Technical Appendix 6: Recent changes in flood frequency and magnitude in Welsh river catchments
Technical Appendix 6
Recent changes in flood frequency and magnitude in Welsh river catchments
14
Technical Appendix 6: Recent changes in flood frequency and magnitude in Welsh river catchments
Contents
Page
List of Figures ii
List of Tables ii
1. Introduction 1
2. Study catchments, data and methods 3
3. Variations in flood magnitude 6
4. Variations in flood frequency 11
5. Flood frequency analysis 13
6. Conclusions 18
List of Figures
Page
Figure 1: Locations of study catchments and flow gauging stations 4
Figure 2: Annual maximum flood series for the four gauging stations 7
Figure 3: POT events exceeding thresholds giving 4.5, 2 and 0.2
events per year 12
Figure 4: Flood frequency analysis using partitioned flow records:
A. Dee at Manley Hall; B. Severn at Abermule,
C. Teifi at Glanteifi; D. Dyfi at Dyfi Bridge 16
List of Tables
Page
Table 1: Comparison of recent trends in flood frequency and
magnitude in selected British rivers 2
Table 2: Characteristics of the four drainage basins 5
Table 3: Inter-annual variability in flood magnitude, estimated
using standard deviation 9
Table 4: Statistical significance of the selected breakpoints
in each record 15
Table 5: Estimated flood magnitudes from frequency analysis
on partitioned records 17
14
Technical Appendix 6: Recent changes in flood frequency and magnitude in Welsh river catchments
Recent changes in flood frequency and magnitude in Welsh river catchments
1. Introduction
Temporal variations, or non-stationarity, in flood frequency and magnitude have been detected in both gauged flow records and longer reconstructed flood histories from many locations (e.g. Booy and Morgan, 1985; Rumsby and Macklin, 1994; Black, 1996; Longfield and Macklin, 1999; Benito et al., 2003; Mudelsee et al., 2003; Sheffer et al., 2003). Such non-stationarity in flood regime violates one of the principal assumptions (the stationarity of the mean) underpinning conventional flood frequency analysis (Beven, 1993; Franks, 2002). The results of a flood frequency analysis depend not only on the length of record available but also on which particular years the flood record encompasses (Kiem et al., 2003). This will be most significant in cases where either trends or step-changes in flood frequency and magnitude are present within the available record (Reed, 1999). The recent occurrence of several major flood events in Welsh rivers (e.g. 1998, 2000, 2002) suggests that there may have been a sufficient change in recent flood regime to affect the results of flood frequency analysis. This could be manifest in two ways: first, the return period for a flood of a given magnitude could change, and second, the magnitude of a flood of a given return period could increase or decrease.
Changes in flood frequency and magnitude have been detected in gauged flow records from western Scottish rivers since 1988, including the occurrence of several maximum recorded floods and increases in flood frequency (Black, 1996). Peaks-over-threshold (POT) frequencies attained maximum recorded values in western Scottish catchments during the 1980s and 1990s (Black and Burns, 2002). In England, the Easter 1998 flood event produced maximum recorded peak flows in many catchments in Warwickshire, Northamptonshire and Oxfordshire (Horner and Walsh, 2000). In the Tyne, in northern England, flood frequency was low between 1970 and 1979 and increased from 1980, with a sequence of major floods after 1986 (Rumsby and Macklin, 1994). Similar results were obtained from the Ouse at York where high flood frequencies and magnitudes were recorded in 1944-1968 and 1978-1996 separated by a period of extremely low flood frequency and magnitude during 1969-1977 (Longfield and Macklin, 1999). Disastrous flooding also occurred in the 1980s and 1990s in south and west Ireland (Kiely, 1999).
An analysis of national changes in UK flood regimes revealed no continuous trends during the period 1940-1990 although systematic fluctuations, similar to those in precipitation values, were present (Robson et al., 1998). Subsequent work confirmed the absence of a long-term trend over the last 80-120 years but found evidence of an increase in protracted high flows during the last 30-50 years (Robson, 2002). This was confirmed by Marsh and Dale (2002) who reported an increased occurrence of high long-duration flows in five large catchments in England and Wales from 1990. This reflects the increase in total winter precipitation between 1961 and 2000 reported by Osborn and Hulme (2002). There was also an increase in the intensity of winter precipitation between 1961 and 1995 (Osborn et al., 2000).
