Streamflow trends in western Britain

Authors


Abstract

[1] Concerns about recent river flow changes are increasing, given their importance for water management, flooding, drought mitigation, and geomorphological and ecological processes. For the first time, this paper develops a full picture of recent streamflow trends over the period 1962–2001 for western Britain across the entire flow spectrum, at annual and seasonal resolutions, for 56 gauging stations and three study periods. Annual and seasonal trends are calculated and mapped using the non-parametric Mann-Kendall test with bootstrap resampling. Several significant, mainly positive, trends are observed at high and low flows, while mean flows are relatively stable. A marked east-west gradient of streamflow trend emerges, with more significant and positive trends in the mountainous west, particularly for autumn and winter high flows. Positive annual low flow trends are found across western Wales, largely driven by increasing summer base flow. Many trends are consistent with previously observed rainfall and North Atlantic Oscillation (NAO) changes.

1. Introduction

[2] There is a growing international perception that flood and drought frequencies are increasing in magnitude and/or frequency [e.g., Andreadis and Lettenmaier, 2006; Burn and Hag Elnur, 2002; Kundzewicz et al., 2005; Lettenmaier et al., 1994; Lins and Slack, 1999; Zhang et al., 2001]. In Europe specifically, severe current and recent droughts and major flooding episodes (e.g., UK 1998 and 2000; Danube and Elbe 2002 and 2006 [Ulbrich et al., 2003]) have heightened concerns, and are reflected in river flow trend analyses for northern, central and eastern Europe [e.g., Bîrsan et al., 2005; Cigizoglu et al., 2005; Hyvarinen, 2003; Lindstrom and Bergstrom, 2004]. However, in western Europe, although a number of useful trend studies have addressed part of the flow spectrum or focused on individual basins [e.g., Dixon et al., 2005; Lawler et al., 2003; Robson, 2002], there has been no systematic definition of streamflow trends for the full streamflow spectrum from low flows to high flows, at annual and seasonal resolutions across several basins. This is an important omission because flow trends may differ significantly between streamflow magnitudes and seasons [e.g., Bîrsan et al., 2005].

[3] In this paper, we address this research gap in western Britain for the first time by quantifying, for a dense network of gauging stations, recent trends across the entire flow spectrum from low to high flows, and for seasonal and annual series over three different time periods. Western Britain lies in a representative high-rainfall, hydrologically dynamic part of the European Atlantic rim [Lawler, 1987], and occupies a dominant leading-edge position for precipitation systems derived from North Atlantic cyclogenesis. Such environments are thought to be especially susceptible to climate change [Rodda and Rodda, 2000], and flows here are less affected by anthropogenic impacts characteristic of populated areas of the eastern lowlands. Given its high runoff, western Britain is also a key national water resource, and knowledge of flow trends is crucial in applied terms. This intensive study provides a key platform for future work linking hydrological responses, including extreme flows, to climate and atmospheric circulation variability in the North Atlantic region [e.g., Kiely, 1999; Kingston et al., 2006; Lawler et al., 2003]. Such comprehensive flow trend analyses also help to constrain predictions of future hydrological change, assess likely geomorphological and hydro-ecological impacts, and inform management of flood, drought and water resource issues.

2. Data and Methods

[4] The study area covers all of Wales and the English Midlands (combined area ≈46,800 km2) and includes the main river basins of Severn, Trent and Wye. Precipitation totals and seasonality decline strongly from the mountains of Wales in the west (c.2800 mm per annum) to the Midland lowlands of the east (c.650 mm per annum) [Lawler, 1987]. Daily Mean Flow (DMF) records were selected from those held within the National River Flow Archives (NERC, National River Flow Archives, http://www.ceh.ac.uk/data/nrfa/). A minimum record length of 25 years of continuous data was set, and stations were chosen that were free from flow retention, notable gauging errors and changes in station location over the record length. A total of 56 gauging stations were selected, with contributory drainage basin areas ranging from 18 km2 to nearly 7500 km2.

[5] DMF records (derived as the average of 24 hours of 96 × 15-minute gaugings) were permutated to calculate time series of flow quantiles at annual and seasonal resolutions. The standard UK Meteorological Office seasons were used: Winter (Dec/Jan/Feb); Spring (Mar/Apr/May); Summer (Jun/Jul/Aug); and Autumn/Fall (Sep/Oct/Nov), to ensure comparability with other relevant hydroclimatological studies. The annual or seasonal x-quantile, Qx, of streamflow (q) was obtained for each year (i) as the value at which the probability Pr(q ≤ Q) = x [e.g., Bîrsan et al., 2005]. Trend tests were applied to time series of Qx(i), where i = 1, …, t years.

