Abstract

In December 2015, northern England experienced two major flooding events with extreme, even in some locations unprecedented, rainfalls and flooding. New 24-, 36-, and 48-hour UK rainfall records were created of 341.4, 401.4, and 405.2 mm, respectively. Three river-flow gauging stations, with flows of around 1,700 m3/s exceeded the previous peak flow record for England and Wales. There was widespread flooding, including major towns and cities, some of which had recent flood alleviation schemes. In Cumbria, the flood events in 2005, 2009 and 2015 compared with previous and historical events raise questions about the stationarity of the flood data and flood-producing mechanisms. These possible effects are less apparent elsewhere in northern England. This paper discusses whether present methods of estimating flood risk are able to cope with such extreme events and suggests topics for future research. In the meantime, for studies where flood estimates are important, practical hydrologists are faced with the difficult task of producing design flood estimates which fit with our understanding of these events.

INTRODUCTION

This paper, which describes the flood events in northern England in December 2015, is based on three presentations at the British Hydrological Society's (BHS) International Conference on 31st August 2016, given by Peter Spencer, Ian Perkins, and Duncan Faulkner, titled ‘The floods of December 2015 in northern England’, ‘What do you do after a major flood?’ and ‘How many 100-year floods before hydrologists lose their credibility?’ respectively. Figure 1 shows the area covered.

Figure 1

Location map, showing river gauging stations and raingauges.

Figure 1

Location map, showing river gauging stations and raingauges.

The floods of December 2015 in northern England were described by the Prime Minster and the Environment Secretary immediately after the events as ‘unprecedented’ (News Outlets 2015). This general description appears reasonable for the event in terms of the overall magnitude, rarity and spatial extent of the flooding, though it does not apply to all locations; the terms ‘extreme’ or ‘exceptional’ might also be applied. Record 24-, 36-, and 48-hour rainfalls for the UK were recorded. Much of northern England was affected, and some locations were flooded more than once. Many river gauging stations recorded their highest ever flows, and three stations with flows of around 1,700 m3/s broke the previous record of 1,520 m3/s for England and Wales.

Since the BHS Conference, a recent review (Marsh et al. 2016), has described the winter floods covering the UK as a whole, including rainfalls from November to February, temperature, and water resources. The review also compares the December 2015 events with previous notable runoff episodes, and concludes that the extent, severity and duration of the resulting flooding has few, if any, close modern parallels.

Since 2015 the Met Office has named storms based on their expected impacts. During December 2015 there were three named storms: Desmond (4th and 5th), Eva (23rd and 24th), and Frank (30th and 31st); the heavy and persistent rainfall on 25th and 26th December was not named. In this paper, for consistency and simplicity the event of 5th and 6th December is referred to as Storm Desmond and the event of 25th and 26th December 2015 is referred to as the Christmas event. Storms Eva and Frank brought only limited flooding.

FLOODING

Overview of flooding

In north-west England, Storm Desmond affected much of Cumbria (except for the far west and south-west), and the catchments of the Lune and Wyre in Lancashire. There was serious flooding in Carlisle, Cockermouth, Kendal, Keswick, Appleby and Lancaster as well as many other towns and villages. The flood defences constructed after the 2005 and 2009 floods were overtopped at Carlisle, Keswick and Cockermouth. In the north-east, Storm Desmond affected the Tyne catchment, both dispersed communities and Corbridge, Haydon Bridge, and Hexham,

The Christmas event affected mainly parts of Lancashire and west Yorkshire. In the north-west, the floods affected the catchments of the Ribble, Douglas, Yarrow, Roch and Irwell. Many towns and villages were flooded, and more than 100 properties were flooded at Whalley, Croston, Parbold, Padiham, Radcliffe, Rochdale and Salford. In the north-east, the main catchments affected were the Calder, Wharfe, Aire, and Ouse, particularly parts of Mytholmroyd and Hebden Bridge (in Calderdale), Tadcaster, Leeds and York (where the majority of flooding was from the Foss rather than the Ouse).

Impacts of flooding

An initial total cost estimate to the UK of £5 billion to £5.8 billion was reported in a paper to the House of Commons (2016). The Association of British Insurers (2016) subsequently estimated the total cost of insurance payouts would reach £1.3 billion. There was one reported death in Cumbria and another in Northern Ireland (House of Commons 2016). Around 17,800 properties were flooded in northern England during December 2015 (Table 1), of which around 25% were businesses. The flooding in Cumbria in 2005 and 2009 had been more localised, the numbers flooded being 2,500 in 2005 and 2,240 in 2009 compared with 6,870 in 2015. Floods have wider impacts on people than just property flooding. For instance, in Lancaster, 61,000 properties were without electricity (Royal Academy of Engineering 2016). In Greater Manchester, the electricity supply to 31,200 properties failed, and the infrastructure damage has been estimated at £11.5 million (Greater Manchester Flood Investigation Report 2016). Such Flood Investigation Reports have been produced by all the affected Lead Local Flood Authorities. The series of heavy forecast and actual rainfalls over the winter created uncertainty and worry. For instance, Appleby and Glenridding were flooded several times (Appleby on the 5th, 22nd, and 26th December 2015, and Glenridding on the 5th, 9th, and 21st December 2015).

Table 1

Numbers of properties flooded in northern England in December 2015

County Properties Main locations Properties 
Cumbria 6870   
  Kendal 2140 
  Carlisle 1930 
  Keswick 730 
  Cockermouth 630 
Lancashire 2590   
  Lancaster 360 
  Croston 340 
Greater Manchester, Merseyside & Cheshire 2500   
  Salford 750 
  Radcliffe 670 
  Rochdale area 230 
North East 300   
Yorkshire 5800   
  Calderdale 3790 
  Leeds 3360 
  Bradford District (includes city) 1010 
  York 630 
County Properties Main locations Properties 
Cumbria 6870   
  Kendal 2140 
  Carlisle 1930 
  Keswick 730 
  Cockermouth 630 
Lancashire 2590   
  Lancaster 360 
  Croston 340 
Greater Manchester, Merseyside & Cheshire 2500   
  Salford 750 
  Radcliffe 670 
  Rochdale area 230 
North East 300   
Yorkshire 5800   
  Calderdale 3790 
  Leeds 3360 
  Bradford District (includes city) 1010 
  York 630 

WEATHER SYSTEMS

The winter weather of 2015/16 was dominated by a series of low pressure systems which tracked in from the Atlantic, and when combined with warm, wet air from the south-west produced sustained and heavy rainfall. Northern England experienced around twice the normal November rainfall. Thus immediately before the heavy rainfalls at the start of December much of the ground was already wet or saturated.

Storm Desmond was an extra-tropical cyclone with a warm conveyor system driven by strong winds which gave a prolonged warm and moist south-westerly (Tropical Maritime) airstream. Where there was significant orographic enhancement these gave prolonged heavy rainfall lasting for around 36 hours. Thus, although rainfalls were not intense, the totals were exceptional in some locations.

The weather systems in Cumbria in January 2005 and November 2009 were of a similar nature to Storm Desmond, though rainfalls were more localised and totals were generally less. The January 2005 event had an added feature called a ‘Sting Jet’ (Environment Agency 2006 (Chapter 3 by the Met Office)).

After Storm Eva, another warm frontal zone passed over the UK on the morning of 25th December, bringing scattered showers and by the evening an occluded front over northern England brought more intense storms, which were followed by a second pulse of frontal rain from the early hours of 26th December.

RECORDED RAINFALL

The spatial pattern of rainfall

Figure 2 shows the rainfall pattern over 2 days (09:00 to 09:00) for Storm Desmond and the Christmas event. In the figure, the northern and southern extents of the rainfall are Environment Agency Area boundaries. Comparison of the two figures shows that although rainfalls over 50 mm were more widespread in the Christmas event, Storm Desmond had significantly heavier rainfalls (over 150 mm), which were concentrated over the central Cumbrian hills.

Figure 2

Two-day (event) rainfall depths (mm) in northern England: left – Storm Desmond; right – Christmas event.

Figure 2

Two-day (event) rainfall depths (mm) in northern England: left – Storm Desmond; right – Christmas event.

Rainfall depths and return periods

Tables 2 and 3 list rainfall depths and indicative return periods at selected raingauges.

Table 2

Rainfall depths (mm) and indicative return periods (years) Storm Desmond

Station Catchment 12 hour
 
24 hour
 
36 hour
 
Rainfall Return period Rainfall Return period Rainfall Return period 
North-west        
 Skelton Eden 75 130–150 138 >1000 183 >1000 
 Brotherswater Eden 179 200–500 293 >1000 369 >1000 
 Thirlmere Derwent 196 >1000 325 >1000 401 >1000 
 Honister Derwent 209 250–600 341 >1000 367 >1000 
 Kentmere Kent 106 25–75 183 60–100 225 100–200 
 Turnerford Lune 68 20–30 110 75–100 137 150–300 
 Abbeystead Wyre 66 10–15 97 20–40 115 25–60 
North-east        
 Chirdon North Tyne 56 25–30 94 100–120 122 180–280 
 Garrigill Noonstones South Tyne 112 130–360 151 160–580 179 180–750 
 Hexham Firtrees Tyne 31 58 13 74 25–30 
 Burnhope Reservoir Wear 62 10 106 35–80 126 40–100 
Station Catchment 12 hour
 
24 hour
 
36 hour
 
Rainfall Return period Rainfall Return period Rainfall Return period 
North-west        
 Skelton Eden 75 130–150 138 >1000 183 >1000 
 Brotherswater Eden 179 200–500 293 >1000 369 >1000 
 Thirlmere Derwent 196 >1000 325 >1000 401 >1000 
 Honister Derwent 209 250–600 341 >1000 367 >1000 
 Kentmere Kent 106 25–75 183 60–100 225 100–200 
 Turnerford Lune 68 20–30 110 75–100 137 150–300 
 Abbeystead Wyre 66 10–15 97 20–40 115 25–60 
North-east        
 Chirdon North Tyne 56 25–30 94 100–120 122 180–280 
 Garrigill Noonstones South Tyne 112 130–360 151 160–580 179 180–750 
 Hexham Firtrees Tyne 31 58 13 74 25–30 
 Burnhope Reservoir Wear 62 10 106 35–80 126 40–100 
Table 3

Rainfall depths (mm) and indicative return periods (years) Christmas event

Raingauge Catchment 12 hour
 
24 hour
 
36 hour
 
Rainfall Return period Rainfall Return period Rainfall Return period 
North-west        
 Great Harwood Ribble 61 20 95 55–90 103 50–75 
 Common Bank Douglas 53 15 73 30–40 81 30–40 
 Holden Wood Irwell 71 25–35 109 50–120 128 50–160 
 Cowm Res Roch 70 15–20 107 40–70 116 30–50 
North-east        
 Scargill Nidd 42 77 20–30 82 15–25 
 Thruscross Wharfe 55 10 102 60–115 108 45–80 
 Skipton Snaygill Aire 48 10 90 60–100 101 60–110 
 Gorpley Calder 71 20–30 113 55–105 126 50–95 
Raingauge Catchment 12 hour
 
24 hour
 
36 hour
 
Rainfall Return period Rainfall Return period Rainfall Return period 
North-west        
 Great Harwood Ribble 61 20 95 55–90 103 50–75 
 Common Bank Douglas 53 15 73 30–40 81 30–40 
 Holden Wood Irwell 71 25–35 109 50–120 128 50–160 
 Cowm Res Roch 70 15–20 107 40–70 116 30–50 
North-east        
 Scargill Nidd 42 77 20–30 82 15–25 
 Thruscross Wharfe 55 10 102 60–115 108 45–80 
 Skipton Snaygill Aire 48 10 90 60–100 101 60–110 
 Gorpley Calder 71 20–30 113 55–105 126 50–95 

Rainfall return periods

This paper quotes the frequency of rainfall and floods in terms of return period, the average interval between years containing one or more floods greater than a given magnitude (Institute of Hydrology 1999). The term ‘return period’ is used because it coincides with that in the Flood Estimation Handbook (Institute of Hydrology 1999). Thus a 100-year event has a likelihood of 1/100 of occurring or being exceeded in any one year, which can also be expressed as an Annual Exceedance Probability (AEP) of 1%.