In Welsh rivers changes in flood frequency and magnitude were first reported by Howe et al. (1966; 1967), who found a substantial increase in flood heights at Welsh Bridge, Shrewsbury, on the Severn during 1940-1964 compared with the period 1911-1940 (Table 1). A similar pattern was found on the Wye (Howe et al., 1966), although Walsh et al. (1982) suggested that the increase there occurred in the late 1920s. A later analysis of flooding in the Severn found that the highest frequency of floods occurred in the late 1940s and early to mid 1960s, followed by a decline after 1968 and an increase in the late 1970s and early 1980s (Higgs, 1987). On the Tawe, in south Wales, a much greater frequency of high-magnitude floods was found during 1929-1981 than in the period 1875-1928, with distinct clusters of major floods occurring in 1929-1933, 1957-1967 and 1979-1981 (Walsh et al., 1982). Similar results were reported from the Ebbw. Black (1996) found no maximum recorded floods in the Teifi, upper Severn and Dee to parallel those which occurred in western Scottish catchments after 1988. However, recent work in the Trannon subcatchment of the Severn found an increase in flood magnitude and frequency since 1988 (Mount et al., 2005).
Table 1: Comparison of recent trends in flood frequency and magnitude in selected British rivers
River / Period of record / Date of change / Increase or decrease in flood frequency and magnitude / ReferenceWales
Severn / 1911-1964 / 1940 / Increase / Howe et al. (1966; 1967)
1953-1983 / 1968
1978 / Decrease
Increase / Higgs (1987)
Wye / 1911-1964 / 1940 / Increase / Howe et al. (1966)
Tawe / 1875-1981 / 1929 / Increase / Walsh et al. (1982)
Ebbw / 1908-1978 / 1929 / Increase / Walsh et al. (1982)
Trannon / 1969-2002 / 1988 / Increase / Mount et al. (2005)
England
Tyne / 1700-1992
(1875-1992
shown here) / 1875
1895
1920
1940
1955
1970
1980 / Increase
Decrease
Increase
Decrease
Increase
Decrease
Increase / Rumsby and Macklin (1994)
Ouse / 1878-1996 / 1904
1944
1969
1978 / Increase FM
Increase FF
Decrease
Increase / Longfield and Macklin (1999)
Notes: FM = flood magnitude; FF = flood frequency
The effects of natural climatic variation on the results of flood frequency (return period) analysis were first demonstrated by Howe et al. (1966; 1967) for Welsh Bridge on the upper Severn. Howe et al. (1966; 1967) split the flood record into two parts (pre-1940 & post-1940) and plotted flood height against recurrence interval. The gradient of the fitted frequency curves, and their position relative to the flood height axis, on the graph increased from the 1911-1940 period to 1940-1964. Higgs (1987) found that there was a significant decrease in the gradient of curves fitted to return period analyses on records from Caersws and Montford on the upper Severn which were divided at 1968/1969. Comparable analyses from other British rivers are lacking. However, Dunne and Leopold (1978) performed flood frequency analysis on a 36 year flood record (1934-1970) from the Tana River, Kenya, partitioned at 1961. The resulting frequency curves had markedly different gradients. Booy and Morgan (1985) divided a long flood record from the Red River at Winnipeg, Canada, into three equal parts and obtained a similar result. More recently Franks (2002) has shown the effect of a marked change in climate on flood frequency curves using flow data from New South Wales while Kiem et al. (2003) examined the effect of climatic oscillations (ENSO, IPO) on flood frequency curves using the same data. Jain and Lall (2000) found systematic variations in the magnitude of the “100 year flood” when estimated using a 30 year moving window on the 82 year flood record from the Blacksmith Fork River.
The aim of this Technical Appendix is to (1) use gauged flow records to determine whether there has been a recent change in the frequency and magnitude of flooding in Welsh rivers, and (2) investigate the potential effects that any change might have on the results of flood frequency analysis. First, the study catchments, data and methods used are outlined. Second, the annual maximum flood records are examined for changes in flood magnitude. Third, the peaks-over-threshold records are analysed for changes in flood frequency. Finally, the records are partitioned to gauge the effect of the observed changes on the resulting flood frequency curves.
This work was undertaken as part of the Predictive and Investigative Modelling of Flood Risk within Welsh River Catchments, a multi-partner project funded by the Welsh Assembly Government, Environment Agency Wales, Countryside Council for Wales, British Geological Survey and University of Wales, Aberystwyth. The data used were provided by the Environment Agency.
