Planning for floods - will we ever learn?18 April 2006
As knowledge of the flood hazard increases, so too has the number of properties and people at risk. Using the example of the extreme flood which occurred in Boscastle, UK, Colin Clark investigates what progress has been made in the assessment of the magnitude and frequency of floods and suggests ways that flood risks could be addressed in the future
Although there is a growing awareness of the imbalance of humankind with nature, there is little consensus as to how this state of affairs should be tackled. This dilemma comes about partly through a lack of knowledge of natural systems and also by the inability to place isolated events into their proper context. Nowhere is this more apparent than with water related issues such as floods. After over 50 years of modern scientific hydrology much progress remains to be made in the assessment of the magnitude and frequency of floods. Paradoxically, as knowledge of the flood hazard has increased so too has the number of properties and people at risk. How has this come about? What is a more realistic approach to assessing extreme floods in the future? This paper addresses these two questions using as an example the flood hazard at Boscastle, North Cornwall (Figure 1) and makes comparisons with the investigation funded by the Environment Agency.
Over 50 years ago Walter Kollmorgen wrote: ‘there is an urgent need for alternative plans dealing with flood problems. Under present procedure the public is confronted with a one-or-nothing program…Flooded floodplains are in no way to be compared to an enemy, although our attack on flood problems suggest the lavish expenditure of money and resources thrown into a military campaign…Settlement control can resolve many of our flood-loss problems.’ (Kollmorgen, 1953).
In England since the 1950’s the number of properties at risk of flooding by the 1 in 100 year standard has grown to an estimated 1.75M (Environment Agency: www.environment-agency.gov.uk). This figure is probably too low since neither gauged nor other estimates of floods at this rarity match what the historic record shows (Williams & Archer, 2002; Clark, 1996). Older parts of towns were often built on largely flood free sites, such as at Louth in Lincolnshire, UK, suggesting some knowledge of river flooding going back before the 19th century (Clark & Vetere Arellano, 2004). Much recent housing has been built on the floodplain or closer to the river than previously. At Boscastle in North Cornwall, UK, shipping warehouses have been converted into residential and commercial use. Floods in the 1950’s and 60’s led to promises of new technology being used and flood protection measures (Cornish & Devon Post, 16 February 1963) without being fulfilled. Even if the instrumentation had been installed its usefulness in forecasting floods would have been very limited since the necessary real-time flow modelling techniques were yet to be developed.
Severe flooding on Mendip in Somerset, UK, during the late 1960’s prompted a call for better methods of flood assessment, resulting in the Flood Studies Report (NERC, 1975) and more recently the Flood Estimation Handbook (IOH, 1999). Measured flow data were used to produce regression equations that describe the relationships between catchment variables and peak discharge. By their nature regression equations optimise the coefficients about the mean. This will result in flood estimates being above or below the correct result. Where regression equations are used to give variables that are then used in further algorithms then errors can become compounded. Confidence limits on the results are given in the FEH but they seem to be ignored for design purposes. The rainfall-runoff method (Houghton-Carr, 1999) provides an alternative set of flood estimates, but then the decision has to be made as to which result to adopt. The revitalised (ReFH) rainfall-runoff method (Kjeldsen et al 2005) may produce greater consistency with the results of the statistical method, but consistency does not automatically mean accuracy.
One reason why the accuracy of estimates of rare floods are poor is because the data set used to calibrate the methods only includes data which were collected since the 1960’s at the majority of sites, which is generally considered to be ‘flood poor’ (Macdonald, Black, & Werrity, 2002). Estimates of historic floods could help to improve the situation, but little progress has been made in this direction. Historic flood reviews are often poorly carried out and their inclusion in Flood Risk Assessments is only one of lip service as in Lewin, Fryer, & Partners (2003). Although some guidance exists to deal with historic flood data (Potter, 1978; Archer, 1999; Bayliss & Reed, 2001) more practical guidance on the collection and interpretation of historic flood accounts together with some checks which should be done is included here. The approach to historic flood frequency analysis should include:
• Efforts to obtain records of floods from all available sources, and to include additional information if and when it is found.
• Conversion of the descriptions of the events into estimates of peak discharge, bearing in mind that changes in channel cross section and bed levels will affect the results and must be taken into account as far as possible.
• An estimate of the rarity of the flood events. This can only be done when there is a uniform coverage of events: special care is needed if there appears to be a flood poor time period, which may be a result of the floods not being recorded. The period of record used to estimate the return period of moderate events may have to be shorter than that for the biggest event.