The FEH Web Service (2015) provides catchment-averaged rainfall depth-duration-frequency (DDF) relationships from two methods, referred to as FEH1999 and FEH2013. The FEH Web Service does not provide point rainfall (such as at a raingauge) DDF relationships from the FEH1999 method, and these return periods were derived using the FEH CD-ROM. FEH1999 follows Volume 2 of the Flood Estimation Handbook (Faulkner 1999) and FEH2013 results from a comprehensive update (Stewart et al. 2013). The range of estimated return periods in Table 2 are based on both methods. At very low return periods the estimates by the FEH1999 and FEH2013 methods are very similar. Above around 20 or 30 years, return periods by the FEH2013 method are nearly always higher than FEH1999 for all the stations in Table 2 except Honister. In some cases return periods using FEH2013 can be much more than 1000 years for durations of 24 hours or more – for example, at Brotherswater the return periods for 24 and 36 hours are 2844 years and 8643 years, respectively.

Although it used improved statistical analyses and is based on more recent data than FEH1999, the dataset for FEH2013 did not include the large events in 2005 (except for a few stations), 2009, 2012, and now 2015. CEH Wallingford are currently investigating the effects of updating the dataset on the FEH 2013 relationships in a study jointly funded by CEH Wallingford and the Environment Agency.

As shown in Table 4 for the Derwent at Camerton and the Leven at Newby Bridge, return periods for catchment averaged rainfall depths by the FEH1999 and FEH2013 methods tend to be more similar than for those individual raingauges which recorded the higher rainfalls; the values for 2009 are taken from Stewart et al. (2012).

Table 4

Catchment rainfalls (mm) in 2009 over 36 hours

Catchment November 2009
 
December 2015
 
Rainfall (mm) Rainfall return period (years) Rainfall (mm) Rainfall return period (years) 
  FEH1999 FEH2013  FEH1999 FEH2013 
Derwent 155.7 157 193 163 200 230 
Leven 200.3 185 485 239 470 1800 
Catchment November 2009
 
December 2015
 
Rainfall (mm) Rainfall return period (years) Rainfall (mm) Rainfall return period (years) 
  FEH1999 FEH2013  FEH1999 FEH2013 
Derwent 155.7 157 193 163 200 230 
Leven 200.3 185 485 239 470 1800 

Rainfall maxima

New record 24-, 36-, and 48-hour rainfalls were recorded during Storm Desmond (Honister 341.4, Thirlmere 401.4 and 405.2 mm, respectively). Brotherswater recorded 293.4 mm in 24 hours and 368.6 mm in 36 hours, which also broke the previous records. Kendon et al. (2016) show eight daily rainfalls of over 200 mm during 2015 in the UK (in their Figure 43), of which three were in Cumbria on 5th December, and these may be compared with the 16 observations exceeding 200 mm in over 100 years from 1866 to 1968 reported by Rodda et al. (2009) for the whole of the British Isles.

Table 5 summarizes the UK maximum rainfall depths after the November 2009 and December 2015 events; the 48-hour maximum is little changed. For durations of 24 to 48 hours, Stewart et al. (2012) assigned return periods to these 2009 rainfalls at Seathwaite Farm as 93 to 172 years by the FEH1999 method and 1,862 to 3,656 years by the FEH2013 method. The new maxima 24-hour rainfall depth at Honister is around 96% of the Probable Maximum Precipitation (PMP) in the Flood Studies Report (FSR) (1975) and at Thirlmere the new 36- and 48-hour maximum depths are around 114% and 109%, respectively, of the PMP in the FSR.

Table 5

UK rainfall maxima (mm) after October 2009 event and December 2015

 Maxima 2009 Maxima 2015
 
Duration (hr) Seathwaite Farm Honister Thirlmere 
103.0 120.2  
12 189.2 209.2  
18 263.2 293.6  
24 316.4 341.4  
36 395.2  401.4 
48 405.0  405.2 
 Maxima 2009 Maxima 2015
 
Duration (hr) Seathwaite Farm Honister Thirlmere 
103.0 120.2  
12 189.2 209.2  
18 263.2 293.6  
24 316.4 341.4  
36 395.2  401.4 
48 405.0  405.2 

Example – Rydal Hall

Rydal Hall (upstream of Windermere) has been selected as an example of rainfalls in November and December because it is located in central Cumbria and has a long record of daily rainfalls. At an altitude of 78 mAOD, orographic enhancement in Storm Desmond would be expected to be less than at more upland gauges such as Honister (358 mAOD). It is not included in Table 2 because there is only a daily raingauge (measured at 09:00 GMT each day), so that 24-, 36-, and 48-hour rainfall depths are not available. The record of daily rainfalls starts in January 1921, with a few gaps or periods when only monthly data are available, making 95 years of record to December 2015 (composed of two raingauge sites at the Hall).

Figure 3 shows daily rainfalls (after quality control by the Meteorological Office) at Rydal Hall in November and December 2015. The total November rainfall of 509 mm and total December rainfall of 781 mm may be compared with average rainfalls in those months of 242 mm and 261 mm, respectively. Only three days in November and December 2015 had no rain. The rainfall on 14 November 2015 is from Storm Abigail, which was the first named storm in the UK.

Figure 3

Daily rainfalls at Rydal Hall, November and December 2015.

Figure 3

Daily rainfalls at Rydal Hall, November and December 2015.

The daily rainfalls from Storm Desmond are amongst the highest in the 95 years of record, the 1-day rainfall of 131 mm is rank 4, the 2-day rainfall of 237 mm is rank 2, and the 3-day rainfall of 280.4 mm is rank 2. By the FEH2013 method these depths have return periods of 125, 1025 and 1320 years, respectively – so at this station FEH2013 appears to over-estimate the return periods of the 2-day and 3-day rainfalls in Storm Desmond.

Comparison of event rainfalls in January 2005, November 2009 and Storm Desmond

The January 2005 event is described in Environment Agency (2006) and the November 2009 event in Environment Agency (2012) and Stewart et al. (2012). December 2015 was much the more significant – heavy rainfalls were more widespread and were generally higher. As an example, Table 6 summaries rainfalls in January 2005, November 2009, and December 2015 at Brotherswater and Honister, two raingauges which recorded amongst the highest rainfalls in these events; Brotherswater is upstream of Ullswater and Honister is upstream of Derwentwater. 36-hour rainfalls at Honister were similar in November 2009 and December 2015.

Table 6

Rainfalls at Brotherswater and Honister

 Brotherswater
 
Honister
 
Duration (hr) Jan 2005 Nov 2009 Dec 2015 Jan 2005 Nov 2009 Dec 2015 
49.4 59.6 103.4 69.0 83.6 120.2 
12 73.6 114.4 178.8 114.4 159.2 209.2 
18 97.6 162.2 240.2 143.0 234.0 293.6 
24 138.4 200.8 292.4 176.6 301.4 341.4 
36 171.0 242.8 368.7 207.4 376.6 379.2 
 Brotherswater
 
Honister
 
Duration (hr) Jan 2005 Nov 2009 Dec 2015 Jan 2005 Nov 2009 Dec 2015 
49.4 59.6 103.4 69.0 83.6 120.2 
12 73.6 114.4 178.8 114.4 159.2 209.2 
18 97.6 162.2 240.2 143.0 234.0 293.6 
24 138.4 200.8 292.4 176.6 301.4 341.4 
36 171.0 242.8 368.7 207.4 376.6 379.2 

Future rainfall events

The rainfall mechanisms in the storms of January 2005, November 2009 and Storm Desmond were broadly similar and the type of event was not unusual. There appears to be no reason why such events should not occur again, possibly with greater rainfall depths.

As part of their work for the National Flood Resilience Review (HM Government 2016), the Met Office assessed the potential for rainfalls to exceed those in the recent past. The Met Office concluded that winter monthly rainfall totals over the next 10 years could plausibly be 20% higher than Storm Desmond for Carlisle and the Christmas event for the Calder Valley. The Met Office considered that there is limited evidence that climate change is affecting rainfall over England and Wales, and that natural variability will continue to dominate extreme rainfall for the next 10 years. The analysis indicated that there is a 10% likelihood of a region experiencing monthly rainfall greater than the existing record in any year within the next 10 years.

RIVER FLOWS

Hydrometric aspects

Under such exceptional conditions, river gauging stations may not continue to record. In Cumbria and Lancashire in December 2015, 44 stations were damaged (including 32 which were flooded), and communication problems meant that no data were received during the events at 27 stations. Post-event survey recovered most of the peak levels which had not been recorded during the events.

At most river gauging stations, the stage is measured and converted to flow using a rating curve. The accuracy of flow estimates in large events has considerably improved in recent years, due to a combined approach using Acoustic Doppler Current Profiler (ADCP) gauging equipment for spot flow measurements, the use of calibrated hydraulic models to derive top-end rating curves, and comparison of flows between stations along the same watercourse as a check on consistency. However, uncertainties in the flow estimates remain due to factors such as extrapolation of the rating curve beyond spot gaugings and the current inability to gauge floodplain flows. Channel morphology may change during the event, so that the bed form before and after the event may not reflect the bed at the flood peak. In the December 2015 events, spot flow gauging was often not possible during the floods due to difficulties of access, safety, and the fast river velocities. A notable exception was a gauging at 1270 m3/s at Bywell on the Tyne, which is the highest known gauging ever in England (Figure 4).

Figure 4

The Tyne at Bywell – photograph during the highest gauging in England. (Photo: Ian Downs, Environment Agency Hydrometry team).

Figure 4

The Tyne at Bywell – photograph during the highest gauging in England. (Photo: Ian Downs, Environment Agency Hydrometry team).

Peak river flows

Figure 5 shows the river flow stations which recorded the highest flow on record in December 2015, and indicates the spatial pattern of the rivers which were most affected.

Figure 5

River flow stations which recorded the highest flow on record.

