Estimating extreme floods6 July 2004
A more thorough search of historic data could help provide reliable estimates of extreme floods. Colin Clark examines the case of the Great Till Flood of 1841
RELIABLE estimates of extreme floods are very hard to obtain. Dam engineers need such data, especially for the design of dam spillways which have to safely pass the probable maximum flood where the dam is positioned upstream of where people live. Traditionally empirical relationships were used to provide this data. More recently in the UK the Flood Studies Report (NERC, 1975) and its successor the Flood Estimation Handbook (IOH, 1999) advocate the use of probable maximum precipitation (PMP) and the rainfall runoff method. More recently historic flood data have been investigated and the results cast serious doubt on the results of these more recent methods, almost always giving much higher estimates of extreme flood events. To date there has not been enough emphasis of the value of historic flood events to produce: 1. Estimates of very rare floods: eg Clark & Vetere Arellano (2004) Clark (in press, a); 2. Flood frequency curves: eg Acreman and Horrocks (1990); Clark, (1999); 3. A longer flood record which is combined with riverflow data: eg Clark (2003a); 4. Estimates of extreme rainfall (Clark, in press b), which can also be maximised to give new estimates of PMP: eg Clark (2003b).
The shelf life of historic flood data is considerable. This is because they can be revisited in the light of new knowledge and understanding of floods, together with techniques of analysis which were not in widespread use at the time when the event was first documented (Clark in press, b). New information may also come to light as a result of a more thorough search of the historic record. This became apparent when the archives relating to the Great Till Flood (GTF) of 1841 were examined. This flood has already been described (Anon, 1841, Cross, 1967) where the latter account is widely quoted (Newson, 1975; Rodda, Downing, & Law, 1976; Bradford & Faulkner, 1997). This paper seeks to look at both existing and new evidence relating to the GTF of 1841 on Salisbury Plain, England: estimates of both the peak discharge and storm producing rainfall are presented. An assessment of the antecedent conditions of the flood has also been made: this is important for flood flow modelling, keeping in mind the possible effects of future climate change on the flood hazard during wintertime. The flood is placed in the context of flood events in 1789 and 1915, and finally a flood frequency curve for the Till is produced.
Accepted view of the Great Till Flood
The GTF is probably the most serious event in documented history on the river Till. Cross (1967) gives details of the events leading up to and including the flood: the autumn of 1840 was wet and in December temperatures were below zero at Salisbury 21km to the south-south-east of Tilshead, for over two weeks. The temperature was reported to have been sub-zero from 4-11 January, with temperatures as low as -11ËšC on 8 January. Heavy snow fell the day after and then more snow on the 13th; a rapid thaw and heavy rain on the 16th on a frozen sub-soil led to the GTF. At Shrewton 36 houses were destroyed, three people were drowned, and about 200 made homeless in the area (Cross, 1967, Salisbury Journal 25/1/1841). In the following year a number of Flood Cottages were erected from the money collected towards the Charity Fund. A commemorative plaque is fixed to the cottages at Shrewton and Tilshead (Figure 1). No estimates of rainfall, peak flow, or flood rarity were given.
The river Till drains an area of 41.3km2 at Tilshead on the northern part of Salisbury Plain (Figure 2). The valley is now almost entirely dry and orientated north-south leading to the river Wylye at Stapleford. The underlying rock consists of middle and upper Chalk; the former is exposed in the valley above Tilshead. Slopes are moderate to steep and general height is about 140m with the highest point 195m at Gibbet Knoll. Soils are largely Andover Association (Findlay et al., 1984) on valley side slopes and bottoms, with the Icknield Association on the western and eastern parts of the drainage basin.
Some 10km downstream is the village of Shrewton where the catchment area is 72.9km2 and the soils are very similar to those above Tilshead. In the late 18th century the land use was chalk grassland grazed by sheep on the downs with some arable land in the valley bottoms (Davis, 1794). By the time William Cobbett visited the area in the 1820's (Cobbett, 1826) this picture had hardly changed. Fifteen years later the Tithe surveys for Shrewton and Tilshead (Wiltshire Record Office, 1840) give little additional detail. At present the land use in much of the catchment has changed very little.