• The use of a frequency distribution which allows a linear fit of events of frequencies from 2-106 years.
• As many checks as possible on the reliability of the results using data from different sources. An example of this might be a water balance check or the outcome of a natural experiment, which often takes place during a flood event.
The value of historic flood information
There are more flood events than floods which have been directly measured. This makes the historic record a valuable source of data (Clark, 2003a). Ideally this should include the timing, peak discharge and flood volume. Even when it is not possible to make realistic estimates of these variables, the basic frequency of flooding should become apparent. A consistent picture has to be built up using as much local knowledge about the catchment as possible, and testing the results against the complete range of events up to and including the probable maximum flood (PMF). Table 1 gives a selection of the sources of historic flood information. County Record Offices may have other sources of information. A search using the national archives web site www.a2a.org.uk can also be helpful.
Care has to be exercised when using these sources. For example flood markers may have been moved such as at Bruton (Figure 2). Every effort should be made to check the flood level against other sources, especially eye-witness accounts and contemporary photographs. Although newspaper reports are useful where possible they should be used with rainfall records. Many flood events go unrecorded especially in remote rural areas, and the coverage before 1850 is often poor. It is essential to search all available newspapers because details of the event can vary. In the case of the Boscastle flood investigation (Environment Agency, 2005, hereinafter referred to as EX5160) newspapers were only examined from 1950. This resulted in the floods of 1894, 1903 and 1926 being missed out, or in the case of the 1932 event, being misrepresented. Personal diaries and journals are valuable sources. At Boscastle there were at least three journals kept in the 19th century: Thomas Pope Rosevear (Raymond-Barker & Clark, 2006); Jabez Brown (CRO DDX 383/1-7) and Lydia Saunders (Jewell 2000). None of these were used in the report EX5160, and they would have helped to unravel the flood hydrology at Boscastle. Two floods were described by these accounts; 28 October 1827 and 27 September 1882. Figure 3 shows the account of the earlier event: its use will be described below. Flooding in 1882 was not recorded in the newspapers. While on holiday at Boscastle Lydia Brown Saunders wrote: ‘Tuesday night, or rather Wednesday morning, land water out. Water rushing through the house, in at the back and out at the front door, and still rising. Uncle, hatted and great-coated, on the stairs; Aunt Ellen, in dressing gown and naked feet, parading about downstairs…Oh! It is awful to see the water rushing out of people’s doors. Poor old Granny Cobbledick’s is worse than ours. Just looked out the back and the water is up to the dining room sill. Uncle says it has never been so high, as does (Faithful) Ferrett’…’ Added to this is the account by Jabez Brown of the same event (CRO DDX 383/2). ‘On Tuesday night we were alarmed, the water in Valency river rose so high and flowed over the bank and rails and into our house before we had time to move many things our passage was like a mill stream…but our clothes partly dried was swept from the wash house table…’
Published accounts include chronologies about bigger towns and cities (Wheeler, 1889, 1901). They have to be treated with some caution especially if the original sources are not given. However, they can give a general picture of floods and provide dates which should be checked against other sources. Photographs have mainly been taken during the second half of the 20th century. If taken during the flood they give a minimum water level. It is vital that the correct date is assigned to each photograph. For example the photographs supposed to be of 30 August 1950 at Boscastle and reproduced in report EX5160 were in fact taken during the 1940’s. Since there was no newspaper report of this event then the photograph is probably the only record that exists (Figure 4). It is possible to estimate flood depth from this picture by measuring the width of the telegraph pole in relation to the height of the building not submerged by the water.
Eye-witness accounts are often reported in newspapers but sometimes floodplain occupants or visitors may provide valuable information, especially if they are interviewed soon after the event. Sometimes an elderly resident will provide unique information of a flood which took place long ago. It is not easy to check these accounts and they should be used in conjunction with other sources. Secondary sources include the University of Dundee flood chronology web site www.dundee.ac.uk/geography/cbhe (Black & Law 2004); the pre-1961 editions of British Rainfall (1862-1960) contain references to floods, as do scientific journals. In all cases the original source material should be consulted.