Figure 5

River flow stations which recorded the highest flow on record.

Tables 7 and 8 summarise the peak flows at key stations. Bearing in mind the comments above, peak flows have some uncertainty, and some flows are still provisional, awaiting more detailed hydraulic modelling and review. Return periods have been derived based on the current estimate of the peak flow and applying standard FEH methods (Environment Agency 2015). The datasets have been updated to include the December 2015 events at all stations affected by the event; values are indicative and different values may be obtained in more detailed studies. QMED is the median annual flood.

Table 7

Peak river flows (m3/s) and indicative return periods (years): Storm Desmond

River Station Rank Years of record Peak flow (m3/s) Peak as ratio of QMED Indicative return period 
North-west       
 Eamont Pooley Bridge 46 268 4.45 350 
 Petteril Harraby Green 47 110 3.26 175 
 Caldew Cummersdale 19 279 1.75 20 
 Eden Sheepmount 49 1680 3.09 300 
 Greta Low Briery 45 350 3.15 175 
 Derwent Ouse Bridge 48 450 4.36 500 
 Derwent Camerton/Seaton Mill 56 600 2.93 300 
 Kent Victoria Bridge 37 403 2.80 150 
 Leven Newby Bridge 77 224 3.09 150 
 Lune Caton 48 1740 2.39 100 
North-east       
 Tyne Bywell 61 1730 2.04 140 
 South Tyne Haydon Bridge 49 915 2.02 60 
 North Tyne Reaverhill 56 716 2.12 60 
River Station Rank Years of record Peak flow (m3/s) Peak as ratio of QMED Indicative return period 
North-west       
 Eamont Pooley Bridge 46 268 4.45 350 
 Petteril Harraby Green 47 110 3.26 175 
 Caldew Cummersdale 19 279 1.75 20 
 Eden Sheepmount 49 1680 3.09 300 
 Greta Low Briery 45 350 3.15 175 
 Derwent Ouse Bridge 48 450 4.36 500 
 Derwent Camerton/Seaton Mill 56 600 2.93 300 
 Kent Victoria Bridge 37 403 2.80 150 
 Leven Newby Bridge 77 224 3.09 150 
 Lune Caton 48 1740 2.39 100 
North-east       
 Tyne Bywell 61 1730 2.04 140 
 South Tyne Haydon Bridge 49 915 2.02 60 
 North Tyne Reaverhill 56 716 2.12 60 
Table 8

Peak river flows (m3/s) and indicative return periods (years): Christmas event

River Station Rank Years of record Peak flow (m3/s) Peak as ratio of QMED Indicative return period 
North-west       
 Yarrow Croston 43 79 2.34 150 
 Calder Whalley Weir 55 501 2.94 450 
 Roch Rochdale 23 93 2.13 80 
 Roch Blackford Bridge 67 192 2.67 150 
 Irwell Irwell Vale 20 200 2.30 70 
 Irwell Bury Ground 43 284 2.54 100 
 Irwell Manchester Racecourse 75 ∼500 1.86 50 
North-east       
 Nidd Hunsingore 52 239 1.99 180 
 Calder Mytholmroyd 28 276 3.26 100 + 
 Ouse Skelton 130 544 1.69 45 
 Aire Armley 45 350 2.49 200 
 Aire Lemonroyd Weir 22 266 1.72 40 
 Wharfe Tadcaster 21 480 2.23 100 
River Station Rank Years of record Peak flow (m3/s) Peak as ratio of QMED Indicative return period 
North-west       
 Yarrow Croston 43 79 2.34 150 
 Calder Whalley Weir 55 501 2.94 450 
 Roch Rochdale 23 93 2.13 80 
 Roch Blackford Bridge 67 192 2.67 150 
 Irwell Irwell Vale 20 200 2.30 70 
 Irwell Bury Ground 43 284 2.54 100 
 Irwell Manchester Racecourse 75 ∼500 1.86 50 
North-east       
 Nidd Hunsingore 52 239 1.99 180 
 Calder Mytholmroyd 28 276 3.26 100 + 
 Ouse Skelton 130 544 1.69 45 
 Aire Armley 45 350 2.49 200 
 Aire Lemonroyd Weir 22 266 1.72 40 
 Wharfe Tadcaster 21 480 2.23 100 

Most of the above flows have been derived using the observed levels and the rating equation, but Low Briery and Ouse Bridge have been adjusted based on hydraulic modelling. The above peak flow at Bywell of 1730 m3/s is a central estimate from an analysis which gave a range from 1670 to 1790 m3/s based on estimation of velocities from gauging during this event and floodplain survey. This flow is considered to rank second only to the great flood on the Tyne of 1771.

Pooley Bridge and Ouse Bridge are just downstream of Ullswater and Bassenthwaite, respectively. The high ratios of ‘Peak flow to QMED’ reflect the lesser attenuation by the lakes of such large flow volumes compared with smaller events such as QMED. The road bridge at Pooley Bridge (upstream of the gauging station), which had been built in 1764, collapsed on 6th December 2015. Lake levels and lake attenuation are discussed in a later section.

At Harraby Green and Camerton/Seaton Mill the station which was operational in 2009 was closed and replaced with a new station in December 2015. Flooding was more extensive near Harraby Green in December 2015 than in November 2009, suggesting that December 2015 was the larger event. The peak flow at Camerton in November 2009 was estimated by detailed hydraulic modelling as 720 m3/s ±10% and rounded to 700 m3/s, with uncertainties due to the changes in the channel during the event. The current estimate for the December 2015 event at the replacement station at Seaton Mill is around 600 m3/s, awaiting further investigation.

Annual maxima

Figure 6 shows the annual maxima series since 1960 at selected river gauging stations, including a number at the downstream end of their catchment. The November 2009 flood at Camerton was more than twice the next highest event in the 49 years of record since 1960. At Sheepmount, the January 2005 flood was 26% higher than the next highest event since 1967, and the December 2015 flood was more than 10% higher again. Marsh et al. (2016) include similar plots for three of these stations.

Figure 6

Annual maxima (water years) from 1960 at selected (a) Cumbrian, (b) Lancashire and (c) north-east river flow stations.

Figure 6

Annual maxima (water years) from 1960 at selected (a) Cumbrian, (b) Lancashire and (c) north-east river flow stations.

Trends in annual maxima

Some of the above stations show consistent upward trends, though without December 2015 the visual trend may be less convincing. Similar apparent trends occur at a number of upstream stations in the north-west.

Table 9 shows the results of a Mann-Kendall test for trend on the annual maxima (from the start of the record) for the stations shown in Figure 6, for all stations with a confidence greater than 90%. Three of the four Cumbrian stations have a confidence of 94% or more when 2015 is included. However, except for the Roch at Blackford Bridge, any trends at the stations in Lancashire, the North-east and Yorkshire are not statistically significant. At Bywell, the trend without December 2015 is decreasing, though the flow record is affected in the early years by rating problems associated with gravel (Archer et al. 2007) and from 1981 by Kielder Reservoir, which drains around 11% of the catchment to Bywell. The sample in Table 9 is small and further investigation of Cumbrian stations will be carried out.

Table 9

Mann-Kendall tests on all years of annual maxima flows, with and without 2015

River Station Start of record   Comment 
With 2015 (%) End 2014 (%) 
Kent Sedgwick 1968 99.9 99.8  
Eden Sheepmount 1966 94.7 88.3  
Leven Newby Br 1938 >99.9 99.9  
Derwent Camerton/Seaton Mill 1960 91.0   
Lune Caton 1968 92.6   
Calder Whalley Weir 1970    
Ribble Samlesbury 1961    
Irwell Bury Ground 1972    
Roch Blackford Bridge 1948 96.0 92.0  
South Tyne Haydon Bridge 1959    
Tyne Bywell 1955  95.2 Decreasing 
Nidd Hunsingore 1934    
Wharfe Wetherby Flint Mill 1936    
River Station Start of record   Comment 
With 2015 (%) End 2014 (%) 
Kent Sedgwick 1968 99.9 99.8  
Eden Sheepmount 1966 94.7 88.3  
Leven Newby Br 1938 >99.9 99.9  
Derwent Camerton/Seaton Mill 1960 91.0   
Lune Caton 1968 92.6   
Calder Whalley Weir 1970    
Ribble Samlesbury 1961    
Irwell Bury Ground 1972    
Roch Blackford Bridge 1948 96.0 92.0  
South Tyne Haydon Bridge 1959    
Tyne Bywell 1955  95.2 Decreasing 
Nidd Hunsingore 1934    
Wharfe Wetherby Flint Mill 1936    

Stations used for trend analysis should be consistent over the period of record analysed (i.e. no significant changes at the station) and the data itself should be of high quality. A classification of ‘OK for pooling’ in the NRFA Peak Flow dataset should be a prerequisite if flows are analysed, but that classification by itself may not be sufficient. The data at the stations above which have test values greater than 90% are considered to be of sufficient quality for the analyses. Quinn (2012) discusses the assumptions behind use of flood peak data – that the data are accurate, independent, homogeneous and stationary.

Discussion

With the wet ground conditions and high lake levels before the rainfall on 4th and 5th December, we would expect that river flow return periods would be higher (i.e. rarer) than rainfall return period estimates. However, Table 10 shows that at many catchments this is not the case. These results reflect, at least in part, the uncertainties in the rainfall and flow return period estimates.

Table 10

Rainfall and flow return periods in December 2015 for selected catchments

    Return period estimate (years)
 
River Location Rain FEH1999 Rain FEH2013 Flow 
Derwent Seaton Mill 200 230 500 
Leven Newby Bridge 470 1800 150 
Eden Sheepmount 140 210 300 
Eamont Pooley Bridge 420 620 350 
Lune Caton 140 320 100 
Calder Whalley 34 48 450 
Irwell Bury Ground 66 160 25 
    Return period estimate (years)
 
River Location Rain FEH1999 Rain FEH2013 Flow 
Derwent Seaton Mill 200 230 500 
Leven Newby Bridge 470 1800 150 
Eden Sheepmount 140 210 300 
Eamont Pooley Bridge 420 620 350 
Lune Caton 140 320 100 
Calder Whalley 34 48 450 
Irwell Bury Ground 66 160 25 

A feature of the December 2015 events is that with initially wet catchments and long duration storms, flood volumes and percentage runoffs were very high. Many volumetric percentage runoffs in Cumbria were around 90% – for example, the catchments of the Eamont at Udford, the Eden at Sheepmount, the Kent at Victoria Bridge, the Yarrow at Croston, the Ribble at Samlesbury, and the Irwell at Manchester Racecourse were in the range 84 to 94%. These values are at the top end of those normally found in an event rainfall-runoff analysis (e.g. Houghton-Carr 1999: Appendix A, Flood Event Analysis). Webster & Ashfaq (2003) in a study of 2328 events in the UK noted an increase in percentage runoff with event magnitude and the need for rainfall-runoff models to reflect the importance of this and of the effect of initial catchment wetness.

