational studies have addressed the influence of climate and catchment properties on drought characteristics and propagation in a wide range of catchments demonstrating climatic and geological diversity. Studies have also tended to focus on a few characteristics representing geology or climate (e.g. Vidal et al., 2010; Lorenzo-Lacruz et al., 2013b; Haslinger et al., 2014) rather than a wide range of physiographic and land use properties, with the exception of the study by Van Loon and Laaha (2015) that used 33 catchment properties.
This study exploits the long streamflow and precipitation records held by the
National River Flow Archive (NRFA) for 121 catchments. Using observed data,
the utility of standardised indicators, the Standardised Precipitation Index
(SPI) and the Standardised Streamflow Index (SSI), for characterising drought
characteristics and propagation behaviour is assessed, specifically
addressing the following key questions:
How do meteorological and hydrological drought characteristics vary
spatially across the UK? Over which timescales are meteorological and hydrological droughts related? Which climatic and catchment properties influence hydrological drought
characteristics and the propagation from meteorological to hydrological drought?
Addressing these questions will supplement the existing knowledge of the
baseline drought hazard and propagation behaviours across the UK, in a set of
catchments with diverse properties, representative of hydro-climatic and
landscape variations. This knowledge is an important foundation for the
development of improved drought M & EW systems (Folland et al., 2015; Van
Loon, 2015), allowing preventative measures to be implemented, resulting in
reduced vulnerability and increased resilience to drought.
The UK has one of the densest hydrometric networks in the world. Hydrometric
data are archived and curated by the NRFA (
Location and cluster membership of UK Benchmark catchments selected for this study, including the nine case study catchments.
The selected Benchmark catchments were required to have at least 30 years of daily streamflow records 1961–2012 and each month was required to have at least 25 days of valid observations (in order to calculate mean monthly streamflow). Two ephemeral streams were excluded from the selection, as the truncation of the flows at zero would have been unhelpful when studying drought propagation. The selection criteria resulted in 121 catchments, providing good spatial coverage of the UK and a range of catchment types (Fig. 1). The selection of Benchmark catchments used here differs slightly to other published studies (e.g. Hannaford and Marsh, 2006; Chiverton et al., 2015a) because of differing selection criteria and the ongoing evolution of the Benchmark network. The NRFA also holds catchment average monthly precipitation data for each catchment based on observed UK Met Office data (Met Office, 2001; Marsh and Hannaford, 2008). At least 30 years of catchment average monthly precipitation data were available for each catchment between 1961 and 2012. In some cases, the catchment average monthly precipitation and mean monthly streamflow period of record differed in length, but all catchments had at least 30 years of data overlapping 1961–2012. Less than 10 % of catchments had a difference in record length of 5 or more years, and less than 3 % of catchments had a difference in record length of 10 or more years. When data completeness was calculated from the start of the catchment average monthly precipitation and mean monthly streamflow record, the proportion of missing data for each catchment was, on average, less than 0.01 % of months for precipitation data and less than 2 % of months for streamflow data.
Catchments were clustered using a previously developed classification system (Chiverton et al., 2015a) based on the temporal dependence in daily streamflow (characterised by calculating semi-variograms), enabling calculated drought characteristics to be analysed regionally. Where the catchments overlapped with those used in Chiverton et al. (2015a), the same cluster allocations were used. The 15 catchments that did not overlap between the two studies were assigned to the cluster for which the semi-variogram was closest to the mean semi-variogram of the cluster. Figure 1 shows the distribution of clusters across the UK for the 121 selected catchments. Clusters one and two are predominantly located in the upland north and west of the UK, have steeper slopes and less storage, are less permeable and have a higher amount of precipitation than the catchments in clusters three and four which are mostly located in the south and east of the UK. Predominant soil types differ between all four clusters. Clusters one and two can also be differentiated by elevation, while clusters three and four can be differentiated by their geology (Chiverton et al., 2015a).
Nine catchments covering a range of catchment types and sizes, as well as each cluster, were selected as case study catchments (Fig. 1) to allow more detailed, catchment-scale results to be displayed in this article.