The river channel itself is a historic document since it is a reflection of the local and catchment hydrological processes. Care must be taken to check that downstream influences have not affected the channel dimension such as at Bourton in Dorset, UK, where the effects of a weir over 200m downstream are apparent. Bankfull discharge has a frequency of 1.5-2 years (Leopold Wolman & Miller, 1963; Dunne & Leopold, 1978) on many rivers, one exception being rivers mainly on chalk where the frequency is lower (Harvey, 1969). Wharton (1992) gives useful guidelines for the estimation of bankfull discharge. Where the channel has been badly eroded during a flood then either a photograph of the pre-flood channel can be used to estimate channel dimensions or the 1:2500 Ordnance Survey plans can be used directly to measure channel width to within 0.1m. Channel depth should be still apparent in the field, while the water surface slope can be surveyed in the field and the roughness coefficient estimated from tables (Chow, 1959) or photographs (Barnes, 1967) and applied in the Manning equation:
Q = A R0.666 S0.5 n-1
Where Q = discharge; A = channel area; R = hydraulic radius; S = water surface slope; n = channel roughness.
In report EX5160 an attempt was made to estimate QMED or the 2-year flood by using data from donor catchments. The variable QMED is about the same as bankfull discharge. However, there is no substitute for local data: rivers are like people, they are very variable and may be alike in all characteristics except the one we are interested in. From all the evidence gathered in the field the bankfull discharge was estimated as 16m3/sec.
Boulders have been used to estimate the peak discharge of floods (Costa, 1983). Care has to be taken in deciding whether or not the boulders were moved during the flood and not merely excavated from the floodplain sediments without being moved. In the case of a new flood the evidence should be examined and photographed as soon after the event as possible.
Estimating peak discharge of historic floods of the combined Valency and Jordan catchments
Contemporary evidence as described above and the contemporary channel cross section, either natural or man made, can be used to estimate peak discharge. The water surface slope should be surveyed as soon after the flood event as possible, and if not the existing water surface slope will approximate the likely value. Where a bridge or other structure has been replaced it is essential to use the contemporary cross section. For example at Boscastle in 1958 the road bridge consisted of two arches, which was replaced two years later by a single span. The estimate of the June 1958 flood by HR Wallingford (EX5160) was 90m3/sec, but this was based on the single span bridge. Using the contemporary bridge plans (Cornwall County Council, CR/222; Drw. No. B27/1) gave a peak discharge of 75m3/sec (Clark, 2005a). Estimates of discharge were based on the velocity head equation, which gives the head to pass a given discharge:
Where V = velocity; g = acceleration due to gravity (9.81m/sec)
The summation of head losses due to entry, exit, and friction losses:
HT = HE + HF + HX
Where the subscripts to the head losses (H) are Total, Entry, Friction, and Exit.
Often an open channel is used and conventional hydraulic calculations performed. It is important that the river channel and floodplain are treated separately otherwise there is a tendency to get an unacceptably high result, because the friction factor is rather high in shallow water: indeed the water may be stagnant. More than one estimate of the flood should be made so that the results can be compared for consistency. Further checks are described below.
It is also possible that the peak discharge can be estimated from measurements of the sediments moved during the flood. The author (Clark, 2005a) made a mistake with the boulder evidence at the Boscastle flood in that the very largest boulders had been excavated from the banks but not moved during the event. This conclusion was realised when a mistake was found in the hydraulic calculations made downstream of the boulder dump site.
The revised peak discharge is 230m3/sec. The channel cross section at the boulder site was probably not fully eroded at the time of peak flow. Assuming that half of the channel erosion had taken place then the revised peak discharge at this site is 243m3/sec. Excluding the two largest boulders from the sample gives a median boulder size of 0.68m which gives a peak discharge of about 230m3/sec. It was noted that large boulders were seen in the river banks upstream of Boscastle, which were not eroded during the flood, but which may appear sometime in the future.
The question then arises as to the magnitude of the flood
which may have brought these larger boulders into the valley of the Valency.
The Jordan catchment drains an area of about 2.1km2. Figure 5 shows the valley cross section just above Marine Terrace before it joins the Valency. While some recent renovation work was being carried out at Frogpits, evidence was found of a flood whose level was about 0.3m higher than in 2004. This may represent the flood of 1827, when pigs had to be taken out of the roofs of houses as noted by Thomas Rosevear (Figure 3). The estimated flows of the floods of 1827 and 2004 are given in Table 2.
