Catastrophic floods: magnitude and frequency investigations18 January 2005
Heavy rainfall in North Cornwall in August 2004 caused devastating floods in the town of Boscastle. Using field evidence and the historic flood record, Colin Clark describes how the event fits into the magnitude and frequency of floods at Boscastle, and the national picture of extreme floods and its implications for dam safety
On the afternoon of 16 August 2004 north Cornwall experienced a band of heavy rainfall lasting several hours. It was centered close to the Valency river catchment which discharges into the sea at Boscastle (Figure 1). The Valency rose by several metres very rapidly, causing widespread damage to property, the road bridge, and nearby gardens. Three properties were demolished, and cars washed away or badly damaged. Upstream of Boscastle the river channel was enlarged to more than twice its former width. The flood began soon after 1500 hrs BST, and was all over five hours later. Fortunately no lives were lost, although there were some lucky escapes. For example, for their own 'safety' several people climbed on top of the mobile building which houses the Visitor Centre. They were soon rescued by helicopter. Shortly afterwards one of the many trees washed down by the flood rammed the building, taking it out to sea.
What was the storm rainfall which caused this flood to occur? What is the likely return period of the flood? What is the history of flooding at Boscastle and how can this be reconciled with the flood of August 2004? What warning, if any, could have been given, and what action might have been taken to reduce the risk of damage to property and possible loss of life? What is the risk of similar floods taking place in the future? How does Boscastle 2004 fit into the national picture of extreme floods, especially in relation to estimates of the probable maximum flood and dam safety?
This paper seeks to provide answers to these questions. The results are broadly consistent with those from other historic floods which have taken place beyond the reach of living memory. As such the Boscastle flood has served as a test of the methodology applied by the author and co-authors to other situations both in England and overseas (Clark, 1997, 2003, Clark & Vetere Arellano, 2004; Rakhecha & Clark, 2002).
Evidence for the flood
Both photographs and eyewitness accounts paint an awesome picture of the flood. Figure 2 shows cars from the nearby car park being swept away by the rising floodwaters, while Figure 3 shows the water level even higher, where the building in the centre is just about to be demolished. The river came out of bank by 15.30 BST, reaching a peak at about 1700 BST and returning to bankfull by 2000 hrs BST. The estimated cost of repairs and loss to business through insurance claims is estimated to be about £4.5M [US$8.4M] (Barham, 2004) and the task of rebuilding property in a conservation area poses a major challenge. There was no evidence of a major damming back of flood debris and trees further upstream (Mason, 2004). Above Boscastle the channel was considerably widened (Figure 4), while a significant boulder dump at SX 102912 has been identified.
During the early stages of the storm, rainfall intensities were about 10-20mm hr-1 at Boscastle while at Lesnewth (SX 134900) the intensities were in excess of 40mm hr-1 as shown by the tipping bucket raingauge. Then came a period of lower intensity rainfall from 14.45 - 15.45 BST, followed by the main storm which had intensities in excess of 80mm hr-1 lasting for 0.25hr. From 15.45-16.15 BST 86mm were recorded at Lesnewth. Figure 5 shows the rainfall profile at 0.25hr intervals. These data have to be increased by 19% as a result of under-recording by the tipping bucket raingauge.
Estimates of peak flow
From the evidence of photographs, eyewitness reports and a field survey, the water level on the channel and valley cross section below Boscastle Bridge is shown in Figure 6. The National Trust store survived the flood but the nearby shop was destroyed. It is not clear if it was still standing at the time of the peak discharge, therefore two estimates of the likely peak flow were made. The slope area method was used to estimate the discharge:
Q = A R0.666 S0.5 n-1
Where Q = discharge m3 s-1; R = hydraulic radius; S = water surface slope m m-1; n = roughness factor; A = channel area m2.