[6] In order to investigate the full streamflow spectrum, trends were analysed for 11 quantiles of the DMF distribution, including the minimum (QMin) and the maximum (QMax) and mean (QMean). At annual and seasonal resolutions, the following quantiles of flow (as opposed to exceedance probabilities) were calculated: 10th (Q0.1) low flow percentile, 20th (Q0.2), 30th (Q0.3), 40th (Q0.4), 50th (Q0.5), 60th (Q0.6), 70th (Q0.7), 80th (Q0.8) and 90th (Q0.9) high flow percentile.

[7] One potential issue is the impact on results of the length of sampling window over which trends are calculated [see Dixon et al., 2006]. In order to assess this, analysis was conducted for three time spans of t = 25 years (1977–2001) for 56 stations; t = 30 years (1972–2001) for 44 basins and t = 40 years (1962–2001) for 24 basins.

[8] Monotonic trends were analysed using the nonparametric “distribution free” Mann-Kendall test for the given quantiles of the cumulative streamflow distribution [see Lins and Slack, 1999]. A detailed explanation of the rank-based Mann-Kendall hydrological trend test used here is given by Kundzewicz and Robson [2000, 2004]: this is often employed for flow trends [e.g., Bîrsan et al., 2005; Lettenmaier et al., 1994; Zhang et al., 2001]. Being distribution free and robust with regard to outliers, it is well suited to the analysis of streamflow time series which tend to be non-normally distributed.

[9] An assumption made in this test is that the data are not serially correlated. As both the annual and seasonal time series analysed in this study are sampled at yearly intervals the impacts of serial correlation are likely to be low [Lins and Slack, 1999]. Also, Bîrsan et al. [2005] found low serial correlation coefficients in similar data sets and concluded that the more conservative method of pre-whitening time series that displayed autocorrelation had little difference in such cases.

[10] Test statistic significance levels were determined using the flexible and robust method of bootstrap resampling (with 1000 resamples) as it makes relatively few assumptions about the independence and distribution of the data set [e.g., Kundzewicz and Robson, 2004]. This paper focuses on the trend results found to be significant at or above the 10%, 5% and 1% significance levels (α ≤ 0.1, 0.05, 0.01; two-tailed test, as the direction of change is assumed unknown).

3. Results

[11] Figure 1 summarises the trend analyses from nearly 1500 annual and 2700 seasonal time series permutations covering 25–40 years of record, and shows interesting, but complex, patterns across the flow quantiles, seasons and study period lengths. It is clear that the percentage of gauging stations exhibiting flow trend is highly dependent on the quantile and season selected (Figure 1). Thus, results are discussed for annual analyses and seasonal breakdowns of the complete flow profile, with particular focus on mean, high and low flows, including spatial patterns. Regional trend-mapping was completed for the entire flow spectrum; however, this paper will focus on spatial results for mean, high and low flow quantiles only, as these represent the key series for important flood and drought issues. Spatial patterns vary with analysis period (Figure 2), further highlighting the influence of record length on trend analyses [e.g., Dixon et al., 2006].

Figure 1.

Percentage of streamflow stations showing linear trends (significant at α ≤ 0.1) in annual flow quantiles for (a) t = 25 years; (b) t = 30 years; (c) t = 40 years and seasonal flow quantiles over 25 years (1977–2001); (d) Dec/Jan/Feb; (e) Mar/Apr/May; (f) Jun/Jul/Aug; (g) Sept/Oct/Nov.

Figure 2.

Geographical variations in annual streamflow trends (significant at α ≤ 0.1) for QMin, Q0.1, QMean, Q0.9, and QMax, analysed at t = 25 years, t = 30 years and t = 40 years. Triangles indicate direction of trend and statistical significance.

3.1. Annual Analysis

[12] Three key points emerge for the annual trends. First, up to 54% (but more usually up to 30%) of stations show streamflow trends (significant at α ≤ 0.1) for a given flow level, and the percentages tend to increase with record length (Figure 1). Second, most significant flow changes are observed for the higher and lower flows, and few trends emerge for average flows: this therefore imparts a strong symmetry in trend prevalence about the median flow (Q0.5) (Figures 1a, 1b and 1c). Third, most trends are positive, especially for the higher streamflows.

[13] Average flows (Q0.3–Q0.6) appear generally stable over time (Figures 1a, b and c), and very few stations (<13%, α ≤ 0.1) show trend, although this rises to 25% for QMean for the 40 year period, 1962–2001. The only spatial pattern evident is a clustering of positive trends in south and mid-Wales over the 40 year period (Figure 2i), but many of these gauging stations lie on the same river (River Wye) or share common runoff source areas.