The catchment average SAAR (standard-period average annual rainfall) 1961–1990 was used as a descriptor of the precipitation climate. The SAAR values were derived from a 1 km gridded map based on Met Office data (Spackman, 1993). In order to investigate the influence of the physical catchment on drought propagation, catchment properties were extracted for each catchment. The selected catchment properties (Table 1) have been found in previous studies to be significant for modifying climate–streamflow associations and in determining the temporal dependence of flows (Laizé and Hannah, 2010; Chiverton et al., 2015a). Base flow index (BFI), calculated from streamflow data (Gustard et al., 1992), although not technically a catchment property, has been found to reflect catchment geology, storage and release properties and so was used as an indicator of catchment storage (Bloomfield et al., 2009; Hidsal et al., 2004; Van Loon and Laaha, 2015). Catchment properties were derived from spatial data held by the NRFA (Marsh and Hannaford, 2008), the British Geological Survey, and in some cases extracted from the Flood Estimation Handbook (FEH; Bayliss, 1999).
The Standardised Precipitation Index (SPI) is calculated by first aggregating precipitation data over a user-defined accumulation period (often 1, 3, 6, 12 or 24 months). A probability distribution function is then fitted to the aggregated precipitation data for each calendar end-month (of the accumulation period) individually. It is then transformed to the standard normal distribution with a mean of zero and a standard deviation of one. This transformation makes the SPI comparable over time and space. The calculated SPI value represents the number of standard deviations away from the typical accumulated precipitation (McKee et al., 1993; Guttman, 1999; Lloyd-Hughes and Saunders, 2002). For SPI calculation, a Gamma distribution is often fitted to precipitation data. Several studies have tested the most appropriate probability distribution to fit to precipitation data and in many cases found Gamma to be acceptable (e.g. Guttman, 1999; Stagge et al., 2015). The Standardised Streamflow Index (SSI) uses the same principle as the SPI, aggregating streamflow data over the given accumulation periods (Vicente-Serrano et al., 2012b; Lorenzo-Lacruz et al., 2013a). In contrast to precipitation and SPI calculation, there is no widely adopted probability distribution function fitted to streamflow data for SSI calculation, and previously, numerous probability distribution functions have been used (e.g. Vicente-Serrano et al., 2012a). Here, we fit the Tweedie distribution, which has been shown to fit the same catchments well (Svensson et al., 2015b), for both catchment average monthly precipitation and mean monthly streamflow. The Tweedie distribution is a flexible three-parameter distribution that has a lower bound at zero (Tweedie, 1981; Jørgensen, 1987). The “SCI” package for R (Gudmundsson and Stagge, 2014) was used to calculate SPI and SSI for the period 1961–2012 and accumulation periods of 1–24 months. A new function enabled the parameter estimation in the “tweedie” package for R (Dunn, 2014) to be called within the SCI package (Svensson et al., 2015b). Accumulation periods are denoted as follows: SPI-x and SSI-x, for example, SPI-6 and SSI-3 correspond to a 6-month precipitation accumulation period and a 3-month streamflow accumulation period, respectively.
Drought events were defined as periods where indicator values were
continuously negative with at least 1 month in the negative series reaching a
given threshold (McKee et al., 1993; Vidal et al., 2010). Thresholds of
Missing catchment average monthly precipitation/mean monthly streamflow data would mean that no SPI or SSI value was calculated, potentially affecting duration/severity characteristics for some events. However, visual inspection of the data confirmed that for major UK drought events (Marsh et al., 2007), the impact of missing data was minimal and isolated to only a few catchments for streamflow data, and there were no missing precipitation data for major events. This and the low proportion of missing data in the data sets as a whole (Sect. 2) suggest the incidental months of missing data are localised and unlikely to have had a significant impact on the extracted drought characteristics.