The values of n which were chosen based on Barnes (1967) and Chow (1959) were 0.05 for the channel and 0.06 for the floodplain. The uncertainty of flood estimation due to the roughness coefficient has been described by Wohl (1998). They are comparable and even lower than those associated with other methods of estimating discharge indirectly. The results of the present investigation suggest a peak discharge of 337m3 s-1 without the shop in place and 301m3 s-1 with it still present. These estimates are associated with channel and floodplain flow velocities of 3.3m s-1 and 3.0/2.9m s-1 respectively.
Further evidence for the peak discharge comes from the boulder dump some 300m upstream of Boscastle (Figure 7). A sample of nine boulders in the channel were measured for their three axes. Normally the mean of these is taken as the boulder diameter (Costa, 1983) which is adequate when the boulders are well rounded, but for more angular specimens a correction may be necessary. Therefore a sample of cobble sized material was returned to the laboratory, the three axes measured and the volume and hence equivalent diameter determined. The results showed that for sub-rounded material the mean of three axes was adequate, while for more angular material a reduction in diameter of about 7% was necessary. Since the median boulder size was taken as representative of flow conditions, and the specimens were sub-rounded, this correction was not necessary. The median boulder diameter was 0.76m.
The use of boulders to estimate flood flows is well established, ever since Brahms first proposed the ‘sixth power law’ in 1753, which states that the weight of a rock particle is proportional to the 6th power of channel bottom velocity. More recent studies include those of Costa (1983); Williams, (1983); Carling & Grodek, (1994); Baker, (1989) and Clarke, (1996). The use of unit stream power, ? and critical stream power, ?c to estimate the flow needed to initiate boulder movement was first described by Bagnold (1980).
Unit stream power is defined: ? = ? Qs/w where ? = unit stream power kg m-1 s; ? = density of water kg m-3 ; Q = discharge m3 s-1s; s = water surface slope m m-1; w = channel width, m.
Using data from flumes and natural rivers Bagnold (1980) was able to define the critical stream power at which boulders of a given size will move:
?c = 290 D1.5 Log (12Y/D)
where ?c = critical stream power kg m-1 s; D = median diameter of boulders, m ; Y = mean depth of flow, m. These two equations are equal at incipient boulder movement. Figure 8 shows the cross section where these hydraulic calculations were made. The estimated depth of flow gives a peak discharge of 310m3 s-1sand a stream power of 325kg m-1 s-1, which is equal to the critical stream power needed to move a boulder 0.76m diameter. A sensitivity diagram was produced for a range of n values, boulder sizes and flood depths. Figure 9 shows the result where the estimated discharges are shown for a range of likely conditions. A result of about 310m3 s-1 agrees well with that obtained at Boscastle using the slope area method while the effect of an uncertain channel roughness and boulder size gives a range of peak discharges of 298-371m3 s-1swith a preference towards the lower end of this range. This can be compared with the probable maximum flood (PMF) as suggested by the Flood Estimation Handbook (IOH, 1999) using the 50% summer storm profile, as 273m3 s-1. Once again the deficiencies in the FEH are all too apparent.
Another sample of boulders was found on the floodplain just upstream of the channel boulder dump. The median diameter of these was 0.29m and it is instructive to apply the same methodology as was used for the in-channel boulders. The results show that a flow depth of about 1.9m would be needed to move boulders of this size. The presence of small hazel trees on the floodplain raises the roughness coefficient to 0.08. Fortunately, at this site there was enough trash caught in the trees to suggest a water depth of at least 1.85m. Further hydraulic calculations at this site are not possible because of the presence of fallen trees and other debris. However, the results show that the stream power in the channel and floodplain were high enough to lift the boulders out of the channel and over the floodplain. The flow velocities associated with the movement of the larger boulders are just over 5m . This compares with a velocity of 4.6m s-1based on Costa (1983) and a value of 5m s-1if the exponent in the equation V = 0.18 D0.487 (Costa, 1983) was 0.5, which is in exact conformity with the sixth power law. Although boulders have been excavated on floodplains in the past, they may have been deposited in the channel itself. The river channel subsequently migrated over the floodplain to a new location, leaving the in-channel boulders apparently deposited on the floodplain. This makes the Boscastle floodplain boulder dump indicative of very high flow velocities and of special interest in terms of sediment transport processes.