Effect of storm duration

The critical storm duration will vary considerably from catchment to catchment, increasing with factors such as catchment size and the effect of lakes and reservoirs. The attenuation in the large lakes (and the system of large lakes and reservoirs in the Derwent catchment) not only delays and lowers the peak flows but also smooths the outflow hydrographs so that lake outflows are relatively insensitive to changes in rainfall intensity during the event. Thus, the peak flow may be sensitive to the rainfall intensity over the critical storm duration at upstream locations such as Appleby but will be less sensitive to this at downstream lake-affected locations such as Cockermouth.

Summary

The peak flows were thus related to both exceptional rainfalls at orographically enhanced locations and very high percentage runoffs, giving in combination large volumes of water which filled the lakes and reservoirs.

LAKE LEVELS

Table 11 summarises the peak lake levels since the start of the gauged record at major lakes. Some of these levels are slightly different from those in Miller et al. (2013) due to a change in the surveyed datum. In the December 2015 event the lake level recorders on Windermere and Derwentwater continued to operate whereas at Bassenthwaite, Thirlmere, and Ullswater peak levels were surveyed after the event.

Table 11

Peak water levels in lakes and reservoirs (mAOD) at current gauging stations

Lake Station Record start Jan 2005 Oct 2008 Nov 2009 Dec 2015 
Bassenthwaite Castle How June 1999 71.25 71.01 72.52 72.70 
Derwentwater Lodore July 1995 77.21 77.13 77.78 77.88 
Thirlmere Thirlmere Reservoir Oct 1997 179.95 179.72 180.11 180.56 
Windermere Far Sawrey Feb 1968 41.78 42.11 42.83 42.75 
Ullswater Glenridding Nov 1961 147.01 146.68 147.70 148.23 
Lake Station Record start Jan 2005 Oct 2008 Nov 2009 Dec 2015 
Bassenthwaite Castle How June 1999 71.25 71.01 72.52 72.70 
Derwentwater Lodore July 1995 77.21 77.13 77.78 77.88 
Thirlmere Thirlmere Reservoir Oct 1997 179.95 179.72 180.11 180.56 
Windermere Far Sawrey Feb 1968 41.78 42.11 42.83 42.75 
Ullswater Glenridding Nov 1961 147.01 146.68 147.70 148.23 

At Bassenthwaite, Derwentwater, Ullswater and Thirlmere December 2015 is the highest event, then November 2009, and then January 2005. However, at Windermere, the order is November 2009, December 2015, October 2008, January 1982 (41.87 mAOD) and then January 2005.

These large lakes have a significant attenuation effect on most floods. However, in Storm Desmond the relatively high initial lake levels and large flood volumes will have caused the attenuation to be much less. Figure 7 gives an example of the principle (in the figure, independent events starting at the same initial lake level are plotted sequentially for clarity). In their review of the November 2009 floods, Miller et al. (2013) drew attention to the reduced attenuation effect of lakes in larger events and recommended that further research be undertaken on lake attenuation and that flood frequency analysis of lake-influenced catchments be reviewed.

Figure 7

Reduced attenuation in Bassenthwaite as flood volume increases (inflow solid lines, outflow dashed lines).

Figure 7

Reduced attenuation in Bassenthwaite as flood volume increases (inflow solid lines, outflow dashed lines).

Derwentwater and Bassenthwaite

Hudleston (1935) describes the flood event in 1932, lake attenuation, flooding mechanisms, and historical information, with a particular interest in the Derwent catchment. He refers to a record of lake levels kept by the Marshall family at Derwent Isle on Derwentwater from 1869. The largest reported event in this record, on 28th October 1888, touched 8 ft (ca. 2.44 m). Other information in the paper suggests that this equates to a level of 77.11 mAOD, which was similar to January 2005 and October 2008. The flow gauging station at Ouse Bridge immediately downstream of Bassenthwaite was opened in 1968, and the largest events are in 2009 and 2015. It thus appears that December 2015 is the highest level on Derwentwater since 1869. Uncertainties include the lack of gauged information between 1935 and 1968, and any difference in level between Derwent Isle and the present Derwentwater gauge at Lodore.

Derwentwater and Bassenthwaite each cover an area of around 6.8 km2, but during flood events (e.g. in the most recent years 2008, 2009 and 2015) they appear to merge and the expanse of water increases to around 22.4 km2. However, the lakes do not act as one, because the lake levels are different – by about 5.2 m between Lodore (Derwentwater) and Castle How (Bassenthwaite) in November 2009 and December 2015. Detailed hydraulic modelling shows that the operation of the Derwentwater/Bassenthwaite system is complex. One control is the road between Keswick and Braithwaite; Hudleston (1935) noted this road as acting as a weir over a length of around 300 m, and the main (A66) road was subsequently raised in the 1970s.

Windermere

The three highest events at the gauging station at Newby Bridge (downstream of Windermere) from 1939 are in October 2008, November 2009, and December 2015, and have estimated peak flows of 145, 239, and 224 m3/s, respectively. Clark (2003) estimated the peak outflow from the event in 1898 as 132 m3/s, and hydraulic modelling to date suggests that this is reasonable when compared with a level on Backbarrow Bridge downstream. From the Chronology of British Hydrological Events (Black & Law 2004): 5th December 1864 – the highest in the last 50 years, the lake rose more than 8 ft, with no damage worth naming; 2nd November 1898 – 1 ft (ca. 30 cm) higher than previously recorded.

An event in July and August 1938 was reviewed by McClean (1940). A record of peak level data from Wray Castle from 1934 has been supplied by the Freshwater Biological Association. A table of annual maxima lake levels from 1934 to the present, compiled from various sources by Peter Spencer and Martin Wilson of the Environment Agency, suggests that November 2009, December 2015, October 2008 and November 1898 (in that order) are the highest lake levels since 1864 or 1814 (50 years from 1864).

Ullswater

The destruction of the road bridge at Pooley Bridge at the outlet to Ullswater suggests that this was the largest event since the bridge was built in 1764.

Summary of lake history

In summary, the available evidence suggests that November 2009 and December 2015 were the highest known events, including the historical record, on Ullswater, Windermere, Derwentwater and Bassenthwaite.

HYDROLOGIC AND HYDRAULIC MODELLING

As part of the post-event work, the Environment Agency and its consultants carried out extensive hydrologic and hydraulic modelling. Table 12 lists the main studies carried out for north-west England. These modelling studies feed into the Cumbria Flood Action Plan and investigations of options at specific locations. Two examples are described in more detail.

Table 12

Summary of hydrologic and hydraulic modelling in north-west England

Title Numbers Details 
Event verification 19 modelling studies Keswick, Cockermouth, Carlisle, Grasmere, Ambleside, Windermere, Appleby, Eamont Bridge, Pooley Bridge, Lancaster, Garstang, St Michaels, Whalley, Padiham, Ribchester, Croston, Wigan, Rochdale, Salford 
Event hydrology reviews 16 catchments  
Scheme performance reviews 4 major flood defence schemes Carlisle, Cockermouth, Keswick, Salford 
National Flood Resilience Review Carlisle Extreme event modelling 
Reservoir Operation Assessment 3 reservoirs Potential operation to improve flood defence 
‘Quick-wins’ 6 modelling studies Feasibility and optioneering 
Flood forecasting models  Models generally performed well. Selected models were reviewed and updated 
Various  Natural Flood Management assessments
Flood volume calculations
Effect of gravel on conveyance (7 reaches) 
Appraisal of potential flood defence schemes  For every potential scheme 
Title Numbers Details 
Event verification 19 modelling studies Keswick, Cockermouth, Carlisle, Grasmere, Ambleside, Windermere, Appleby, Eamont Bridge, Pooley Bridge, Lancaster, Garstang, St Michaels, Whalley, Padiham, Ribchester, Croston, Wigan, Rochdale, Salford 
Event hydrology reviews 16 catchments  
Scheme performance reviews 4 major flood defence schemes Carlisle, Cockermouth, Keswick, Salford 
National Flood Resilience Review Carlisle Extreme event modelling 
Reservoir Operation Assessment 3 reservoirs Potential operation to improve flood defence 
‘Quick-wins’ 6 modelling studies Feasibility and optioneering 
Flood forecasting models  Models generally performed well. Selected models were reviewed and updated 
Various  Natural Flood Management assessments
Flood volume calculations
Effect of gravel on conveyance (7 reaches) 
Appraisal of potential flood defence schemes  For every potential scheme 

Event verification

Event verification has been a critical activity to provide confidence in event hydrology and hydraulics. Important components of such modelling are good quality observed datasets and hence the need to review data consistency and stage-discharge relationships for events beyond their calibration range.

Example – model verification of the Derwent catchment

Low Briery is the key gauging station upstream of Keswick on the River Greta. The river reach at and upstream of Low Briery suffered serious erosion and deposition on the 4th and 5th December 2015. The high river flows undermined the side slopes such that much sediment and many trees fell into the river. The river section at Low Briery gauging station suffered considerable gravel deposition, raising the bed by more than 0.5 m. The existing rating gave a peak flow of 491 m3/s, but when this was applied to the hydraulic model the modelled flooding at Keswick far exceeded that recorded. A flow of 343 m3/s at Low Briery was necessary to replicate the actual flood extent. The existing rating curve was unreliable because it was extrapolated from the highest flow gauging of only 152 m3/s, and because of the changes in bed cross-section during the event.

The National Flood Resilience Review

As a result of the winter floods, HM Government set up a review to assess how the country can be better protected from future flooding and extreme weather events (HM Government 2016). For the Environment Agency, this required running four river and two tidal models under extreme weather conditions. The first two studies were of Carlisle and the Calder Valley in the north-west and north-east, respectively. An initial scoping study for Carlisle was carried out using fast and stable flood forecasting models with an initial estimate of likely rainfalls.

Based on many runs of their high resolution global climate model the Met Office concluded that winter monthly rainfall totals over the next 10 years could plausibly be 20% higher than Storm Desmond for Carlisle and the Christmas event for the Calder Valley. The Met Office then supplied the Environment Agency with new detailed gridded rainfalls based on these recent events with the 20% uplifts. These rainfall inputs were used to generate inflow hydrographs to the Environment Agency's detailed river and floodplain models of the present situation, and these models were then run to predict the flooding associated with these extreme rainfall scenarios. The results showed that the flooding remained within the areas and depths defined by the current Environment Agency's Extreme Flood Outline. This enabled HM Government and the asset owners to review the resilience of key local infrastructure assets such as energy, water, health, transport and telecommunications within the Extreme Flood Outline.

ARE CURRENT FLOOD ESTIMATION METHODS ADEQUATE?

Most current UK flood estimation is based on the Flood Estimation Handbook (FEH) (Institute of Hydrology 1999), which whilst retaining many of the basic approaches in the Flood Studies Report (FSR) (NERC 1975) made important improvements, particularly in flood frequency analysis and rainfall DDF relationships. The FEH has been updated by several major research projects (Kjeldsen 2007; Kjeldsen et al. 2008), and subsequent papers.