Streamflow, and so the SSI, integrates catchment-scale hydrogeological
processes. As such, a comparison with the SPI provides an indication of the
time taken for precipitation deficits to propagate through the hydrological
cycle to streamflow deficits. SPI accumulation periods of 1–24 months and
SSI-1 time series were cross-correlated using the Pearson correlation
coefficient to analyse the most appropriate accumulation period of SPI to
characterise to SSI-1. The 1-month SSI also provides a good description of
low flows, similar to the 30-day mean flow, which is often used in studies of
annual minimum flows (e.g. Gustard et al., 1992). The SPI accumulation period
with the strongest correlation with SSI-1 was denoted SPI-
Independence of data is a requirement for many statistical analyses. However, because of temporal dependence, or autocorrelation, in the SSI-1 and in all the series of SPI accumulation periods exceeding 1 month, data are not independent. Correlations between two autocorrelated time series have fewer effective degrees of freedom than is assumed in a standard significance test. As such, using a standard significance test can result in an increased chance of concluding correlations are statistically significant (i.e. an increased rate of Type 1 error; Pyper and Peterman, 1998). In order to address and control Type 1 error rates, the “modified Chelton” method outlined in Pyper and Peterman (1998) was adapted to account for missing data, and used for calculating the effective degrees of freedom for a given data series. Details of the modified Chelton method are provided in the Supplement (Sect. S1).
Hydrological drought characteristics were plotted against SAAR and the corresponding correlation coefficients calculated. Spearman's correlation was used because of the non-linear relationships between the hydrological drought characteristics and SAAR. Clusters one and two were grouped together because of their location in the windward mountainous north and west of the country and clusters three and four were grouped together because of their location in the sheltered lowland south-east. Spearman's correlations were also used to quantify the relationship between the hydrological drought characteristics and catchment properties described in Table 1.
Summary of catchment properties used (after Chiverton et al., 2015a).
Median drought characteristics calculated for selected SPI and SSI
accumulation periods using thresholds of
For each accumulation period and catchment, drought events were identified
using thresholds of
Boxplots showing meteorological drought characteristics based on
SPI using thresholds of
For SPI-1, SPI-6 and SPI-18, there is little variation between the four clusters of catchments for the number of events and the median drought duration/severity characteristics (Fig. 2). This indicates that meteorological drought characteristics vary only modestly across the country over shorter accumulation periods once the precipitation has been standardised. The maximum duration/severity characteristics showed more differences between clusters, often showing a gradual change from clusters one to four. For SPI-1 the maximum duration of droughts in cluster one was generally short (between 4 and 9 months), whilst those in cluster four were longer (between 4 and 11 months). Similarly, for SPI-1 maximum severity, droughts in cluster one were less severe than those in cluster four. In contrast, the maximum duration and severity for SPI-6 was similar across all clusters, whilst for SPI-18 the median of the maximum duration decreases when moving from clusters one to three; the median of cluster four is higher than that of cluster two. The median maximum severity shows a different pattern for SPI-18 than for the shorter accumulation periods – median values increase (i.e. become less severe) moving from cluster one to three; cluster four has a lower (more severe) median severity than cluster three. Over these longer accumulation periods, inter-annual variability starts to become more influential; however, as will be discussed below (Sect. 5.1), the findings are somewhat surprising given that cluster one (mostly north-west Britain, the wettest and most upland part of the country) displays the longest drought durations and most severe events.
Maps of meteorological drought characteristics based on SPI-1 and SPI-6 (Fig. 3) again show little spatial variability in either the number of events or event duration and severity. The number of events at the 18-month accumulation period also shows little spatial variability; however, the duration and severity maps show longer, more severe meteorological drought events occurring in northern England and Scotland.
For SSI (Fig. 4), there is a larger difference between the clusters for SSI-1 and SSI-6 than is seen in SPI for the same accumulation periods (Fig. 2). As was the situation for SPI-1, the differences between clusters occurs gradually from cluster one to four. For SSI-1 fewer, but longer and more severe, events are identified in cluster four than cluster one. As the SSI accumulation period increases to 18 months, there is less difference between the clusters (Fig. 4); much like the spatial trends seen for SPI-18 (Fig. 2), whereby cluster one has a much greater range in maximum duration and severity than the other three clusters.