Water balance assessment for the flood
Conformity of the results of peak flow estimation using more than one method is not a guarantee of their accuracy. A more powerful and telling test is a water balance check on the volume of rainfall compared with the volume of runoff, and any losses due to the soil moisture deficit (SMD) and percolation:
Rainfall = Runoff + SMD + percolation
Rainfall and runoff are linked by the SMD, the percentage runoff, and soil hydraulic conductivity. These data are used in a non-linear flow model (Clark, 2004a) which is already in use by the Environment Agency SW Region. The model also requires the rainfall input, which was measured at Lesnewth and estimated by the radar station at Cobbacombe Cross, which is situated about 90km ENE of Boscastle. The radar data (Figure 10) were corrected using data gathered at Lesnewth, to give a weighted area rainfall for the Boscastle catchment of 177mm.
The MORECS (Meteorological Office Rainfall and Evaporation Calculation System) SMD for real land use in the local area for 10 August was 92.6mm. This is the depth of rainfall needed to bring the soil to field capacity, at which point runoff would be controlled by the saturated hydraulic conductivity (KSat) of the soil. Using the daily rainfall at Lesnewth and an estimate of actual evaporation from weighing lysimeters at Charldon Hill Research Station (CHRS), gives an antecedent SMD of 63mm. However, MORECS SMD has been shown to be in serious error (Clark, 2002a) with the estimates of SMD being in excess of 50mm of the measured values, especially in the summer. This finding is supported by means of water balance checks and field observations of the timing of runoff. Therefore the SMD for Lesnewth was recalculated using local rainfall and measured evaporation at CHRS from March 2004 onwards. Generally, temperatures at CHRS are higher than in the Lesnewth area, but the rainfall from March to August was about the same at both locations. The result gave an SMD of zero on the morning of 16 August. As a comparison, on 10 August MORECS SMD at CHRS was 110.6 whereas the measured value was 37mm. From 10-15 August at Lesnewth 40mm rainfall were recorded.
The condition of field capacity is not likely to be applicable to the whole catchment because of the strong environmental gradients in the area. At an altitude of 210m, Lesnewth has a higher rainfall than the lower part of the Valency catchment. The SMD of the lower area was assessed using the relationship between rainfall data gathered at Bossiney, 4km SW of Boscastle and at an elevation of 88m. The strong relationship between the rainfall from April-July for previous years at these two sites allowed the SMD to be estimated for the lower catchment as 39mm. This probably covered an area of about 6km2. The remaining 13.8km2 will be expected to produce both slope runoff and KSat runoff (Clark, 2004a), while the lower catchment would only produce slope runoff until the SMD is reduced to zero. Thus the weighted area SMD for the morning of 16 August was 12mm.
The relationship between rainfall intensity and percentage runoff was assessed by a field survey of the soils within the catchment. The saturated hydraulic conductivity was measured both in the field and in the laboratory using the core method (Hollis & Woods, 1989) for the two main land uses: grassland and woodland which cover 95% and 5% of the area respectively. Other land uses such as roads and buildings are insignificant. Figure 11 shows the results where the woodland soils are less permeable than other woodland soils sampled elsewhere. The grassland soils appear very impermeable. By applying rainfall at varying intensities to the soils data the percentage runoff can be predicted and shown on the same diagram (Figure 11).
The corrected rainfall profile at Lesnewth was then applied to the non-linear unit hydrograph model and the model optimised to give a peak discharge of 300m3 ss-1 (Figure 12). The timing of the flood is comparable with events on 16 August, with the river rising about 15-30 minutes too early. The depth of runoff is 131mm. The percentage runoff can be estimated:
% runoff = Depth of runoff/Rainfall = 74%
This result compares with 79% based on the SMD and the relationship in Figure 11. Thus the water balance for the flood can be written:
Rainfall (171) = runoff (131) + SMD (12) + percolation (28)
where the units are mm. The implied rainfall depth of 171mm is slightly less than the measured depth of 177mm. This result also suggests that the depth of runoff is too low by 6mm, which is well within the level of uncertainty in the other three variables. It is important to note here that the SMD and percolation are based on measurements made in the catchment, the unknown variable being the depth of runoff. A much higher estimate of the peak discharge is unwarrented by the accounts of the flood, antecedent SMD and the soils data, while a lower peak discharge, such as 200m3 s-1 would substantially reduce the percentage runoff to 59%. This represents a loss of 0.5Mm3 or 27mm rainfall and could not be sustained in view of the soil hydraulic conductivity and antecedent SMD.