After the 2005 and 2009 floods, flood alleviation schemes were built at Carlisle, Keswick and Cockermouth, with quoted standards of protection against floods with return periods of 200, 75, and 100 years, respectively. Detailed flood estimates had been made using the FEH methods with the updates current at the time. All three schemes were overtopped in Storm Desmond.

Statistically, the chance of two rare events occurring only a few years apart is remote. For example, considering the design standard of 1 in 100 years for the scheme at Cockermouth and applying the Poisson distribution, there is a 9% chance of two 100-year events occurring in the period of record of 56 years at Camerton/Seaton Mill, but the chance of two such events within 7 years falls to 0.23%.

Such large events occurring close together may be just chance – the improbable has happened. Alternatively, perhaps the standard methods do not apply to these large events, or perhaps weather systems and rainfall have changed.

Any flood estimate for rare events has considerable uncertainty. Table 13, from Miller et al. (2013), demonstrates the uncertainties in flood frequency estimation for such extreme events, and the differences in results using the FEH statistical method with and without the November 2009 flood. In Table 4 the return period of the catchment rainfall over 36 hours is 485 years.

Table 13

Return period flow estimates for the November 2009 event at Pooley Bridge

 Return period (years) using data to 2008
 
Return period (years) using data to 2009
 
Method Return period 95% confidence Return period 95% confidence 
Single site >50,000 12,767 to >50,000 280 39 to >50,000 
Pooled (Enhanced single site) 5,877 1,066 to >50,000 460 122 to 4,289 
 Return period (years) using data to 2008
 
Return period (years) using data to 2009
 
Method Return period 95% confidence Return period 95% confidence 
Single site >50,000 12,767 to >50,000 280 39 to >50,000 
Pooled (Enhanced single site) 5,877 1,066 to >50,000 460 122 to 4,289 

Although it is well understood that flood estimates are uncertain, how we allow for the uncertainty is less clear. In standard applications of scheme design, values will be based on the best estimate, with a freeboard or uncertainty allowance which includes a range of sources of uncertainty.

After Storm Desmond, Peter Spencer expressed a concern as to whether flood estimation methods are adequate and asked the following questions: Should FEH methods be applied differently? Should we make more use of historical data? Should we make more use of continuous simulation? Are there long-term trends or short-term ‘clustering’? Are we seeing the effects of climate change? Are very large floods different? Are flood responses in Cumbria different?

ISSUES ARISING IN FLOOD ESTIMATION

Should FEH methods be applied differently?

There has been considerable discussion about the application of FEH methods over the years. Kjeldsen (2013, 2015) outlines some feedback about FEH methods and discusses some possible subjects for further research. Some hydrologists feel that there has always been a dichotomy between the use of hydrological judgement and a straightforward application of set procedures. Whatever the complexities, nuances, and comprehensiveness of the FEH and other publications, set procedures are attractive when time, budget or skills are short. The paragraphs below discuss some ideas on how the application of the FEH statistical methods might be improved.

The FEH statistical method

The FEH statistical method has two key components – estimation of an Index Flood, QMED, and derivation of growth factors to higher return periods based on a Pooling-group of hydrologically similar catchments.

Some aspects of the FEH statistical and rainfall-runoff methods can be based on general relationships (e.g. between QMED and catchment descriptors). Such general relationships can produce poor results in some locations. To quote a question from Tim Hunt (Environment Agency) ‘You keep saying the FEH methods work reasonably well across the UK, but how come they rarely work for my smaller ungauged catchments which make up the bulk of the flood estimation workload?’

One way to improve flood estimates is by use of gauged data. The HiFlows-UK project (Spencer et al. 2004a, 2004b), which has now become the NRFA Peak Flow Dataset, was set up in 2001 to establish an accurate dataset for use in UK flood estimation. Although the quality of the data has improved considerably, work still continues to improve both the data and the metadata. An important aspect is the metadata which enables users to form their own judgement about the applicability of a station for their particular use. More small catchments (both rural and urban) with reasonable peak flows would enable gauged data to be applied more widely. Some stations are operated as ‘levels-only’, without flows. Given the constraints on resources, one practical approach to improve the peak flow estimates for such catchments would be to target them for limited spot flow gauging and use small in-bank hydraulic models to estimate at least QMED.

Index flood, QMED

Error in the estimate of QMED is probably the most important source of error in flood estimation. QMED estimates based on catchment descriptors have considerable uncertainty, the equation having a structural factorial standard error (fse) of 1.431 (Kjeldsen et al. 2008) and a sample fse of 1.47 with an updated dataset (Kjeldsen 2013, 2015). Vesuviano et al. (2016) note that any (hydrological) method is likely to underestimate QMED in extremely urbanised catchments (URBEXT ≥ 0.6). The principle of improving flood estimates at ungauged sites by using the transfer of flood data from similar but gauged catchments is fundamentally sound. The practicalities are in the selection of the donor catchment(s) (e.g. which hydrological features to use?) and possible trade-offs between catchment similarity and data quality. There is some diversity of opinion on the selection of donors. Different practitioners may: consider similarity is more important than proximity, others vice versa; select donor sites for QMED based on published pooling-group criteria, which do not include BFIHOST (an estimate of Base Flow Index from soil types), whereas others regard similarity of BFIHOST as important; or use several donors, perhaps up to six (Kjeldsen et al. 2014), whilst others prefer one or two carefully selected stations.

If a short period of record is used to derive QMED but there is a neighbouring or similar catchment with a longer record, following the Robson & Reed (1999) and FEH Guidelines (2015), that longer record will be used to adjust the QMED estimate based on the shorter record. A section below discusses non-stationarity. If the data are not stationary it is suggested that it would be better to place more weight on recent years and not make such an adjustment.

Pooling groups

Pooling groups are used to derive a growth curve to extrapolate from the Index flood (QMED) to the desired return periods. The stations in a pooling-group are meant to be hydrologically similar to the subject site and are selected based on the catchment descriptors of area (AREA), standard average rainfall (SAAR), the lake and reservoirs index (FARL), and floodplain extent (FPEXT), with different weights applied to each descriptor (Kjeldsen et al. 2008). With the weights and the range of descriptor values, stations tend to be selected mainly on AREA and SAAR. Catchments in such a pooling-group can have very different values of flood characteristics which some practitioners might think important (e.g. QMED or QMED/AREA) and very different percentage runoffs (as represented by, for instance, SPRHOST and BFIHOST). The results are often seen in a wide scatter of the individual growth curves of all the stations in a pooling-group. This is particularly the case for small catchments where, because of scarcity of small catchments in the NRFA Peak Flow Dataset, the same stations keep being selected. Although it would seem logical to exclude from the pooling-group all stations with very different values of QMED/Area and BFIHOST to the subject site, such an approach is not standard practice.

The guideline that a pooling-group should have more than 500 station years is commonly rigidly adhered to, yet the graph on which this is based (Kjeldsen et al. 2008, Figure 6.3) suggests that fewer station years (perhaps as low as 300) are generally sufficient. It may be better to have fewer stations of good quality data and which are genuinely hydrologically similar to the subject site rather than to insist on increasing the number of station-years to 500.

A standard pooling-group has been composed of only catchments which are less than 3% urban, and adjustment for urbanisation at the subject site is then done subsequently. Particularly when the analyst wishes to include a station in its own pooling group (the Enhanced Single Site method), it might be better to include catchments with more urbanisation, and version 4 of WinFAP-FEH now allows analysts to do this. Further, if urbanisation is important, why not include all urbanised catchments as candidates for the pooling group, with similarity of URBEXT as one of the selection criteria?

If local stations are used more in a pooling group the spatial correlation will be increased, and this may mean that the growth curve is based to an increasing extent on the same flood events, which are not independent (e.g. Quinn 2012).

Confidence limits

Figure 8 of the Eden at Sheepmount in Carlisle shows a single-site flood frequency plot (not including December 2015) from WinFAP-FEH v4 with 95% confidence limits. The upper confidence limits do not take into account any physical constraints, such as whether the flow is more than the Probable Maximum Flood or the ability of the river to convey such flows to the target site.

Figure 8

Single-site flood frequency analysis for Sheepmount.

Figure 8

Single-site flood frequency analysis for Sheepmount.

Should we make more use of historical data?

Historical data are generally taken to refer to any data before the gauged record. They can include newspaper reports, books, flood marks, photographs, and anecdotal information. Importantly, in some cases historical information may suggest that recent apparently exceptional events have been matched in the past. However, changes over time and the scarcity of information may make the results from such an approach less certain than desired.

The recent Environment Agency R&D project, which is referred to as ‘FEH Local’, (Technical Guidance, Prosdocimi et al. 2016; Environment Agency 2017) addresses the use of local information such as historical data and paleoflood techniques to reduce uncertainty.

Version 4 of WinFAP-FEH enables historical data to be used in a single-site flood frequency analysis. At present, it is not able to include historical data within a pooling group, and more investigation of how this could be done would be required.

The FEH Local Technical Guidance refers to the now often-quoted idea that many gauged records are dominated by what are now thought to be a sequence of relatively flood-poor decades in the 1960s–1990s. The period from 1998 to 2015 has seen more severe flooding in many UK catchments, but historical and paleoflood evidence often shows that these recent large floods are not unprecedented. The technical guidance suggests that inclusion of historical data tends to increase flows compared with use of the gauged record alone. Black & Fadipe (2009) found that estimated 100-year flood flows at three out of four sites increased by more than 50% as a result of incorporating reliable historical information. Archer (2010) found that in a study of twelve catchments in north-east England historical flood estimates were greater than from a pooling group in all cases. Archer et al. (2016), studying chronologies of flash floods in north-east and south-west England, noted strong natural variability. The second half of the 20th century showed the lowest frequency of such events, so that they considered that using this period to project future flood risk is likely to result in serious underestimation. Figure 9, taken from the FEH Local project, shows a significant increase in flow estimates by including historical data compared with the pooling-group results.

Figure 9

Wear at Durham: use of historical data in flood frequency analysis.

Figure 9

Wear at Durham: use of historical data in flood frequency analysis.

Macdonald (2014) considered that the current ‘flood-rich’ period (from 2000) is not exceptional and that several comparable periods of increased flooding can be identified historically, although the data used appear not to include the more recent high flows. Macdonald & Black (2010) considered that once very large periods are considered (250 years) climatic variability becomes inescapable, and the inclusion of flood-rich and flood-poor periods leads to more robust flood frequency estimates.

Inclusion of historical data may reduce flood estimates. Macdonald & Black (2010) used documentary records for York dating back to 1263 AD which lead to a suggested 20% reduction in the 100-year peak from the FEH pooled estimate.

Example of historical data – Carlisle

After the flooding in Carlisle in January 2005, the resulting scheme designs were based on standard FEH methods and on the event, which was assigned a return period of around 180–200 years. This January 2005 event was estimated as having been around a metre above the highest event in the historical record for Carlisle, and at Sheepmount the December 2015 event was around 0.4 m above that in 2005.