Although the estimates for the 2004 event are close, the report EX5160 gave a combined estimate of the peak discharge for the Valency and Jordan in 1827 as 30m3/sec. From Rosevear’s description of the 1827 event, the east river, that is to say the Valency to the east of the bridge did not flood and so caused no damage, while to the west, below the confluence of the Jordan and Valency there was considerable damage. This means that the likely discharge below the road bridge was about 55m3/sec.
Flood frequency analysis
In classical flood frequency analysis the events are ranked in order of size and the return period calculated using a plotting formula. The most commonly used is that of Gringorton (1961) which gives similar results to that of Clark (1983).
Rp = [ I – 0.44/ n + 0.12] –1 (Gringorten, 1961)
Rp = M/ (N – a) (Clark, 1983)
Where Rp = Return period, I = rank order, n = record length (Gringorten, 1961); M = length of record; N = rank order; a = a variable but = 0.3 for records longer than 50 years (Clark, 1983). Table 3 shows the schedule of floods at Boscastle for which it has been possible to estimate the peak discharge. The table includes the results in EX5160. Several points are worth highlighting:
• The flood of 1952 at Boscastle did not take place; this event may have been confused with the Lynmouth flood on the same day.
• There was no overflowing of the Valency on 10 June 1932. The newspaper report stated that heavy rain fell at Boscastle doing comparatively little damage apart from the roads (Cornish & Devon Post 11/6/1932).
• Report EX5160 has omitted the floods of 1882, 1894, 1903 (twice), 1926 and the flood of the 1940’s. It is possible that there was a second flood during the 1940’s but no details have yet been found.
• The estimate in EX5160 of the flood of 1958 was based on the single span bridge which replaced the two arch structure in 1960.
• There was partial blockage of the road bridge during the February 1963 event which makes the assessment of peak discharge difficult. If a value of 60m3/sec is correct then a rainfall intensity of 16mm/hour lasting for at least one hour is needed. The Council Engineer estimated the flow on the Jordan as 2.27m3/sec, which gives a peak rainfall intensity of 7mm/hour. This is much more typical of frontal rainfall during the winter. On the higher ground above Boscastle the rainfall intensity may have been 10mm/hour giving an upper estimate of 37m3/sec.
• When the results from EX5160 are plotted onto extreme probability paper (Figure 6) they show an exponential increase in discharge with increasing rarity even on the extreme probability scale. This is in contrast to the growth curve of local rainfall such as at Lesnewth and Bossiney, the former site being located in the centre of the catchment (Figure 7). In EX5160 it is stated that the rainfall growth curves in SW England are steeper than flood growth curves, but a comparison of Figures 6 and 7 show just the opposite. In the present paper the flood growth curve is similar to the rainfall growth curve. As a further comparison the flood growth curve for Horner Water on Exmoor in SW England is also shown in Figure 6.
• The length of record chosen in EX5160 was 200 years, even though their flood data extended from 1827-2004, a period of 178 years. However, there are no floods in their analysis during the period 1827-1932, a period of 104 years. Not only is this unjustified but it makes the effective time period from 1932-2004 or 73 years. But the flood of 1932 did not result in the Valency overflowing so the final valid time period for EX5160 is 1950-2004, a period of 55 years. These observations alone invalidate their return periods in Table 3.
• Although the author overestimated the peak discharge of the 2004 flood, this has not affected the present outcome of the flood frequency analysis: the result is essentially the same (Clark, 2005a).
Estimation of the return period of the 2004 flood
The return period of he 2004 flood cannot be assessed using conventional methods. This is because it is so much bigger than all other events and also because it is so rare in relation to the length of record. Its rarity can be estimated using a combination of antecedent soil moisture deficit and the return period of the storm. Table 4 shows two estimates of the return period of the storm. The much lower result from the FEH is an effect of the method of extrapolation of estimated rainfall depths at high return periods (Macdonald & Scott, 2001). The supposed relationship of the storm rarity to flood rarity has recently been changed with the publication of the Revitalised Rainfall-Runoff method, ReFH, (Kjeldsen et al. 2005). Whereas in the FEH (IOH, 1999) for example, a 50 year storm was believed to produce a 30 year flood, with the two values converging at floods in excess of a 1 in 200 year return period, the ReFH now equates storm and flood peak return period up to the 1 in 150 year event. In reality during the time of more common storms the catchment wetness will exert a significant effect on the resulting flood, while with rare storm events this effect will tend to be less.