Maps showing selected meteorological drought characteristics based
on SPI-1, SPI-6 and SPI-18 using a threshold of
Maps of hydrological drought characteristics based on SSI show more spatial variability (Fig. 5) than the meteorological drought characteristics (Fig. 3). For SSI-1 and SSI-6, fewer, longer, more severe events occur in the south and east. As the accumulation period lengthens to 18 months, longer, more severe events occur in Scotland and the north of England. Despite this, the number of events remains fewer than 10 throughout the UK, with the most events occurring in the south-east of England.
Time series plots of SPI for selected accumulation periods in Fig. 6 and SSI in Fig. 7 show the highly variable time series for the 1-month accumulation period. As the accumulation period increases to 6 and 18 months, the time series become smoothed, with both wet and dry periods becoming more prolonged. Figure 6 also shows that at the longer accumulation period (SPI-18) for the two Scottish case study catchments (the Dee and Cree), the early time series is dominated by dry events, while the later time series is dominated by wet events. This is in contrast to the remaining case study sites in England and Wales, which show more regular fluctuations between wet and dry events throughout the SPI time series. Similar long-term trends can be seen in the SSI time series for the case study catchments in Fig. 7. The implications of these patterns for application of the SPI and SSI will be returned to in the discussion (Sect. 5.1).
Pearson correlations between SSI-1 and different accumulation periods of SPI
(1–24 months) showed that for the majority of catchments, SPI-
Figure 8 shows that for catchments in the north and west of the UK, SPI-
Figure 9 shows the correlations between all SPI accumulation periods
(1–24 months) and SSI-1. The strength of the correlations reflects the
spatial variability seen in SPI-
When SPI values (for accumulation periods of 1–24 months) were correlated
with lagged SSI-1, the strongest correlation was found at a lag of zero
months (i.e. no lag) for all catchments. One would expect the SPI
accumulation period most strongly correlated with lagged SSI-1 (lagged
SPI-
Boxplots showing hydrological drought characteristics based on SSI
using thresholds of
Case study catchments in the south and east (Harpers Brook, Thet, Lambourn
and Great Stour) show stronger and significant (
Table 3 shows the Spearman correlations between hydrological drought
characteristics (based on SSI and includes a propagation indicator, SPI-
Maps showing selected hydrological drought characteristics based on
SSI-1, SSI-6 and SSI-18 using a threshold of
Figure 11 shows the relationship between SAAR and hydrological drought
characteristics for all catchments, with points coloured by BFI to give an
indication of the relationship between the hydrological drought
characteristics and catchment storage. The plots show BFI decreasing as SAAR
increases, a reflection of the fact that most high BFI, i.e. high storage,
catchments are located in lowland south-east England that receives less
precipitation. Figure 11 shows positive relationships between SAAR and
median/maximum severity, but as SAAR reaches
Correlation coefficients for Spearman correlations between
hydrological drought characteristics and SAAR (
Case study catchment SPI time series for selected accumulation periods.
Case study catchment SSI time series for selected accumulation periods.
Map of catchments showing the SPI accumulation period most strongly
correlated with SSI-1 (SPI-
Figure 12 shows the relationship between SAAR, hydrological drought
characteristics and propagation but for catchments in clusters three and four
only (the results for clusters one and two are not shown as they are broadly
similar to the results for the full data set). The relationship with SAAR for
these clusters, as shown in Table 3, is weaker than those for all catchments
(Table 3, Fig. 11). Instead, it is clear that catchments from clusters three
and four can be split into two groups, those with higher BFI values and those
with lower BFI values (Fig. 12); catchments were split based on the median
BFI for clusters three and four. Each group separately follows the same
relationship with SAAR, as described for the full data set in Table 3 and
Fig. 11. This is with the exception of the
Hydrological drought characteristics for clusters one and two showed strong correlations with elevation properties. This, in conjunction with the strong correlations between the hydrological drought characteristics and SAAR (Table 3), indicate that the climatological control is the dominant factor influencing hydrological drought characteristics in the typically wet, upland catchments of clusters one and two mainly located in the north and west of the UK. The variation in precipitation across the lowland south and east is relatively minor in comparison to the north and west of the UK, but exhibits heterogeneity in geology and land cover, allowing catchment properties to exert a greater control on the hydrological drought characteristics in clusters three and four. As such, in the following sections, only results for clusters three and four are presented and discussed. The correlations between hydrological drought characteristics and catchment properties for clusters one and two can be found in the Supplement (Sect. S3; Fig. S5).