After the January 2005 event, to put that flood into context, Peter Spencer, with guidance from David Archer and from Jeff Davison at the Environment Agency's Penrith office, put together a historical level series from 1771, which was included as Appendix 5 of the Environment Agency's Cumbria Floods Technical Report (2006). A number of changes will have affected the stage/flow relationship over the years. The most important of these was of the building of Eden Bridge and the associated channel changes in 1815, and the construction of the West Coast Main (railway) line with the embankment across the floodplain in 1848. There have also been numerous improvements to the flood defences, and dredging after the Second World War probably reduced the effect of the flood in 1954. However, in broad terms the river may be regarded as reasonably consistent since 1848. Figure 10 shows estimated peak levels from 1848 to December 2015.

Figure 10

Approximate flood levels at Eden Bridge, Carlisle, from 1848.

Figure 10

Approximate flood levels at Eden Bridge, Carlisle, from 1848.

This historical record is perhaps unusually complete because the council maintained a staff gauge on the downstream side of Eden Bridge from 1850 to the 1930s, and early newspaper reports are detailed. The historical stage record is an amalgam of the Council's staff gauge, newspaper and other reports, and the Environment Agency's gauging station at Sheepmount. Hydraulic modelling was used to establish a relationship between the river levels at Eden Bridge and Sheepmount, though the River Caldew joins the Eden in between, and these were then converted to flows using the rating at Sheepmount.

The derived flows were analysed using the methods to apply historical data in flood frequency estimation described by Bayliss & Reed (2001). The analyses were repeated with different timespans and a range of additional flows to make some allowance for the changes over the years. The range of results suggested an average estimate of the return period of the January 2005 flood of around 300 years. Comparing these results to the scheme design calculations suggested that the scheme design was, if anything, conservative. Re-doing the Bayliss and Reed method after adding the December 2015 flood gives a return period for the January 2005 flood of around 150 years, half the previous estimate.

In a standard application of the FEH statistical method to the Sheepmount gauge in Carlisle, Sheepmount has the steepest growth curve of all the stations in the pooling-group. Single site analysis therefore gives significantly different results from the pooling group; the return periods of the December 2015 flood by the Single site, Single site + historical, and Pooling group (Enhanced single site, which includes Sheepmount) are 100, 130, and 430 years, respectively.

Parkes & Demeritt (2016) applied a Bayesian approach to reduce the uncertainty in flood estimates for the Eden at Carlisle incorporating historical information. The paper describes the history of the River Eden at Carlisle in useful detail. They started with the same base historical record at Carlisle as applied above. 2D hydraulic models were built for various historical conditions and these results were applied to adjust peak flows for the changes over the years – this is a more sophisticated modelling approach to estimating flow in the early years than was applied in the analyses described above. Incorporating historical data into the Bayesian model reduced the spread of the 95% confidence limits by half for the 75-year and 100-year floods, compared with standard methods using solely gauged data. The Bayesian method by itself had little effect on the width when applied to just single-site gauged data. The Bayesian software used was considered suitable for academic rather than commercial use.

Thus, in this example where historical peak flows appear to have been less than those recently recorded, incorporation of the historical data produces lower peak flow estimates than standard FEH estimates. This shows the large uncertainties of any flood estimate, even incorporating historical data.

Paleoflood data

The FEH Local project includes the recent interest in Paleoflood data. Work is ongoing using Cumbrian lake sediments as proxy records for paleofloods. Preliminary results for Cumbria (Chiverrell et al. 2016 reported by the BBC 9 September; Liverpool University 2016), suggest from sediment cores at Brotherswater that the November 2009 and December 2015 floods were the most extreme in 600 years, and that two-thirds of the biggest floods in the Bassenthwaite sediment record have occurred in the past 20 years.

Should we make more use of continuous simulation?

Continuous simulation models have been used to varying degrees in a number of practical studies, and can have advantages over standard FEH-type methods. Lamb et al. (2016) set out the desired advantages of continuous simulation and describe four practical applications. The method is not commonly used because of the time and expense, because the methods still require calibration, and because applying results may not be straightforward. For example, the derived long-term rainfall sequence requires calibration, yet you would expect the results of such a calibration in Cumbria done after December 2015 would give very different results to one done before November 2009. Nonetheless, such methods may be worth the extra effort where an understanding of complex processes within a catchment will produce more confidence in the results.

Are there long-term trends or short-term ‘clustering’?

Rainfall

Recent studies suggest changes in upland rainfall in northern England. Burt & Ferranti (2010) studied nine raingauges with long-term records in an east-west line across northern England. They found large increases in winter rainfall in the 1980s and 1990s at all the upland sites, but not the two lowland sites. Summer rainfall had declined in recent decades. Barker et al. (2004) created a 200-year monthly rainfall index from 1788 for the Central English Lake District (CELD) based at Grasmere. They found a significant decline in summer rainfall since the 1960s, offset by increased precipitation in winter and spring. Wilby & Barker (2016) relate December 2015 rainfalls to the CELD index, and noted that December 2015 was the wettest December and November 2015 was the fourth wettest November in the series. November 2009 was the wettest month on record, with 731 mm compared with December 2015 of 720 mm. Malby et al. (2007) analysed rainfall patterns in the Lake District and identified the intensification of winter rainfall with altitude and increasing winter overspill precipitation to the region's rainshadow (for example, increasing penetration of rainfall into the Eden Valley). Jenkins et al. (2009), in their review of recent trends in the UK climate, include detailed maps of changes in rainfall which show a 10–15% increase in winter precipitation between the 1961–1990 and 1971–2000 periods over much of Cumbria and a 5–10% increase over much of the rest of north-west England.

Flood peaks

Long-term changes in peak flows might be caused by climate change, urbanisation, changes in land use or soil or land drainage, or by man-made changes such as reservoirs and flood alleviation schemes.

The graph in Figure 6(a) and the Mann-Kendall tests in Table 9 show a clear increasing trend at the Cumbrian river stations. Hannaford & Marsh (2008), using a network of near-natural catchments in the UK, found increasing trends in high flow data in maritime upland northern and western areas. The dataset ended in water year 2003/04. Pattison & Lane (2012), found no statistically significant trends in the Sheepmount record from 1770.

In recent years there has been considerable academic interest in non-stationarity (e.g. Kjeldsen et al. 2012; Prosdocimi et al. 2014, 2015; Serinaldi & Kilsby 2015; Sraj et al. 2016). Kjeldsen et al. (2012), using the best 388 stations in the HiFlows-UK dataset to water year 2009/10, concluded that 20% had a trend at 5% significance, and that most of the trends were positive, and were located in the north and west of the UK. Prosdocimi et al. (2014) mapped the decadal magnification factor M10, where a value of more than 1 indicates that, for a certain probability, the magnitude of peak flows occurring is increasing. M10 showed no significant effects for most of England and Wales, but in Cumbria and parts of the north-east there were many stations where M10 suggested an increase.

The historical record at Carlisle suggests some clustering, in that of the 31 events in the Cumbria Floods Technical Report (Environment Agency 2006), 12 occurred in two periods (1851–1857 and 1924–1931) totalling 15 years. There are relatively few specific publications on clustering. Merz et al. (2016), in a detailed study of 68 river flow stations in Germany concluded that all the methods he applied suggested the presence of clustering at a high fraction of the gauges, but that although such clustering was very pronounced for smaller events there was no evidence for larger floods. Ongoing but as yet mainly unpublished work includes Towe 2014 and others, with support from the JBA Trust, who looked into relative risk; preliminary results suggested that extreme results can cluster; for example, if a 10-year event has occurred, the chance of a 50-year event occurring within the next year might double.

Hurst (1951) created an interest in long-term hydrological persistence, or memory, such that wet years cluster into multi-year wet periods (and dry years similarly). There have been many attempts at explanation for the effect, such as those summarised by Koutsoyiannis (2005). Franzke et al. (2015) developed a non-stationary model which can show both persistent and switching regimes. They related the presence of the Hurst effect in the atmosphere to the North Atlantic Jet Stream, noting that persistent jet states are self-maintaining.

Example – Newby Bridge

Newby Bridge at the outlet of Windermere has the longest flood peak record (from 1939) in Cumbria. The flow record at Newby Bridge is compiled from three main stations, and therefore has some uncertainty; however, the time overlap between the stations and comparison with lake levels supports the reliability of the flow data.

A generalised extreme value distribution was fitted under stationary and non-stationary assumptions, and the best fit was found by allowing both the location and scale parameters to vary linearly with time as a covariate. The distribution was fitted using the generalised maximum likelihood estimation method of El Adlouni et al. (2007). The results are summarised in Table 14 and Figure 11. The location and scale parameters both increase strikingly over the period of record, and the non-stationary flood frequency analysis yields flood estimates that are about 15–25% higher than those from a stationary analysis.

Table 14

Newby Bridge: summary results from stationary and non-stationary flood frequency analyses

 % Increase over period
 
Estimate at water year 2015/16
 
Analysis In QMED In 100-year QMED (m3/s) 100-year (m3/s) 
Stationary – – 71 213 
Non-stationary (Location and scale) 44 98 86 248 
 % Increase over period
 
Estimate at water year 2015/16
 
Analysis In QMED In 100-year QMED (m3/s) 100-year (m3/s) 
Stationary – – 71 213 
Non-stationary (Location and scale) 44 98 86 248 
Figure 11

Newby Bridge: stationary and non-stationary flood frequency analyses.

Figure 11

Newby Bridge: stationary and non-stationary flood frequency analyses.

The results of this analysis are subject to considerable uncertainty, with the non-stationary analysis having a wider confidence interval associated with the larger number of parameters to be estimated. An implication is that design floods at Newby Bridge could be underestimated by up to 23% (depending on return period) if conventional methods of flood frequency analysis are applied. The implied return period for the flood on Windermere following Storm Desmond drops from 120 years under an assumption of stationarity to 60 years from the non-stationary analysis. The above analysis does not reveal whether and how the trend might continue

In summary, there appears to be considerable support for the possibility of increasing trend in river flows, or recent clustering. If river flows are increasing, reliance on short gauged records or a long-term historical record may underestimate future flood peaks and lead to under design. Also, non-stationary flood data would question the underlying assumptions of most statistical methods that probabilities are static.

Are we seeing the effects of climate change?

An obvious question from the December 2015 floods in Cumbria is whether recent years have showed the first effects of climate change in increasing the magnitude and frequency of winter floods. Schaller et al. (2016), in modelling the storms affecting southern England in the winter of 2013/14 considered that, as well as increasing the amount of moisture the atmosphere can hold, anthropogenic warming caused a small but significant increase in the number of January days with westerly flow, both of which increased extreme precipitation.

The attribution of any change is beyond the scope of this paper. However, there is specific advice as to what allowances should be made for climate change (Environment Agency 2016), so this is relevant to practical considerations for those involved in flood estimation. Much of the present advice on how to allow for climate change is based on analysis of historic records for a standard period (1961–1990) with modelling to derive factors for increases in future flood flows (Environment Agency 2016). However, if climate change has already affected the flow record used in the analysis, the advice may lead to double-counting of the effect of climate change and, if present non-stationary effects are not accounted for, projections of future climate change may have little effect on the promotability of flood alleviation schemes.