[14] For high flows (≥Q0.9), significant positive trends (α ≤ 0.1) emerge for up to 18% of gauging stations over the 30 year period (Figure 1b; QMax), and up to 54% of stations for the 40 year period (Figure 1c; Q0.9). Some coherence in spatial pattern emerges for the positive trends (Figure 2): in particular, at Q0.9 over the 40 year period, a southwest-northeast gradient is evident, with most of the positively-trending sites, and all of the highly significant ones, concentrated in south and mid Wales, with none to the north-east (Figure 2f).

[15] Annual low and base flow conditions are represented in this study by quantiles ≤Q0.2. Positive trends predominate at all periods (up to 32% of all stations at QMin), although negative trends become more important in the 40-year time series, affecting 17% of stations (α ≤ 0.1) at QMin (Figures 1a–1c). Although the east-west gradient is less striking than for high flows, there is a tendency for increasing low flows to occur in the southwest and the montane and upland zones of north Wales (Figure 2), especially at t = 25 (Figures 2j and 2m) and t = 30 (Figures 2k and 2n). The few negatively-trending sites tend to be confined to the north east (e.g., Figure 2l).

3.2. Seasonal Analysis

[16] A seasonal breakdown analysis of streamflow quantiles allows key seasons for flow change to be identified, and results to be compared with previous seasonal analyses of precipitation inputs (Figures 1d–1g). This paper will present analysis over the 25 year period, 1977–2001; however similar patterns were found over 30 and 40 year periods. The percentage of stations displaying significant seasonal flow trends is less than 43% (Figure 1), although all seasons display statistically significant trends for most quantiles.

[17] Average seasonal flows, as for annual series, show few trends (<20% of gauging stations, at α ≤ 0.1), most trends emerging for the higher and lower flows: this underscores the need to include flow extremes in such analyses (Figures 1d–1g). Summer is the main season affected (Figure 1f): for QMean, for example, 9% of stations show trends, but this represents just five stations, and all are at α ≤ 0.1 significance. Three of these are positive and lie to the west of the two sites displaying negative trends in the Midlands (Figure 3k).

Figure 3.

Geographical variations in seasonal streamflow trends (significant at α ≤ 0.1) over t = 25 years for QMin, Q0.1, QMean, Q0.9, and QMax.

[18] For high flows, the key seasons are winter and autumn, and the dominant trend is positive. For winter QMax, 43% of sites show significant trends (α ≤ 0.1), more than half of which are positive (Figure 1d). In autumn, all 29% of sites showing significant trends emerge as positive (Figure 1g). Winter QMax changes (Figure 3a) show clear east-west differences, with positive trends occurring in northern, central and southeast Wales, and a smaller region of declining high flows to the northeast. At Q0.9, flow increases are largely confined to South Wales (Figure 3e). Interestingly, in autumn, all stations displaying increasing high flows (Figure 3d) are found to the east of the region, and very few affected stations lie in Wales.

[19] For seasonal low flows the key season is summer (Figure 1f), when up to 25% of gauging stations reveal significant trends at α ≤ 0.1, most of which are positive. For QMin, for example, 18% of stations registered an increase over time but this percentage systematically declines as flows increase (Figure 1f). Positive trends also emerge in spring for QMin at 10% of sites (Figure 1e). However, there is also a reasonably constant, but low, percentage (<10%) of stations with declining flows in spring and summer: indeed this is visible for most quantiles (Figures 1e and 1f). Again, a coarse east/west divide is evident, with increasing summer flow minima in the west and southwest, and all declining minimum flows concentrated in the northeast (Figures 3o and 3s).

4. Discussion

[20] This paper develops, for the first time, a full picture of recent streamflow trend for western Britain across the entire streamflow spectrum for 56 stations, at annual and seasonal resolutions. The results reveal complex seasonal and regional patterns of hydrological change: although many flow records show little trend, for those that change significantly trend is much more common at high and low flows, especially for the high streamflow quantiles. However, while some negative trends emerge, few individual gauging stations show significantly increasing high and low flows, so mean and median flow conditions generally have been relatively stable (Figure 1) and, if this trend continues, it should ease some water resource planning concerns. This low percentage of significant trends in the middle of the streamflow spectrum in some cases reflects a balancing of significant increasing trends at high flows and negative trends at low flows. Indeed, analyses restricted to average tendencies would have failed to capture the numerous trends in streamflow extremes most relevant to flooding, drought and river system impact issues.

[21] For example, increasing annual Q0.9 high flows are observed for more than half the sites, and the trend is especially evident in southern Wales (Figure 2f). Major flooding at Easter 1998 and autumn 2000, towards the end of the time series, may reinforce trends in some basins. Separate seasonal analyses revealed that most high-flow increases occur in winter and autumn (Figure 3): this provides some support for the suggestion of Green and Marsh [1997] that runoff seasonality from some steep, more responsive catchments of north-west Europe has become more pronounced.