For floods on the Upper Brue in SW England, the relationship between the ratio of Flood return period to Storm return period and the antecedent SMD is shown in Figure 8. For dry catchments a rare storm will produce a flood of lower rarity than the storm itself, but when the catchment is wet the opposite takes place. This relationship assumes that both storm and flood rarity and SMD have been correctly assessed. Given an antecedent SMD value of 12mm on 16 August 2004 and a storm rarity of about 10200 years gives the return period of the flood as 10200 x 0.744 = 7588 years. When this result is combined with a peak discharge of 230m3/sec a very good comparison with the expected result in Figure 6 is apparent.
Application of the same methodology to the Jordan catchment is much more difficult because the antecedent SMD for this smaller area is less certain. By reversing the Rational equation for a peak discharge of 20m3/sec and a lag time of 0.5 hours, and using the saturated hydraulic conductivity data collected in the catchment gives a flood producing rainfall of 33mm/hour. Using local rainfall analyses this has a return period of 398 years. The SMD was zero since any remaining deficit, which was estimated at about 20mm would have been taken up by rainfall taking place several hours before peak discharge. The return period of the flood event then becomes 398 x 2.253 = 897 years. When this result is combined with an estimate of bankfull dicharge of 1.6m3/sec and a PMF of about 90m3/sec a flood frequency curve (not shown here) reveals that the new culvert for the Jordan river has a likely design standard of only 50 years. This is below the 1 in 75 years standard expected by insurance companies.
Checking the reliability of results
There will always be uncertainty in estimates of floods. This uncertainty can be reduced by making certain checks. In addition to those already mentioned three other methods will be described.
1. Water balance of flood volume.
2. Comparison of flood discharges at one location and another during a flood event.
3. Consistency of flood estimates with both bankfull discharge and the PMF.
A water balance check on flood volume is essential. Report EX5160 did not include this check for the Boscastle flood. Figure 9 shows the estimated hydrographs for this event where the non-linear flow model was able to reproduce the timing of the flood as made by eye-witness accounts. The percentage runoff for the two estimates are 51% and 65% assuming that the areal depth of rainfall was about 160mm. The complete water balance thus becomes:
Rainfall = runoff + SMD + percolation
161 = 113 + 12 + 29
This gives a mismatch of about 16mm rainfall in the present water balance, but is within 10% of the expected value. One of the problems with the method is that a reliable estimate of the SMD has to be made. At present both MORECS (Meteorological Office Rainfall and Evaporation Calculation System) and MOSES (Meteorological Office Surface Exchange System) often gives results which are too high to be of practical value (Clark, 2002, 2003b, in prep.). Direct measurements using lysimeters or soil moisture probes are the best methods, followed by estimates of potential evaporation using empirical data from sunken pans which are related to air temperature, humidity and windspeed. The SMD for 15 August 2004 was estimated from measurements of actual and potential evaporation at CHRS which were then combined with daily rainfall at Lesnewth. The result was an SMD = 0.0mm. Towards the bottom of the catchment area rainfall was less and the record at Lesnewth was adjusted downwards based on the rainfall record collected at nearby Bossiney. The weighted SMD for the whole catchment was 12mm. Comparison of the rainfall regime at Lesnewth and CHRS (Figure 10) shows that Lesnewth was wetter by about 60mm. Since the measured SMD at CHRS before the flood was just over 50mm then at Lesnewth soils would have been at field capacity with an SMD of zero. Deep percolation has to be estimated from a site survey of the saturated hydraulic conductivity of local soils.
The second check which can be made is to use the estimates of bankfull discharge. In discussion with the Environment Agency some doubts have been expressed about a bankfull discharge of about 16m3/sec as compared with a value of 6.6m3/sec (EX5160). This problem can be solved by asking a question regarding the simultaneous behavior of the river level above and below Boscastle road bridge, given that in the early stages of the 2004 flood little flow was produced by the Jordan stream which joins the Valency close to the bridge: if the lower estimate of bankfull discharge is correct was the car park above Boscastle bridge flooded before the discharge reached about half bankfull discharge below the bridge where the bankfull flow is about 26m3/sec? In this case the answer is negative. A similar question can be posed for the June 1993 event. A photograph taken during this event shows the Valency near bankfull flow below Boscastle bridge. If the lower estimate of bankfull flow above the bridge was correct then about 10-12m3/sec of overbank discharge would be expected to flow across the floodplain and bridge for some time since the hydrograph was attenuated. This overflow did not take place. From this natural experiment a bankfull discharge above the bridge of 14-16m3/sec is suggested.