Historic floods at Boscastle
In the absence of a long term measured flow record, placing an extreme flood in its proper context means looking at the evidence for historic floods. Table 1 gives a summary of the floods for which documentary evidence is available. Some writers (Doe, 2004) appear to have used uncritically the information on the British Hydrological Society website (www.dundee.ac.uk/geography/cbhe) which mentions flooding at Boscastle in 1847 and 1957.
Both dates do not apply to floods at Boscastle, with the earlier event affecting the rivers Inney, Allen, and Camel (Cornwall Royal Gazette 16/7/1847). Indeed, during the following week Lady Launceston and the Rev Gibbons were reported to have set up their ‘summer residence at this delightfully romantic watering place’. (Cornwall Royal Gazette, 23/7/1847). The flood of 1957 took place in 1958.
Table 2 shows the estimates of the peak discharge of four historic floods for which enough evidence exists. Boscastle Bridge was rebuilt following the 1958 flood and the author was kindly provided with the plans of the old bridge by the Cornwall County Council which showed that the bridge consisted of two arches, one large and the other small. Head loss equations were applied to the bridge in order to help determine the peak flow of the 1958 flood.
The historic record starts and ends with a flood. It is known that a flood took place in the area in about 1854 (Launceston Weekly News 17/11/1894) but no evidence for this event has been found. Therefore the time period was extended to half way from 1854 to 1894, making the record 130 years long. The return period of the 2004 event is clearly far greater than 186 years. Its likely return period can be better estimated when a flood frequency curve for the Valency river can be established.
Flood frequency analysis
A robust estimate of flood frequency can be obtained by using the historic flood record as described above and an estimate of bankfull discharge, which has a return period in the range 1.5-2 years, and an estimate of the probable maximum flood using the non-linear flow model and revised estimates of PMP (Clark, 1995, 2002b). The estimate of bankfull discharge was based on the pre-flood channel which since 1907 has remained stable as shown by Ordnance Survey maps throughout the 20th century.
Bankfull discharge was estimated as 18m3 s-1sand the PMF 516m3 s. The frequency distribution on which these results are plotted was first used in meteorology (Rakhecha & Clark, 1999) and hydrology (Rakhecha & Clark, 2002). More recently, Koutsoyiannis (2004) has proposed a similar frequency distribution for analysing rainfall records.
The rarity of the PMF, in line with risk assessment at dam sites (Hughes et al, 2000) is taken as 106 years. Figure 14 shows the flood frequency curve, where the data for the historic floods, bankfull discharge, and PMF are drawn as circles in recognition of the uncertainty of the estimates of both discharge and return period, (Clark 2004b). As a comparison, the FEH 100 year flood using the 50% summer profile, has been plotted on Figure 14, showing how it has severely underestimated the flood hazard at this site.
The flood frequency curve suggests that the flood of 2004 had a return period around 10000 years. This does not mean that a similar flood will not occur during this century. The depth of water on the floodplain in the 2004 flood was about 1.7m. A flood depth of 0.8m would have caused considerable damage, including washing away cars and some damage to property.
Such an event has a return period of about 400 years, while an increase of flood depth to 1m might occur once in 1000 years. In terms of the impact on people, flood depth is far easier to understand than flood return periods. More importantly, it is the risk of an event taking place during, say, the next 50 years which is more critical in helping authorities decide on the best course of action. This problem must now be considered.