Figure 13 shows that when clusters three and four are grouped together, both
the median and maximum hydrological drought duration have a strong positive
correlation with catchment properties related to storage, such as the
percentage of highly productive fractured rock (
PROPWET has significant correlations with all the hydrological drought
characteristics (except the
Heat map showing correlations of SPI accumulation periods of 1–24 months with SSI-1 for all catchments.
Heat maps for case study catchments showing correlation between
SSI-1 lagged by 0–6 months and SPI accumulation periods of 1–24 months. The
lagged SPI-
Relationship between hydrological drought characteristics based on
SSI-1 using a threshold of
Relationship between hydrological drought characteristics based on
SSI-1 using a threshold of
Heat map showing the correlations between selected hydrological
drought characteristics based on SSI-1 using a threshold of
The percentages of arable land and grassland were significantly correlated
for all hydrological drought characteristics (
All hydrological drought and propagation characteristics were weakly correlated with catchment properties such as area, slope, the percentage of mountain, heathland and bog and elevation properties (generally non-significant). The use of “near-natural” Benchmark catchments meant that they are little influenced by urban areas or regulation; as such, the catchment properties urban extent and FARL were excluded from the analysis.
Drought characteristics were extracted from SPI and SSI time series from a wide and representative sample of UK catchments. This provides a comprehensive view of meteorological and hydrological droughts at the national scale, assessed using the standardised indicators that have been relatively under-used in the UK. Overall, the results show that, for shorter accumulation periods, there is comparatively little difference between catchment types (as shown by the clusters, Fig. 2) or around the country in meteorological drought characteristics extracted from SPI time series (Fig. 3). Although the UK has an order of magnitude precipitation gradient across the country, there is little difference in the median of the meteorological drought characteristics. Similarly, Van Loon and Laaha (2015) found little spatial variation in the number and average duration of meteorological events between clusters of Austrian catchments. However, this study shows that there are pronounced regional differences in the maximum drought duration and severity, which is supported by Folland et al. (2015), who note that the north-west has a more variable climate and the south-east is subject to longer dry spells, and that in practice the two regions experience droughts in opposition. Regional differences in meteorological drought duration and severity have also been found elsewhere, e.g. in Valencia, where spatial variation was found to be the result of both catchment relief and climatic variability across the region (Vicente-Serrano et al., 2004).
In contrast, hydrological drought characteristics extracted from SSI time series show distinct regional variations and differences between catchment types. SSI-1 and SSI-6 results show fewer, longer, more severe droughts occurring in southern and eastern regions of England, which are dominated by groundwater-fed rivers on permeable aquifer outcrops (Figs. 4 and 5). These results parallel those seen in Vidal et al. (2010), who found fewer, but longer, and more severe events in gridded, modelled streamflow data in northern France, which is dominated by groundwater-fed rivers and large aquifer systems, than in southern France. These results show that although standardisation is carried out for each month, the month-to-month autocorrelation in streamflow means that droughts defined using a given SSI threshold can take on very different characteristics around the country, according to hydrological memory.
Given the climatological gradient in the UK, the long, severe droughts identified using SPI-18 and SSI-18 in Scotland were unexpected (Figs. 3 and 5). Previous studies characterise droughts in Scotland as being shorter and less severe than those in the south and east of England (Jones and Lister, 1998; Marsh et al., 2007). These apparent long droughts are a result of strong long-term increasing temporal trends in run-off, primarily driven by the inter-decadal variability in the North Atlantic Oscillation, as have been widely reported (e.g. Hannaford, 2015). As there is a strong trend, the standardised approach makes it appear that there is one long drought in the early record and pronounced wetness at the end (Figs. 6 and 7). In one sense, this is a perfectly valid finding; the dryness of the early period is important when examining long meteorological droughts. However, in another sense, it is misleading, as “droughts” (in terms of triggering a particular impact) with a duration of 18 months are less influential on reservoir levels and water resources planning in the north and west of the UK. This is, in part, due to the lack of sub-surface storage in these responsive catchments. A short and intermittent wet spell can return the catchment to normal conditions as there is limited storage in which to build up deficits. The dangers of using standardised indicators in the presence of non-stationarity and multi-decadal variability in atmosphere–ocean drivers have been highlighted elsewhere (e.g. McCabe et al., 2004; Núñez et al., 2014).