The third check which can be applied to the flood frequency analysis is the linkage between bankfull discharge, the historic flood estimates, and the PMF. Figure 6 shows that using the historic record alone a bankfull discharge of about 16m3/sec is suggested. The estimate of the PMF is about 500m3/sec. Application of the non-linear flow model used an estimate of PMP based on transposition (WMO, 1986) of the Martinstown storm (Clark, 2005b) and local soils data and a time to peak formulation (Clark, 2006) which is based on historic storm and flood data:
Tp = 1/ (0.340 + 0.0044R)
Where Tp = time to peak (hours), R = rainfall mm/hour lasting for 0.5 hours.
The result was 580m3/sec, which came about because of the low time to peak at very high rainfall intensities. However, the result is close enough to 500m3/sec to give support for a PMF of this order of magnitude. The two estimates of PMF give runoff rates of 25-29m3/sec per km2 which is in excess of the catastrophic flood of Allard, Glasspole & Wolf (1960) by a similar margin to the 1771 flood on the river Tyne (Archer, 1993) and the Divie flood in Morayshire (Acreman, 1989).
Fundamental to any check of results is the maxim that facts should be used to demonstrate models as opposed to models demonstrating facts. As soon as the model takes priority over observations then there is a danger that false conclusions will be obtained. In 1952 the Lynmouth flood caused a revision of the ICE 1930 normal maximum flood standards: future estimates of floods were substantially increased (Dobbie & Wolf, 1953). In the current estimation of PMF for spillway design floods, the model time to peak from the FEH is reduced by a factor of 0.67. For the Boscastle flood study (EX5160) the time to peak was reduced by a factor of 0.5. This helped to increase the estimate of peak discharge to 130m3/sec which is 47m3/sec lower than that obtained in the same report from hydraulic modelling, and 100m3/sec lower than the estimate in this paper. In the hydrological modelling of EX5160 the percentage runoff was increased to levels in excess of what the FEH could produce, while the rainfall was also increased by a factor of 1.3. In spite of all these changes to the model it was then stated: ‘It would be inappropriate to suggest that the FEH methodology should be modified on the basis of one flood event but it has highlighted the need for further data and study in this area.’ (Report EX5160, 2005). Clearly historic floods can provide this data and it would be negligent to wait until another catastrophic flood took place, especially if it resulted in loss of life. Facts should demonstrate models.
Planning for floods includes planning new housing in flood free locations at least to a 1 in 100 year design standard on the one hand, and the design of dam spillways to safely pass the PMF on the other. Enough freeboard in both cases should be included to allow for possible under design. The Boscastle flood will remain in the memory of those present at the time for many years to come. The memory of hydrologists must not be confounded by a poorly researched flood history. Adoption of the 75 year design standard as given in report EX5160 would leave Boscastle vulnerable to floods greater than the 1 in 35 year event as assessed in this paper; it would not be capable of safely passing the Boscastle flood of 1958 in which Charles Berryman was drowned. This situation could lead to a life threatening situation in the case of a more serious event, though less severe than the flood of 2004. The search for historic floods must always continue because it only gives a minimum estimate of the flood hazard. It cannot tell us what the future holds, which may mean either more or less flooding than expected. Although the past must be our best guide to the future, changes in land use and weather patterns may lead to further revisions of current findings.
Colin Clark, Charldon Hill Research Station, Shute Lane, Bruton, Somerset, England BA10 0BJ. Email: email@example.com.
The author would like to thank Anne and Rod Knight for drawing his attention to the account of Lydia Saunders, and Mrs Iris Olde for allowing him to photograph some pictures in her collection and information about the flood in the 1940’s.
|Figure 3 - Thomas Rosevear's account of the 1827 flood|
Sunday 28. One of the most awful days I ever experienced at Boscastle. It rained very heavily in the morning & whilst we were in the Chapel increasingly so – when about to leave the whole street was filled with a body of water rolling down & carrying all the materials with that devastation & ruin were its concommitants – by about 1 o’clock the rain ceased leaving the fine Mcadamized road in complete ruin from Polrunny to Dunn Street – At Bridge teams of waggon Horses were saved with difficulty & pigs also belonging to the cottagers were taken out of thro ye Roofs of Houses – the Langford & cottages the west side of the Bridge suffered mutch – But thro the goodness of God on the East river the waters were raised but little & our property preserved in safety – I would mark the finger of Divine providence & Acknowledge his loving kindness