Meteorological situation, January 1841
No systematic local measurements made during January 1841 have been found. The nearest such data were gathered at Hadspen House 40km to the west-south-west at an elevation of about 120m. Here data for air pressure and temperature were recorded at 9am and 2pm (Hobhouse, 1841). While Britton (1927) stated that the temperature data were of doubtful value, when the mean temperatures are compared with the daily Central England Temperature record (Parker, Legg, & Folland, 1992), a good comparison is apparent (Figure 3) with temperatures at Hadspen about 2°C warmer during the first six days, warmer still from 7-9 January and then very similar thereafter. To help judge whether these data are realistic in the context of antecedent flood conditions, temperatures at Boscombe Down 16km south-east of Tilshead were compared with those at Cambridge Botanic Gardens for January 1963. The result, not shown here, indicates that differences in excess of 4°C between the two sites can occur in cold conditions while temperatures are more comparable around 0°C. The site at Cambridge was chosen because it was one of the sites used to construct some of the CET record (Jones, 1987).
The pressure record made at Hadspen is comparable to that at London (Figure 4) although the numerical values are not important, the pattern of changes show the existence of an area of low pressure around 4 January, its replacement by a high during the next five days, which was then followed by relatively low pressure thereafter. Associated with these changes was heavy snow recorded at Hadspen and Salisbury on 9 and 13 January, while Charlotte Downes who lived at Donhead St Andrew 25km south-south-west of Tilshead at an elevation of about 110m, noted that it snowed a great deal on 14 January, (Downes, 1841). The depth of snow over the Till catchment is unknown, but the nearest account comes from the diary of J.H.Chandler of Stockton, 11km south-south-west of Tilshead - which was copied by Thomas Baker into his unpublished 2nd Edition of Record of the Seasons (Baker undated). At Stockton the depth of snow was given as four inches (100mm).
The high pressure from 6-9 January gave rise to a cold spell with frost for much of the day. By 10 January Charlotte Downes noted both freezing and thawing. On 13 January a thaw had set in at Salisbury and by 15 January a gentle thaw had set in at Hadspen and Donhead St Andrew. Only Charlotte Downes noted heavy rainfall in her Journal on 16 January (Downes, 1841), although rain was reported at Hadspen and over Salisbury Plain.
Flood of Saturday 16 January
Useful data comes from the Devizes and Wiltshire Gazette (21/1/1841) and the Wiltshire Independent (21/1/1841). The first report tells us that at Tilshead: ‘The waters from Can Down began to flow into this village about the middle of the day on Saturday, from the north, but although they soon rose to the height of two feet in that part of Tilshead, the upper part of the parish viz - the west end, was almost entirely free, and no danger was apprehended. This state of things continued until half-past five, when the flood came down the whole width of the street at the rate of at least ten miles an hour, and about three feet deep, rising very rapidy by until about half past, when it began gradually to decline.’
The Wiltshire Independent takes the story further: ‘At Tilshead six houses were completely overthrown, but fortunately no lives were lost, the inmates in every instance escaped, although with the greatest of difficulty, some of them having been taken out of their chamber windows and removed to places of safety. Besides the dwelling houses which were destroyed, several outhouses and sheds, and many walls were washed down. In places the water was from 10 to 15 feet deep in the street. In many places the road is covered to a depth of full two feet with the broken stones which were loosened and washed down by the current during its course.’
Downstream at Maddington, which is contiguous with Shrewton, the same newspaper stated: ‘The rise and fall of the water were remarkably sudden, not much above 12 hours having elapsed between the commencement and end of the flood; it attained its greatest height from 8 to 10 o'clock at night, and at 3 o'clock in the morning the road was dry.’
At Tilshead the estimate of peak discharge is based on the distance between houses at the east end of the village and a flow depth of 3.04m, the lower of the two estimates of the flood waters. A higher depth of water would have led to reports of water reaching bedroom level as suggested during a site visit. The water surface slope was taken as the land surface slope in the village surveyed in the field, while a rough unmade road would have a roughness coefficient of 0.04. These data were then applied in the Manning equation:
Q = AR0.666 S0.5 n-1
Where Q = discharge (m3 s-1); A = channel area (m2); R = hydraulic radius, A/P where P = wetted perimeter of the channel (m); S = slope m m-1; n = Mannings roughness coefficient. In this case A = 26.6m2, P = 14.83, R = 1.793, n = 0.04, s = 0.0017.