SSI-1 was cross-correlated with SPI accumulation periods of 1–24 months to
identify the timescale over which precipitation deficits propagate through
the hydrological cycle to produce streamflow deficits. The mapping of SPI-
Figure 9 shows the strong correlations of SSI-1 with a range of SPI accumulation periods. Although, in general, catchments in the south and east can be said to be permeable, there is a range of geologies and not all catchments are highly permeable or are largely influenced by groundwater. In less permeable catchments, the strong correlations at long SPI accumulation periods are likely to be partially a result of the stronger seasonality in the south and east where evapotranspiration is higher (Kay et al., 2013). Where this enhanced seasonality in effective rainfall (precipitation minus evapotranspiration) induces a stronger relationship between streamflow on successive days, autocorrelation in streamflow increases (Chiverton et al., 2015b). This autocorrelation favours the longer SPI accumulation periods.
The lagged correlations for the Lambourn, Thet and Great Stour all show
strong correlations across both a range of lags and SPI accumulation periods
(Fig. 10) just as lagged correlations of SPI do with the SGI (Bloomfield and
Marchant, 2013). While we find (along with Folland et al., 2015) the
strongest correlation with streamflow occurs when SPI is not lagged, the
presence of strong correlations even at lags of several months (up to 6
months, in some cases) demonstrates potential for early warning of
hydrological drought based on persistence of meteorological anomalies.
Indeed, this characteristic is already used for making successful seasonal
streamflow forecasts based on persistence in the UK (Svensson, 2016). This
forecasting method currently estimates whether flows are likely to be in
high, medium or low bins, but the results presented here suggest that further
work could focus more specifically on drought indicators. Rivers in the north
and west do not benefit from the slow release of stored groundwater; instead,
methods that reflect the expected meteorological conditions are needed for
making skilful streamflow forecasts (Svensson et al., 2015a). The closeness
of the lagged SPI-
Analysis of SAAR and hydrological drought characteristics showed that for upland catchments (clusters one and two), the general precipitation climate (characterised by SAAR) was much more important in influencing hydrological drought characteristics (Table 3) than in lowland catchments (clusters three and four; Table 3, Fig. 12). Table 3 and Fig. 11 also show the strong relationship between SAAR and the hydrological drought characteristics for all catchments together, a result of the prominent precipitation gradient seen between the north-west and the south-east of the UK. Similarly, Haslinger et al. (2014) found that climate was more influential than catchment properties in Austrian catchments, where small-scale geological differences could not explain the variation in correlation significance between streamflow and meteorological drought indices across four geologically similar regions. Precipitation was found to be necessary to produce a significant model of median discharge drought duration, in addition to catchment properties for Austrian catchments (Van Loon and Laaha, 2015). A combination of the weather type (based on the objective Grosswetterlagan weather classification) and catchment properties was found to contribute to the hydrological response time in catchments across the UK and Denmark (Fleig et al., 2011). On a broader scale, in a study of 808 near-natural streamflow records in Europe and the USA, Tijdeman et al. (2015) found that climate classification systems that included absolute precipitation were best for differentiating catchments based on hydrological drought duration. In addition, BFI, the seasonality of precipitation and the occurrence of hot summers were important individual controls on hydrological drought duration.