Risk analysis of floods at Boscastle
The 1958 flood was the worst event in the 20th century and has been estimated to have a return period of 76 years (Table 2). The chances of this event taking place in the next 50 years can be assessed by using the risk equation:
p = 1 - (1-1/T)n
Where p = probability of occurrence; T = return period (years); n = period of interest (years).
Application of this formula shows that the 1958 event has a 49% chance of taking place in the next 50 years, while the 400 year event has a 12% chance of occurrence. This places the concept of return period into a different perspective. Furthermore, analysis of historic floods in NE Spain (Barriendos et al 1998) shows that they do not occur at regular intervals: rivers have both quiet and busy times. This phenomenon can be explored by means of Monte Carlo simulation. A perfect 100,000 year record was generated. Sampling from this record was performed for ‘15000 years’ recording the return periods in each year. The results were divided into 50, 100, and 200 year non-overlapping periods. The frequency of events with return periods 100-499, 500-999, 1000-9999, and +10000 years for each time period are shown in Table 3.
Two of the 150 100 year time periods had four events with a return period of 100-499 years while 28 of the same sample had at least one event whose return period was in excess of 1000 years. Similar observations could be made about the 300, 50 year time periods. The results are a sobering reminder that the 1 in 500 year event may occur more than once in a 50 year time period. If this pattern were to be observed over a large area which included many different drainage basins then we might conclude that either our return period assessment was at fault or that a change in environmental conditions was taking place - or both.
For many people the philosophy behind estimates of event rarity are hard to understand. During the next 50 years the 1000 year or greater event has a 1 in 20 or 5% chance of taking place as shown by the risk equation. On the other hand the Monte Carlo analysis in table 3 suggests a 1 in 11 or 9% chance of occurence (p = 21 + 2 x 2 + 1 + 1 x 2)/300. This equates to a flood depth of 1m on the floodplain at Boscastle. Thus the risk equation may not always be a reliable guide for planning future flood mitigation measures. Perhaps a better approach is to consider the chances of a certain depth of flooding taking place. This stands a better chance of being understood by the public and may avoid the criticism of the assessment made after a flood which is said to be ‘so rare that it cannot happen again’. A similar event, though much less rare, can, and probably will happen in the next few decades at Boscastle. If it is greater than the 1958 flood but less than the 2004 flood then considerable damage will be done. This leads to the question of future flood warning measures which could be implemented.
Flood warning measures
Flood warning systems usually consist of: 1) assessment of the hydrometeorological situation; 2) flow modelling; 3) issue of a warning; 4) responses to the warning.
At present the radar data are not reliable enough and cannot be processed in time to provide a reliable flood warning at Boscastle. Small, steep sided catchments with a rapid response time mean that accurate rainfall data must be collected in real time. The tipping bucket raingauge at Lesnewth is not as yet telemetered. This deficiency should be regarded as a priority together with the installation of a second telemetered gauge in the catchment which would give more detail about the storm event and also serve as a back-up in case one or the other should fail.
Using the non-linear flow model and telemetered rainfall data from Lesnewth on the 16 August 2004, a flood warning could have been given by 1400 BST, giving about 1.5 hours warning. In order to provide a reliable estimate of the SMD, a weighing lysimeter should be installed at or near Lesnewth. The antecedent SMD for the 2004 event was provided from data from the weighing lysimeter at CHRS, which may not be reliable under all conditions.
Once a decision has been made regarding the likely flood flows then a warning system must be activated. This can take the form of an automatic voice message to consenting householders and businesses, messages over the radio, television, or Environment Agency Website. Arrangements for those not in residence have to be the responsibility of the householder. A Flood Warden could also be considered.
According to Handmer et al (2001) the warning message should be: timely and reliable; have local and individual meaning; be forward looking; suggest suitable responses; be reinforced socially; be sent to those at risk
In the event of a marginal risky situation occurring, an all clear message should be given. The purpose of a flood warning system is to save lives. It should also relieve the stress of those who live in a flood prone area. That means the warning must be accurate. Given 1.5 hours warning on 16 August, residents could have had time to remove valuables upstairs, move any vehicles to high ground, fit flood boards to doors and windows; and help elderly and disabled neighbours. Visitors would also have had time to move vehicles and phone friends and relatives.