[22] Lawler [1987] observed a strong east-west climatological gradient in western Britain, with increasing precipitation totals, rainfall intensity, effective rainfall, rainfall seasonality and frequency of heavy daily rainstorms in winter as one moves westwards towards the Welsh mountains. We find here that a marked east-west gradient also occurs in streamflow trend and is manifested in a number of ways: (a) at annual resolutions, all very significant high-flow increases are in the west; (b) high-flow increases tend to occur in autumn in the east, but in winter in the west: this trend should be energising already dynamic hydrological systems (Figures 3a and 3e); and (c) low flow increases are mainly found in the west of the region, with declining base flows in the northeast. This east-west (or occasionally northeast - southwest) divide may reflect the dual influence of increased oceanicity and greater basin elevation and orographic rainfall impacts towards the west.

[23] Indeed, many of the trends identified here are consistent with changes to rainfall climatology and circulation indices observed by others. For example, winter high-flow increases are likely to be related to a tendency for increasing prolonged heavy rainfall events in northern and western regions of Britain, including North Wales [Fowler and Kilsby, 2003], and increases in winter precipitation totals and the proportion of heavy rainfall events in the winter precipitation regime identified by Osborn and Hulme [2002]. This trend is also found in central England [Osborn et al., 2000], where autumn maximum flows are generally rising (Figure 3).

[24] One possible driver of increases in winter rainfall, and possibly streamflow in Wales, is the North Atlantic Oscillation (NAO). Preliminary basin-scale work in north Wales by Dixon et al. [2005] found significant seasonal correlations between riverflow, rainfall and the NAO index. Positive NAO index values tend to be associated with winter storm track trajectories from the North Atlantic, with their attendant oceanic moist air parcels. Overall, the NAO index has been increasing since the early 1960s, and positive correlations have also been found between the winter NAO index and rainfall amounts in parts of the UK receiving considerable orographic precipitation [Osborn and Hulme, 2002], as does the western part of our study area. Similar NAO-rainfall links have also been found for northwest Britain [Wilby et al., 1997], northern and western Europe [Shorthouse and Arnell, 1999], and in neighbouring Ireland [Kiely, 1999].

[25] Although fewer low-flow changes were observed (mainly in the summer in the western half of the region) the trend for rising seasonal low flows may relate to the increasing prevalence of light and moderate summer rainfall events observed by Osborn et al. [2000]. It may also partly reflect the relatively dry period at the start of the time series around the mid-1970s.

[26] It is significant that such a degree of spatial variation exists in the several river flow trends quantified for a region of this size. Furthermore, although some coherent regional patterns of change have been identified, it is also clear that neighbouring basins, or even different gauging stations on the same river, can exhibit dissimilar river flow trends, and may vary in direction and strength. Our results show that it is crucially important to evaluate the representativeness of the sites before extrapolating regional river flow trends from a small sample of rivers or a low density of gauging stations. The sizeable sample of 56 gauging stations used here facilitated initial trend interpretation against previous precipitation and circulation studies at comparable scales. This has now provided a platform for further, basin-specific, studies to define the controls of this comprehensive suite of streamflow trends. Future work should focus on analyses of linkages between the quantified trends for the complete streamflow profile at annual and seasonal resolutions for several rivers in this important ‘leading-edge’ region of northwest Europe, and precipitation distributions, thermal inputs, evapotranspiration controls and atmospheric circulation indices. To conclude, the key trends identified here tend to assume an east-west pattern and include increasing high flows in winter in the higher-altitude basins to the west of the region: this is consistent with recent increases in winter rainfall here and a positive NAO index which is associated with higher orographic precipitation. Increasing summer low flows are also important, especially in the southwest, and some low-flow declines emerge for north and northeastern regions. The implications of these and other observed patterns in constraining predictions of future hydrological change, defining hydroclimatological relationships, understanding river system process and freshwater habitat impacts, and assisting water resources planning, and flood and drought management at national scales, are potentially considerable.

Acknowledgments

[27] This research was carried out as part of an Engineering and Physical Science Research Council (EPSRC) Industrial Case Studentship (Ref. 03302661) in association with Hydro-Logic Ltd and Advantage West Midlands. River flow data was kindly provided by the National River Flow Archive at the Centre for Ecology and Hydrology (CEH) Wallingford (http://www.ceh.ac.uk/data/nrfa/). The authors are very grateful for the support of Rod Hawnt, John Powell and Paul Webster at Hydro-Logic and Barry Hankin (formerly at Hydro-Logic; now JBA Consulting).

Ancillary