For clusters three and four, mainly located in the lowland south and east of the UK, SAAR was weakly correlated with hydrological drought characteristics (Table 3, Fig. 12). A small range of, and generally lower, average annual precipitation, and the presence of permeable aquifer outcrops, means that catchment properties, particularly those related to catchment storage (for example, BFI, percentage of highly productive fractured rock and PROPWET), are more influential than SAAR in determining the drought and propagation characteristics (Fig. 13). Groundwater storage and responsiveness have been found to be important in determining drought duration and severity. In Austrian catchments, Van Loon and Laaha (2015) found that the mean duration of discharge droughts had a strong positive correlation with BFI, as was found in this study for clusters three and four (Fig. 13). Van Loon and Laaha (2015) also found that other catchment properties representative of catchment storage, such as aquifer depth and the presence of lakes and wetlands, had weak correlations with mean discharge drought duration. In the present study, stronger positive correlations were found between drought duration and the percentage of highly productive fractured rock and the percentage of the catchment with no gleyed soils. The weaker relationships in the Austrian study were thought, however, to be a result of missing data for some of the catchment properties, rather than a lack of influence (Van Loon and Laaha, 2015).
Laizé and Hannah (2010) found slope, BFIHOST (a measure of catchment
responsiveness derived using the HOST (hydrology of soil types)
classification; Boorman et al., 1995), percentage of arable land, elevation
and bedrock permeability to significantly influence seasonal flows in UK
Benchmark catchments. They classified catchments into upland impermeable,
lowland permeable and lowland impermeable groups. In lowland permeable
catchments, regional climate was a poorer predictor of streamflow due to the
climate buffering provided by the permeable catchment. Chiverton et al. (2015a)
found that BFIHOST, the percentage of highly productive fractured
rock, the depth to the gleyed soil layer, the percentage of arable land,
slope and PROPWET were all significant catchment properties influencing the
temporal dependence of flows in UK Benchmark catchments. Temporal dependence
can be thought of as indicative of the average lag between meteorological
and hydrological signals; catchments in clusters one and two (three and
four) showed less (more) temporal dependence in streamflow (Chiverton et
al., 2015a). A similar pattern was found here in SPI-
Chiverton et al. (2015a) found the percentage of arable land to be the best
catchment property to distinguish clusters based on the temporal dependence
of streamflow. However, they argued that this was likely due to the positive
relationship with the percentage of highly productive fractured rock and the
negative relationship with other catchment properties such as high elevations
and PROPWET. Arable land, in effect, characterises permeable, lowland
well-drained catchments with high storage. In Austrian catchments, forest
cover was positively, but weakly, correlated with both discharge drought mean
duration and mean deficit (Van Loon and Laaha, 2015). In the present study,
the percentage of woodland was significantly (
The SPI is widely used in existing drought M & EW systems, but the SSI is less widely adopted (Bachmair et al., 2016). This may be a result of the poorer availability of streamflow data in comparison to precipitation data, especially at the short timescales involved in producing useful drought M & EW products. However, the monitoring of hydrological variables and the incorporation of hydrological drought indices is beneficial for effective drought planning and management, and it is particularly useful for communication purposes if both precipitation and streamflow are monitored in a comparable way. In locations where streamflow data are available, the SSI can be used directly in drought M & EW. While this is preferable, SPI could potentially provide a surrogate for hydrological impacts, provided appropriate response times are known.
The correlation results, Fig. 8, showing the spatial variability in SPI-
The results also highlight some of the problems with using the SPI and SSI when calculated for long accumulation periods for locations that have seen an increasing, or decreasing, long-term trend in precipitation or streamflow such as Scotland. Although being able to calculate SPI or SSI for any user-defined accumulation period makes the indicators more flexible, it is essential that meaningful accumulation periods should be chosen to capture the drought characteristics typical of current meteorological or hydrological conditions (Vicente-Serrano and López-Moreno, 2005). For example, the long, severe droughts shown in Scotland for both SPI and SSI for the 18-month accumulation period (Figs. 3 and 5) are not typical of the droughts that have been observed (Jones and Lister, 1998; Marsh et al., 2007). The short droughts that have been most influential for water resources in Scotland are better captured by shorter accumulation periods that are less confounded by the long-term increased precipitation and streamflow trends. Moreover, the findings reaffirm that accumulation periods should be chosen based on likely impacts; Bachmair et al. (2015) observed that short SPI accumulation periods are strongly linked to impacts in north-west Britain, while longer SPI accumulation periods trigger impacts in the south-east.