Implications for dam safety
There is still a need to reassess dam safety especially in the light of new evidence from extreme floods. The Boscastle event showed that a peak discharge of about 300m3 s-1scan occur in a catchment area of about 20km2. This is nearly the rate of runoff equal to the catastrophic flood as defined by Allard, Glasspole, & Wolf (1960). When compared with the likely discharge of the 1893 flood on the Cheviot Hills (Clark, in press), the August 1970 flood in Morayshire (Acreman, 1989) and the 1771 flood on the Tyne (Archer, 1993), it is clear that the catastrophic flood has been exceeded over a wide range of catchment areas. The high rates of runoff associated with these floods should be compared with the existing PMF at dam sites in order to judge the integrity of the spillway.
There is a growing interest in historic floods as an aid to estimating extreme floods (Archer, 1999, Williams & Archer 2002, Acreman 1989). Baker (2003) makes the point that palaeoflood data, and by implication historic data – when carefully collected, are just as valid as gauging station records of extreme floods because the latter are only based on an extrapolation of out of bank conditions, and, in some cases, made following the destruction of the measuring equipment. In addition, when placed in the context of the historic time scale as compared with the duration of gauging station records, the historic records allow a more robust flood frequency curve to be
produced. If the frequency curve can be simulated by suitable flow modelling then any changes in future land use can be accommodated and future flood frequency predicted in respect of any land use changes.
If the Boscastle event had taken place 100 years ago, and assuming that no other flood of a similar size had occurred since, what would we have made of the evidence? Providing that the contemporary newspaper report gave a good indication of the depth and timing of flooding, by using the methodology as described in this paper a good estimate of both peak discharge, rainfall depth, and flood rarity could have been produced. Even without a suitably detailed newspaper report the boulder evidence, soils data and a realistic estimate of the SMD would have given similar results.
The Boscastle flood of 2004 has generated much interest in extreme events. However, both in America and to some extent in England, as the research into extreme events has increased, so has the cost of flood damage (Baker, 2003). This paradox may be explained both by the failure of hydrologists to correctly assess the flood hazard – estimates seem to get bigger as time goes by – and by the failure to estimate the geographical scale of flooding. Many houses were probably built in a quiet time so far as rivers were concerned. A failure to assess historic floods as compared with just listing them must also be addressed. The Flood Estimation Handbook (IOH, 1999) also has a case to answer. It has encouraged rapid assessments of floods at the click of a switch without as much as a visit to the site itself or to the local Record Office to investigate floods further back in time.
When called upon to make sense of an extreme event it fails. Questions raised in the past regarding some of the methodology in the FEH remain unanswered. Although the advice to undertake a historical review of floods is given in the FEH, its represents less than 1% of its entire contents. No doubt historic flood analysis would take longer than the FEH methodology, but the result is often much more reliable.
If the story of King Canute commanding the waves to retreat is true, then it appears that his bad example has been followed by both scientists and public alike. However much we think we know about a particular river it may surprise us yet again. Had we looked at all the evidence of how rivers or the sea behaves, then we might not fall foul of the elements as did our erstwhile King all those years ago.
Colin Clark, Charldon Hill Research Station, Shute Lane, Bruton, Somerset, England BA10 0BJ
The author would like to thank Tim Wood of Cornwall County Council, Graham Bartlett of the National Meteorological Library, Exeter, Ian Barker, Environment Agency, Frank Law for encouragement and advice on sources of historical flood information, Cornwall Fire Brigade, the Cornwall Local Studies Library, Redruth, the British Library, and residents of Boscastle. The non-linear flow model web site was set up by Peter Clark.
|Table 1: Summary of the historic floods at Boscastle|
2004 16 August
1958 3 June
1950 30 August
1926 18 July
1903 27 October
1894 12 November