Are very large floods different?

FSR and FEH rainfall-runoff models are based on a constant unit hydrograph, which gives a linear response in speed and magnitude to the effective rainfall. In reality, runoff responses are not linear but are complex.

When the FSR and FEH rainfall-runoff methods are applied to Probable Maximum Flood estimation, it is recommended to reduce the Time to Peak by multiplying by 0.67, which increases the flood peak by 1.5. The reasons for this recommendation are partly pragmatic, because the FSR rainfall-runoff method under-predicted in tests on six notable events, and partly to ‘represent more rapid and intense response that is believed to occur under exceptional conditions’ (Houghton-Carr 1999; Section 4.2.1).

As discussed in the section above on Lake Levels, lakes and reservoirs (and also some floodplains) tend to absorb small events but have less attenuation effect as the flood volume increases – hence the results for Pooley Bridge and Ouse Bridge in Table 7. This effect is important to many locations in the Lake District.

Flash floods, mainly caused by short intense convective summer storms, are quite different to the long-duration 2005, 2009 and 2015 floods in Cumbria, but raise some of the same questions. Wass et al. (2008), describing a flash flood at Helmsley in North Yorkshire noted an unusually fast response to high intensity rainfall, with the possibility that FEH methods may underestimate the magnitude of extreme flows in more rapidly responding catchments. Archer & Fowler (2015) describe five events to illustrate examples of flash floods. They note the rapid rates of rise and the possibility of steep wavefronts with fast travel times. The report by Fenn et al. (2005) on the Boscastle flood of 2004 noted that in order to replicate the event the Time to Peak had to be reduced to 50% of that predicted by catchment descriptors, and a variable percentage runoff was applied, rising to 95%. Kjeldsen et al. (2016) analysed runoff from a small steep mountainous catchment in South Korea with typhoon rainfalls up to 571.8 mm in 90 hours. They found that watershed lag time decreased with rainfall intensity and runoff volume increased with antecedent soil moisture. Quinn (2010) applied the Richards-Baker Flashiness Index to 64 benchmark catchments. There was an increase in flashiness in winter in the north-west and parts of the south and Wales, though of the 64 catchments only 4 were in north-west England and 4 in north-east England.

In rainfall-runoff methods, the Time to Peak is often derived from the catchment descriptors and is rarely calibrated, though this can be done using stations which are ‘levels-only’; perhaps calibration of the Time to Peak should be more common.

Northern England has many small catchments which according to these examples may respond to intense rainfall events in a non-linear fashion. A further suggestion is that in Cumbria this effect may be reinforced by a non-linear response in lake outfalls. In summary, such observations support the ideas of non-linear response and hence that rare events may be larger than generated by standard methods.

Are flood responses in Cumbria different?

Compared with much of England, distinctive topographic features in Cumbria include steep hill slopes with thin vegetation, flat valleys and many lakes which attenuate flood response. Large rainfall depths can result from long-duration storms and orographic enhancement. Flood events can be preceded by many days of rainfall so that heavy storm rainfall may fall on already wet or saturated catchments, which may partly explain the high percentage runoffs in the December 2015 events. Several of the above effects may combine to cause large floods, so that general UK methods might give worse results in Cumbria than elsewhere.

CONCLUSIONS

The December 2015 events

The weather systems of Storm Desmond were similar to the events in January 2005 and November 2009, creating orographically enhanced rainfalls over the central Cumbrian hills which were not intense, but continued for around 36 hours. Of these three events, rainfalls for Storm Desmond were more widespread and generally heavier. Storm Desmond created new UK maximum rainfalls over 24 to 48 hours.

With exceptional rainfall depths falling on ground which was already wet or saturated due to rainfalls twice the average for the month in November, Storm Desmond caused many rivers to rise to flooding levels. At many locations Storm Desmond appears to have caused the largest known floods in both the gauged and known historical records. In the north-east, flooding was confined largely to the catchment of the Tyne, and although return periods were high, the event was not as exceptional as in Cumbria.

The Christmas 2015 event caused serious flooding at many locations in Lancashire and West Yorkshire, but the event was not as exceptional as Storm Desmond.

The Mann-Kendall test statistics in Table 9 indicate an increasing trend in annual maxima flows for the Cumbrian stations, in some cases even without the December 2015 event. With the exception of the Roch at Blackford Bridge, this increasing trend is not shown for the stations in Lancashire and the north-east. The sample size for each county in Table 9 is small, and further and more detailed investigations will be carried out.

Implications for flood estimation

Extreme events happen. It is therefore necessary for flood estimation methods to be sufficiently flexible that extreme effects can be included in the modelling.

This paper has discussed some of ways in which flood estimation can be made more flexible. Most of these accord with the FEH maxims (Reed 1999) which stress the importance of gauged data, data transfers, and search for (and therefore using) additional information. When extreme events occur such data may become more important, and the FEH Local project should help practitioners make more use of local and historical data. Detailed study of specific local and regional events may reveal important factors such as extreme rainfalls, high percentage runoffs, non-linearity in response and non-stationarity over time. Flood hydrology is sufficiently complicated that such factors may be masked in a general regression analysis of UK-wide data.

Trend and non-stationarity

Current flood estimation practice is based on the assumption of stationarity. The January 2005, November 2009, and December 2015 events suggest that a simple assumption of stationarity in Cumbria may be misleading. Where there is evidence that the gauged and historical data are not stationary, non-stationarity should be incorporated into the flood estimation process. However, considerations of non-stationarity in flood estimation are still in the research and investigation stage and there are no generally accepted practitioner guidelines. The possibility of clustering may be a further complication.

Local and historical data

More use should be made of local gauged data. Where a local or donor gauging station has been operated as ‘levels-only’, it will often be worth deriving even an approximate rating curve to estimate QMED.

Reliable historical data can give context on the variability of flood occurrence. Historical data can be particularly useful where past events are larger than those in the gauged record. Where recent events are larger than those in the past, an understanding of historical data may help improve the confidence in the flood estimate. WinFAP-FEH v4 allows historical data to be incorporated in single-site flood frequency estimates, though there are at present no standard methods to include historical data into pooling-group analyses.

Rainfall

A number of reports and papers have shown increases in winter rainfalls in north-west England. CEH Wallingford are currently investigating the effects of updating the dataset on the FEH 2013 DDF relationships.

Runoff response

The December 2015 floods occurred after preceding heavy rainfall, creating very wet or saturated ground conditions and very high percentage runoffs. Runoff may have been rapid, and lake attenuation less effective with such large flood volumes. Rainfall-runoff modelling for extreme events should therefore take into account the possibilities of such large rainfall and large volumes as occurred in December 2015 and possible non-linear response mechanisms. Where there is concern about rapid catchment response and there is a gauged level record, more efforts should be made to calibrate the Time to Peak.

Practical application

Future flood research and development may address some of the gaps in knowledge about how to allow for extreme floods. In the meantime, where flood estimates are important (such as for consideration of flood alleviation schemes), practical hydrologists should consider the questions raised by the December 2015 events and supplement standard FEH flood estimation methods by additional approaches such as those discussed in this paper.

ACKNOWLEDGEMENTS

Funding for these studies was provided by the Environment Agency. Original data has come from the Environment Agency, unless otherwise acknowledged; annual maxima data were based on the NRFA Peak Flow Dataset, updated to water year 2015/16 by the Environment Agency. The authors thank the three reviewers for their comments, which have led to improvements to the paper.