With the importance of groundwater, particularly in the south and east of England for water supply, to understand fully hydrological drought characteristics and propagation, it is necessary to include a groundwater component to the analysis. Furthermore, although catchment storage plays a key role in determining hydrological drought characteristics and propagation in the south and east, the seasonality and autocorrelation of streamflow, caused by evapotranspiration, will also be influential. Undertaking propagation analysis though the hydrological cycle, using the Standardised Precipitation Evapotranspiration Index (SPEI; Vicente-Serrano et al., 2010), the Standardised Streamflow Index (SSI; Vicente-Serrano et al., 2012b; Lorenzo-Lacruz et al., 2013a) and the Standardised Groundwater Index (SGI; Bloomfield and Marchant, 2013) would therefore give a clearer picture of the climate and catchment properties influential on drought duration, severity and propagation, paving the way for more integrated drought M & EW. In addition, the seasonal variability in drought propagation should be investigated. Studies in Spain found there were distinct differences between the duration of the SPI and SPEI accumulation period most strongly correlated with standardised monthly streamflow (SSI-1) depending on whether full time series or individual months were cross-correlated (Vicente-Serrano and López-Moreno, 2005; López-Moreno et al., 2013). The seasonal component is particularly important in the south and east of England, where summer streamflow and water resources are highly dependent on winter recharge.
The use of near-natural Benchmark catchments in this study has allowed the investigation of hydrological drought characteristics and propagation processes without results being confounded by artificial influences. However, it is often man-made systems that the human population is most reliant upon for water supply, agriculture, etc. Understanding these processes in catchments affected by anthropogenic activities is crucial for truly effective drought M & EW systems. Further work on the drought and propagation characteristics in these modified systems will be much more challenging (Van Loon, 2015). It is likely that the combination of human activities, alongside natural catchment and climate characteristics, will produce more divergent results. However, the results from this study could be used to stratify catchments based on their climate and catchment properties when tackling the challenges of quantifying drought hazard in catchments with anthropogenic modifications.
Finally, as with most observation-based studies of drought, the brevity of
available hydrological records is a constraint. The period of analysis
(1961–2012) does not capture the full range of known hydrological
variability, and previous studies have highlighted the importance of earlier
droughts in the UK (Marsh et al., 2007). Longer records could influence the
drought characteristics presented here, although the same regional picture
and propagation characteristics would undoubtedly emerge. Similar methods to
those used here will be applied in the future to longer records. Localised
reconstructions of drought are becoming available (Lennard et al., 2016;
Spraggs et al., 2015), while national-scale reconstruction research is in
progress (e.g. Historic Droughts;
Meteorological and hydrological drought characteristics and propagation behaviours of 121 near-natural UK catchments were analysed using SPI and SSI over a range of accumulation periods. Meteorological drought duration and severity characteristics showed little spatial variability, whilst hydrological drought characteristics showed many (few), short (long), less (more) severe events in the north and west (south and east) of the UK. Catchments underlain by aquifers tended to show more of a delay in the propagation of drought from meteorological to hydrological drought, with longer SPI accumulation periods most strongly correlated with SSI-1. Standard-period average annual rainfall was found to be important for hydrological drought duration, severity and propagation in the north and west of the UK where catchment storage is generally low, whilst in the south and east, catchment storage and other catchment properties are more influential on hydrological drought duration, severity and propagation. The greater understanding of the UK drought hazard provided by this study can be used as a foundation for future developments of M & EW in the UK, laying the foundations for better drought preparedness and increased resilience to drought.
This study is an outcome of the Belmont Forum Project DrIVER (Drought Impacts: Vulnerability Thresholds in monitoring and Early warning Research). Financial support was provided by the UK Natural Environment Research Council (grant NE/L010038/1). We thank the UK National River Flow Archive (NRFA) for all streamflow and precipitation data, and both the NRFA and the British Geological Survey for catchment property data. The streamflow and precipitation data used here can be obtained from the NRFA on request, as can the majority of catchment properties. The authors would like to thank Sergio Vicente-Serrano, an anonymous reviewer and the editor, Jean-Philippe Vidal, for their constructive comments that helped improve the paper. Edited by: J.-P. Vidal