2. Study catchments, data and methods
Four Welsh rivers, the Dee, Dyfi, upper Severn and Teifi, were selected for this study, of which two drain to the west and two to the east of the Cambrian Mountains (Figure 1). These rivers have relatively large catchments (> 400 km2), long-standing flooding problems, particularly in their lower reaches and comparatively long flow records. Very few Welsh rivers were gauged prior to
Figure 1: Locations of study catchments and flow gauging stations.
1960, when a major expansion of the gauging network began. Most records are therefore around 30 to 40 years in length (Table 2). The catchments are similar in having relatively high relief and high mean annual rainfall. Rainfall is lower in eastward draining catchments, as would be expected. The large difference in mean annual rainfall between the Dyfi and Teifi, which both drain west, is due to the difference in relief and catchment configuration. Mean and maximum daily flow data were available for the four gauges. A difference of opinion exists over the suitability of mean daily flow data for analysis of flood frequency and magnitude. For example, Robson and Reed (1999) regard daily mean flow data as unsuitable for flood frequency analysis due to the short response times of many UK catchments while Svensson and Jones (2004) consider daily mean flows to be indicative of the magnitude of the maximum daily flow. For two of the gauges, Abermule and Glanteifi, the
14
Technical Appendix 6: Recent changes in flood frequency and magnitude in Welsh river catchments
Catchment characteristics (above gauging station) / Mean annual rainfall (mm) / 1369 / 1834 / 1259 / 1382Relief (m) / 857 / 901 / 657 / 588
Basin area
( km2) / 1019 / 471 / 580 / 894
Mean flow (m3s-1) / 31.15 / 23.2 / 14.64 / 28.82
Altitude of gauging station (m O.D.) / 25.4 / 5.9 / 83.0 / 5.2
Period of record / 1970-2003 / 1979-2003 / 1962-2003 / 1959-2003
Record type / Maximum daily flow / Maximum daily flow / Mean daily flow / Mean daily flow
Gauging station / Manley Hall / Dyfi Bridge / Abermule / Glanteifi
Catchment / Dee / Dyfi / Severn / Teifi
maximum daily flow records began in 1990. The mean daily flow records began in 1962 and 1959, respectively and these were used in all analyses. Since these gauges record runoff from a relatively large catchment area daily mean flows approach maximum daily flow values. Greater differences may occur at very high flows due to uncertainty in the upper part of the rating curves. Some caution should therefore be exercised when interpreting these results. For Dyfi Bridge maximum daily flow data were available from 1979. The mean daily flow record from this site began in 1962 but four years of data were missing during the early 1970s. For this reason the maximum daily flow record was chosen for analysis. For Manley Hall on the River Dee mean and maximum daily flow data were available from 1970. Mean daily flow data from 1937-1968 were available from the Erbistock gauge, the predecessor of the current Manley Hall gauging station. The two stations are less than 1 km apart. However, preliminary analysis showed the Erbistock data to be of significantly lower quality then the Manley Hall data. These data were therefore excluded from the analysis and the maximum daily flow data from Manley Hall were used.
Analysis was based on water years (starting October 1st) since winter flooding dominates the regime in these largely un-urbanised catchments. Water years were labelled using the calendar year in which they begin, following Robson and Reed (1999). Part-years of data at the beginning and end of the record were excluded. All records end on 30th September 2004. Annual maximum and peaks-over-threshold series were extracted from the record for each gauge. Independence of peaks was established using the methods outlined by Robson and Reed (1999). Annual frequencies were derived from the peaks-over-threshold series based on three thresholds set to give 4.5, 2 and 0.2 floods per year over the period 1979-2003.
3. Variations in flood magnitude
The annual maximum flood series have been plotted for each of the four gauging stations and the pattern of variation smoothed using a 5-year moving average (Figure 2). Temporal variations in the magnitude of the annual maximum flood are apparent at all four stations. The four types of temporal behaviour associated with climatic variability recognised by Reed (1999) can all be identified in the annual maximum series (trend, step change, periodicity, and quasi-random behaviour). However, long-term trends, persisting through the entire period of record, are notably absent at these sites. In the latter part of the Abermule record (after 1987) a trend of increasing flood magnitude is apparent, but this appears to be the only example in the four records. It coincides with a period of high inter-annual variability where there is no apparent trend in the lower-magnitude floods. Therefore, the apparent trend is primarily the result of the increasing frequency of high-magnitude flood events. Step changes in flood magnitude are more common. The abrupt increase in flood magnitude at Manley Hall in the period 1998 to 2000 and the sudden decrease in flood magnitudes from 1967 to 1968 at Abermule are particularly pronounced examples of this type of behaviour. There is some