The result was a peak discharge of 41m3 s-1. This would have an average velocity of just over 1.5m s-1 with the highest velocity being around 2.5m s-1. This compares with an estimated speed of 10 miles an hour or 4.5m s-1. The higher velocity would be the highest visible to the naked eye, not an average velocity.
Figure 6 shows the cross section during the flood. An increase in flood depth of 0.1m was made on account of the raising of the road level since 1881 as shown by a field survey. Again, using local values of slope and roughness gave a peak discharge of 48m3 s-1. An estimated hydrograph at Tilshead is shown in Figure 7. The rise of the river in the afternoon was due to snow melt and light rain, whereas the main flood was a result of heavy rainfall and more rapid snowmelt in the afternoon.
Role of frozen ground
It is clear that there was not a period of continuous frost during the first part of January (figure 3). Cross (1967) made the contrary conclusion based on a temperature record published in the Salisbury Journal, which he believed to be the only record available. On 6, 7, 8, and 12 January the readings at Salisbury were made early in the morning and late at night (not representative of the day as a whole). The reading of 3°F (2.2°C) should have been accompanied by a thaw as stated in the newspaper report but this was omitted by Cross (1967). For 14 and 15 January only one reading at 0800 hours was published. Another temperature reading at Salisbury during the afternoon of 15 January would have suggested a thaw in progress since at Hadspen the temperature was 2.2°C. An alternative way of assessing the presence of frozen ground is to estimate soil temperature from the equations of net radiation and heat transfer:-
Where: Qn = net radiation W m-2; a = albedo; Qs = incoming radiation W m-2; Qld = downward longwave radiation W m-2; Qlu = outgoing longwave radiation W m-2;
Qs = Qa (a + b n/N)
Where Qa = extra terrestrial radiation W m-2; a and b are constants; n = number of hours sunshine; N = possible number of hours sunshine. Outgoing longwave radiation was estimated using the Stefan-Boltzmann Law:
Where e = emmisivity (varies from 0-1.0); s = Stefan Boltzmann constant = 5.67 x 10-8 W m-2 K4; T = surface temperature °K, where 0°C = 273K.
While Monteith (1973) proposed Qld = 208 + 6T, to estimate downward longwave radiation. Greene and Nelson (1983) proposed Qld = 258 + 3Ts, where Ts is surface temperature °C. The lower gradient of the Greene and Nelson equation (1983) gave unacceptable values of net longwave radiation. Monteith and Unsworth (1990) gave a slightly modified version of the Monteith (1973) equation with an uncertainty of 30 W m-2. Therefore the Monteith equation has been modified by taking a gradient value of 6T and an average of the constants in the three equations:
Qld = 226 + 6T where T = screen temperature °C.
This equation gives values of Qld which fall within the band of expected uncertainty (Monteith & Unsworth, 1990) over the temperature range -10 to 25°C.
In practice it is the change in net radiation which determines the change in soil temperature at shallow depths, so that the constant in the equation for Qld has no real effect on the results. Chalk soils in the catchment are generally less than 0.3m deep and will approximate the air temperature when averaged over one day.
The rate of warming in the soil layer will be:
Where Ts = soil temperature ËšC; t = time s; Qs = net radiation W m-2; z = soil depth, m; Cs = heat capacity 106 J kg-1 K-1. As an example if the change in net radiation = 20 W m-2, soil depth = 0.3m; Cs = 2.0 (Oke, 1987), then the change in temperature in one day = 1.44°C.
When these equations are applied to the Hadspen temperature record for two depths of soil the results are shown in Figure 8. Taking the soil as a whole, complete freezing probably never took place during January 1841. For the shallower layer freezing took place over a period of six days followed by another six days above freezing. These results are averages for each day. Some freezing at the surface at night may have taken place after 9 January. Rain was reported at Salisbury on 10 and 11 January and a thaw also took place on 13 and 15 January.
While this is based on clear sky conditions, in reality there was some cloud cover during the days leading to the flood and on at least three days snow was present. However, in winter the effect of cloud cover on net radiation is positive (Cogley & Henderson-Sellers, 1984; Groisman et al., 1994) with the effect increasing with decreasing temperature, and positive for all cloud types when the surface is covered with snow, Cogley & Henderson-Sellers, (1984). On 13 and 14 January cloudy conditions were reported at Salisbury. The effect of snow cover on net radiation is negative but it will also insulate the ground to temperature extremes. Snow was reported at Salisbury at 10pm on 9 January, after soil temperatures had exceeded 1°C. Snow was reported at Salisbury during the afternoon of 13 January, by which time the average temperature of the top 0.3m of soil was above freezing point after a thaw at 0800 hours the same day. The overall effect of snow and cloud cover after 13 January would be to prevent sub surface freezing of the soil.
Estimating storm rainfall
The snow cover probably had a rainfall equivalent depth of 10mm. From Hough & Hollis (1998) a snow melt rate of 1.5mm hr-1 is suggested. If frozen ground existed and this rate of melt continued for 6 hours then a runoff rate of 17m3 s-1 would have taken place, far in excess of events at Tilshead during the early afternoon of 16 January. From the estimated hydrograph (Figure 7) the volume of runoff was about 795,600m3, equivalent to 19mm runoff above Tilshead. If the ground was not frozen then the rate of percolation must be measured in the soil profile. The saturated hydraulic conductivity of chalk soils is high (Findlay et al., 1984) so it is important to find the lowest and critical value in the soil profile. Figure 9 shows the average results from 27 soil profiles in the catchment, while Figure 10 shows the frequency distribution of the results at the critical depth of 15cm, which has an average saturated hydraulic conductivity of 5.6mm hr-1. The estimates of peak discharge at Tilshead and Shrewton were used to produce the estimates of storm rainfall in Table 1. A storm duration of 6.5 hours is based on a time of concentration at Tilshead of 2.5 hours and a rising limb of the hydrograph lasting 4 hours. Thus the storm rainfall can be estimated from the depth of percolation and runoff: 6.5 x 5.6 + runoff (19mm) = 55.4mm. This gives an average runoff intensity of 8.5mm per hour. If 5.6mm per hour are lost through percolation then the estimated peak discharge using the rational equation is 33m3/sec, which is 8m3/sec lower than that obtained from the historic flood report and field survey data. Applying the same principles to Shrewton a peak discharge of 46m3/sec is produced, very close to the value of 48m3/sec from field data and flood level. Table 1 shows the estimated depth of rainfall wherein the non-frozen senario gives results which are more consistent with the reports of heavy rainfall in the area: estimates of rainfall intensity in the frozen ground scenario are 1.4 and 0.6mm per hour. Estimates of precipitation with snowmelt are also given.
Other historic floods on the Till
A central problem with estimating flood frequency on small rural catchments is often the lack of a historic record. The flood of 1841 on the Till has been quoted as the only significant flood partially caused by a frozen sub-soil and snowmelt. A closer look at the historic record showed that in 1841: ‘Fifty-one years ago last Monday, an inundation from melted snow took place; the body of water is said to have been greater than on this (January 1841) occasion, but the damage was at that time confined to walls and outbuildings, no dwelling-houses having been destroyed, nor the lives of any human beings lost.’ (Wiltshire Independent, 21/1/1841).
Close examination of the meteorological record for 1790 failed to link any day or conditions with this flood. However, in January 1789 there was a thaw during January and much damage reported in neigbouring Dorset. At Stroud, about 50km north-north-west of Tilshead, Thomas Hughes recorded 0.93 inches (23.6mm) rainfall over a two day period 13-14 January, (Hughes, 1789), while at an unidentified site near Stockbridge some 35km east-south-east of Tilshead 1.05 inches (26.7mm) were recorded on 14 January, (Anon, 1789). At this latter site drifting snow was present. Temperatures were sub-zero at both sites in the days leading up to this event.
The daily CET record also gave sub-zero temperatures every day in January until the 14th, including a temperature of -9.2°C on the 12th indicating a flood at Tilshead which took place on frozen ground. Correcting the CET record using the 1841 data from Hadspen and calculating the day degree index from January 1st, allows a further comparison of the events leading up to the floods of 1789 and 1841. Figure 11 shows the much colder conditions of the earlier event.
The other recorded historic flood took place in January 1915 (Cross, 1967). At Shrewton the rainfall for December 1914 was 177mm. On 1 January 22.4mm were measured and the following day 10.9mm. The combined rainfall on these two days coupled with wet ground conditions led to a serious flood at Salisbury lasting for 24 hours, with the water falling rapidly on the 3rd. It rose again the following night and early next day at Salisbury, and also at Shrewton, where for the 4th, 30.5mm were measured.
Flooding of the Till at Shrewton
The historic flood record for three events allows an estimate of the flood frequency of the Till at Shrewton. The peak flow of the 1841 flood was about 48m3s-1 while that for the 1915 flood based on a field survey and the water level in the photograph in Cross (1967), which must be regarded as a minimum level was about 12m3 s-1. The estimate for the 1789 flood is rather less certain but would have been in the range 25-40m3 s-1 based on the mention of damage to walls and outbuildings. Table 2 shows the estimated peak discharge and return period of the floods based on the time period 1789-2003. Rather than plotting the data as points which imply a level of certainty not wholly justified, circles encompassing the most probable results are shown (Figure 12). It is important to stress that other floods may have occured but as yet have not been found, such that the return period of the 1915 flood could be substantially reduced.
Also shown in figure 12 is an estimate of bankfull discharge which has a return period in the range 5-10 years on many chalk catchments (Harvey, 1969), as compared with about 1.5-2.5 years on clay catchments. The generalised flood frequency curve is considerably in excess of that produced by the Flood Estimation Handbook (IOH, 1999). Even a large overestimation of the peak discharge cannot close the huge difference in the results. The results also raise the question of a probable maximum flood for the Till at Shrewton. While the highest rainfall intensities are in the summer, the catchment is much more responsive to winter rainfall when the catchment is wet and baseflow discharge high. On 28 June 1917 111mm were recorded at Shrewton, mostly within the space of eight hours (British Rainfall, 1917). This gives an average hourly rainfall intensity of about 14mm, but with no reported flooding. It is likely that there would have to have been rainfall in excess of 25mm hr-1 such as occured in 1920 over the Louth catchment in Lincolnshire, another chalk catchment, when an estimated 160mm took place in about three hours (Clark and Vetere Arellano, 2004). This produced an estimated peak discharge of about 160m3 s-1 and with a catchment area of 51km2 makes the more severe flood frequency curve in Figure 11 more realistic than the FEH based estimate. It also suggests a probable maximum flood for the Till at Shrewton in excess of 200m3 s-1.
The scenario of a frozen sub-soil prior to the event cannot be correct. It was proposed on the basis of unrepresentative data collected at Salisbury. It took no account of the thaws before 16 January. When the situation is examined from an energy balance approach a similar conclusion is reached, although the top soil layer may have been frozen during the period 6-9 January. Rainfall depths based on the frozen ground scenario range from 14-21mm which are modest for the time of year and severity of the flood.
The behaviour of chalk catchments has been explained better by the sub-surface hydraulic conductivity data. In general chalk catchments produce floods during winter when the ground is wet, baseflow is high, and when rainfall intensity is high enough and prolonged to be affected by the horizon of relatively low hydraulic conductivity. The very high conductivity in the upper horizon in chalk soils means that lateral subsurface flow takes place when rainfall intensity exceeds about 6mm hr-1 and saturated soil conditions. Only when the upper layer of soil is completely saturated and the lateral flow cannot escape fast enough will overland flow take place.
This would be a very rare event such as at Louth in 1920. Normally during the summer heavy rainfall will be absorbed by the very permeable chalk soils and little direct runoff takes place, an effect which decreases during winter when rainfall depth is higher. However, winter rainfall depths rarely exceed 10mm hr-1 for several hours and this limits peak discharges.
In the future England will witness an increase in urban area and other land use changes which will lead to a greater flood hazard. For less developed countries this process could have a more dramatic effect. The estimate of the 100-year event may itself be in error by about 50%, more usually as an underestimate. This paper has produced estimates of three historic floods on the Till. Uncertainty exists with both the rarity and magnitude of the events, but this is a more honest and realistic view to take as compared with other methods whose apparent sophistication hides an ignorance of the past which we are only just beginning to uncover.
For further information, email: [email protected]
The author would like to thank Terry Marsh of CEH Wallingford for his comments on an earlier version of this paper.Related ArticlesEnvironment Agency releases new flood mapTablesTable 1 Table 2