REFERENCES

REFERENCES
Archer
,
D. R.
2010
Applying historical information to flood risk assessment in north-east England
. In
British Hydrological Society, 3rd International Symposium
,
19–23 July
,
Newcastle
,
UK
.
Archer
,
D. R.
&
Fowler
,
H. J.
2015
Characterising flash flood response to intense rainfall and impacts using historical information and gauged data in Britain
.
J. Flood Risk Manag
.
doi: 10.1111/jfr3.12187
.
Archer
,
D. R.
,
Leesch
,
F.
&
Harwood
,
K.
2007
Assessment of severity of the extreme River Tyne flood in January 2005 using gauged and historical information
.
Hydrological Sciences Journal
52
,
992
1003
.
doi: 10.1623/hysj.52.5.992
.
Archer
,
D. R.
,
Parkin
,
G.
&
Fowler
,
H. J.
2016
Assessing long term flash flooding frequency using historical information
.
Hydrology Research
47
.
doi: 10.2166/nh.2016.031
.
Barker
,
P. A.
,
Wilby
,
R. L.
&
Borrows
,
J.
2004
A 200-year precipitation index for the central English Lake District
.
Hydrological Sciences Journal
49
,
769
785
.
Bayliss
,
A. C.
&
Reed
,
D. W.
2001
The use of Historical Data in Flood Frequency Estimation. Report to MAFF
.
CEH
,
Wallingford
,
UK
.
Black
,
A. R.
&
Fadipe
,
D.
2009
The use of historic water levels for re-assessing flood frequency
.
WEJ (Journal of CIWEM)
23
,
23
31
.
Black
,
A. R.
&
Law
,
F. M.
2004
Development and utilization of a national web-based chronology of hydrological events
.
Hydrological Sciences Journal
49
,
246
.
doi: 10.1623/hysj.49.2.237.34835
.
Burt
,
T. P.
&
Ferranti
,
E. J. S.
2010
Changing patterns of heavy rainfall in upland areas: a case study from northern England
.
Int. J. Climatol
.
doi: 10.1002/joc.2287
.
Chiverrell
,
R.
,
Sear
,
D.
,
Warburton
,
J.
&
Schillereff
,
D.
2016
reported by the BBC on 9 September. http://www.bbc.co.uk/news/science-environment-37304856 (accessed 24 November 2017)
.
El Adlouni
,
S.
,
Ouarda
,
T. B. M. J.
,
Zhang
,
X.
,
Roy
,
R.
&
Bobee
,
B.
2007
Generalized maximum likelihood estimators for the non-stationary generalized extreme value model
.
Water Resources Research
43
,
W03410
.
doi: 10.1029/2005WR004545
.
Environment Agency
2006
Cumbria Floods Technical Report. Factual Report on the Meteorology, Hydrology and Impacts of the January 2005 Flooding in Cumbria
.
EA
,
Bristol
,
UK
.
Environment Agency
2012
Cumbria Floods Technical Report. Factual Report on the Meteorology, Hydrology and Impacts of the November 2009 Flooding in Cumbria
.
EA
,
Bristol
,
UK
.
Environment Agency
2015
Flood Estimation Guidelines. Version 5. Operational Instruction 197_08
.
EA
,
Bristol
,
UK
.
Environment Agency
2016
Adapting to Climate Change: Advice for Flood and Coastal Erosion Risk Management Authorities. LIT 5707
.
EA
,
Bristol
,
UK
.
Environment Agency
2017
Using Local Data to Reduce Uncertainty in Flood Frequency Estimation. Technical Guidance 12_17
.
EA
,
Bristol
,
UK
.
Faulkner
,
D.
1999
Flood Estimation Handbook
,
Vol. 2, Rainfall
.
Institute of Hydrology
,
Wallingford, UK
.
FEH Web Service
2015
.
Fenn
,
C. R.
,
Bettess
,
R.
,
Golding
,
B.
,
Farquharson
,
F. A.
&
Wood
,
T.
2005
The Boscastle flood of 16 August 2004: Characteristics, causes and consequences
. In
Proc. of the 40th Defra Flood and Coastal Management Conference
,
5–7 July 2005, University of York, UK. HRPP 341
.
Franzke
,
C. L. E.
,
Osprey
,
S. M.
,
Davini
,
P.
&
Watkins
,
N. W. A.
2015
Dynamical systems explanation of the hurst effect and atmospheric low-frequency variability
.
Sci. Rep.
5
,
9068
.
doi: 10.1038/srep09068
.
Greater Manchester
2016
Flood Investigation Report 26 December 2015
.
EA
,
Bristol
,
UK
.
Hannaford
,
J.
&
Marsh
,
T. J.
2008
High-flow and flood trends in a network of undisturbed catchments in the UK
.
Int. J. Climatol.
28
,
1325
1338
.
doi 10.1002/joc.1643
.
HM Government
2016
(September) National Flood Resilience Review
.
HM Government
,
London
,
UK
.
Houghton-Carr
,
H.
1999
Flood Estimation Handbook
,
Volume 4
.
Centre for Ecology & Hydrology
,
Wallingford
,
UK
.
House of Commons
2016
Winter Floods 2015–2016 Briefing Paper Number CBP7427
.
HM Government
,
London
,
UK
.
Hudleston
,
F.
1935
The Floods of the Lake District (Part1)
.
Printed for private circulation by the River Derwent Catchment Board
.
Hurst
,
H. E.
1951
Long-term storage capacity of reservoirs
.
Trans. ASCE
116
,
770
.
Institute of Hydrology
1999
Flood Estimation Handbook
.
Centre for Ecology & Hydrology
,
Wallingford
,
UK
.
Jenkins
,
G. J.
,
Perry
,
M. C.
&
Prior
,
M. J.
2009
The Climate of the United Kingdom and Recent Trends
.
Met. Office Hadley Centre
,
Exeter
,
UK
.
Kendon
,
M.
,
McCarthy
,
M.
,
Jevrejeva
,
M.
&
Legg
,
T.
2016
State of the UK Climate 2015
.
Met. Office Hadley Centre
,
Exeter
,
UK
Kjeldsen
,
T. R.
2007
The Revitalised FSR/FEH Rainfall-Runoff Method. Flood Estimation Handbook. Supplementary Report No.1
.
Centre for Ecology & Hydrology
,
Wallingford
,
UK
.
Kjeldsen
,
T. R.
2013
The lessons learned from the FEH improvements
. In
Irish National Hydrology Conference
,
Tullamore
,
Ireland
. .
Kjeldsen
,
T. R.
2015
How reliable are design flood estimates in the UK?
J. Flood Risk Manage.
8
(
3
).
doi: 10.1111/jfr3.12090
.
Kjeldsen
,
T. R.
,
Jones
,
D. A.
&
Bayliss
,
A. C.
2008
Improving the FEH Statistical Procedures for Flood Frequency Estimation. Defra/Environment Agency Science Report SC050050
.
DEFRA, HM Government
,
London
,
UK
.
Kjeldsen
,
T. R.
,
Svensson
,
C.
&
Miller
,
J. D.
2012
Large-scale attribution of trend in UK flood flow data
. In
BHS 11th National Symposium
,
Dundee
.
doi: 10.7558/bhs.2012.ns29
.
Kjeldsen
,
T. R.
,
Jones
,
D. A.
&
Morris
,
D. G.
2014
Using multiple donor sites for enhanced flood estimation in ungauged catchments
.
Water Resources Research
50
,
6646
6657
.
Kjeldsen
,
T. R.
,
Kim
,
H.
,
Jang
,
C.-H.
&
Lee
,
H.
2016
Evidence and implications of nonlinear flood response in a small mountainous watershed
.
J. Hydrol. Eng.
21
(
8
).
doi:10.1061/(ASCE)HE.1943-5584.0001343
.
Koutsoyiannis
,
D.
2005
Hydrologic persistence and the Hurst phenomenon
. In:
Water Encyclopedia
,
Vol. 4
,
Surface and Agricultural Water
(
Lehr
,
J. H.
&
Keeley
,
J.
, eds).
Wiley
,
New York
, pp.
210
221
.
Lamb
,
R.
,
Faulkner
,
D.
,
Wass
,
P.
&
Cameron
,
D.
2016
Have applications of continuous rainfall–runoff simulation realized the vision for process-based flood frequency analysis?
Hydrol. Process.
30
,
2463
2481
.
doi: 10.1002/hyp.10882
.
Liverpool University
2016
Study to examine extreme floods in northern England. https://news.liverpool.ac.uk/2016/02/02/nerc-award-to-study-the-recent-extreme-floods-in-northern-england/ (accessed 6 March 2017)
.
Macdonald
,
N.
2014
Millennial scale variability in high magnitude flooding across Britain
.
Hydrol. Earth Syst. Discuss.
11
,
10157
10178
.
doi: 10.5194/hessd-11-10157-2014
.
Macdonald
,
N.
&
Black
,
A. R.
2010
Reassessment of flood frequency using historical information for the River Ouse at York
.
UK Hydrol. Sci. J.
55
,
1151
1162
.
Malby
,
A. R.
,
Whyatt
,
J. D.
,
Timmins
,
R. J.
,
Wilby
,
R. L.
&
Orr
,
H. G.
2007
Long-term implications in orographic rainfall: analysis and implications for upland catchments
.
Hydrol. Science. J.
52
,
276
291
.
doi: 10.1623/hysj.52.2.276
.
Marsh
,
T.
,
Kirkby
,
C.
,
Muchan
,
K.
,
Barker
,
L.
,
Henderson
,
E.
&
Hannaford
,
J.
2016
The Winter Floods of 2015/16 in the UK – A Review
.
CEH
,
Wallingford
,
UK
.
McClean
,
W. N.
1940
Windermere basin: rainfall, run-off and storage
.
Quarterly Journal of the Royal Meteorological Society
66
,
33
62
.
Merz
,
B.
,
Nguyen
,
V. D.
&
Vorogushyn
,
S.
2016
Temporal clustering of floods in Germany: do food-rich and flood-poor periods exist?
J. Hydrol.
541
,
824
838
.
doi: 10.1016/j.hydro.2016.07.041
.
Miller
,
J. D.
,
Kjeldsen
,
T. R.
,
Hannaford
,
J.
&
Morris
,
D. G.
2013
A hydrological assessment of the November 2009 floods in Cumbria, UK
.
Hydrology Research
44
,
180
197
.
doi: 10.2166/nh.2012.076
.
NERC
1975
Flood Studies Report 1975 Vols 1–5
.
NERC
,
Swindon
,
UK
.
News Outlets
(Date accessed 25/11/2017): Guardian 8 Dec 2015 ‘Minister says flooding is unprecedented event, as she defends spending’ https://www.theguardian.com/environment/2015/dec/08/minister-says-flooding-is-unprecedented-event-as-she-defends-spending
,
North-west Evening Mail 8 Dec 2015 ‘David Cameron to visit flood-hit Cumbria’ http://www.nwemail.co.uk/News/Barrow/David-Cameron-to-visit-flood-hit-Cumbria-0c19eaf2-ed38-4906-884c-6da2af5e2a25-ds
,
Huffington Post 27 Dec 2015 ‘David Cameron tells flooding victims he'll do ‘Whatever is needed,’ despite cutting flood defence spending in 2011’ http://www.huffingtonpost.co.uk/2015/12/27/david-cameron-flooding-victims_n_8880952.html
,
Independent 28 Dec 2015 ‘UK flooding: ‘This wouldn't be allowed to happen in the South’ - David Cameron accused of neglecting the North’ http://www.independent.co.uk/news/uk/politics/uk-flooding-this-wouldnt-be-allowed-to-happen-in-the-south-david-cameron-accused-of-neglecting-the-a6787961.html
.
Prosdocimi
,
I.
,
Kjeldsen
,
T. R.
&
Svensson
,
C.
2014
Non-stationarity in annual and seasonal series of peak flow and precipitation in the UK
.
Nat. Hazards Earth Syst. Sci.
14
,
125
1144
.
doi: 10.5194/nhess-14-1125-2014
.
Prosdocimi
,
I.
,
Kjeldsen
,
T. R.
&
Miller
,
J. D.
2015
Detection and attribution of urbanization effect on flood extremes using non-stationary flood-frequency models
.
Water Resources Research.
51
,
4244
4262
.
doi: 10.1002/2015WR017065
.
Prosdocimi
,
I.
,
Stewart
,
L.
,
Faulkner
,
D.
&
Mitchell
,
C.
2016
FEH: Local Improving flood estimates using historical data
. In
FLOODrisk 2016 – 3rd European Conference on Flood Risk Management
,
17–21 October 2016, Lyon, France. doi: 10.1051/e3sconf/20160701006
.
Quinn
,
N.
2010
Are trends in flashiness evident in natural catchments in the United Kingdom?
In
BHS Third International Symposium, 2010, University of Newcastle
,
19–23 July
,
Newcastle
,
UK
.
Quinn
,
N.
2012
Flood peak data and engineering design: a house of cards?
In
BHS 11th National Symposium
,
9–11 July 2012
,
Dundee
.
Reed
,
D. W.
1999
Flood Estimation Handbook
.
Volume 1
.
Institute of Hydrology
,
Wallingford
,
UK
.
Robson
,
A.
&
Reed
,
D.
1999
Flood Estimation Handbook
,
Volume 3
.
Institute of Hydrology
,
Wallingford
,
UK
.
Rodda
,
H. J. E.
,
Little
,
M. A.
,
Wood
,
R. G.
,
MacDougall
,
N.
&
McSharry
,
P. E.
2009
A digital archive of extreme rainfalls in the British Isles from 1866 to 1968 based on British Rainfall
.
Weather
64
,
71
75
.
doi: 10.1002/wea.354
.
Royal Academy of Engineering
2016
Living Without Electricity. One City's Experience of Copying with Loss of Power
.
Royal Academy of Engineering
,
London
,
UK
.
Schaller
,
N.
,
Kay
,
A. L.
,
Lamb
,
R.
,
Massey
,
N. R.
,
van Oldenborgh
,
G. J.
,
Otto
,
F. E. L.
,
Sparrow
,
S. N.
,
Vautard
,
R.
,
Yiou
,
P.
,
Ashpole
,
I.
,
Bowery
,
A.
,
Crooks
,
S. M.
,
Haustein
,
K.
,
Huntingford
,
C.
,
Ingram
,
W. J.
,
Jones
,
R. G.
,
Legg
,
T.
,
Miller
,
J.
,
Skeggs
,
J.
,
Wallom
,
D.
,
Weisheimer
,
A.
,
Wilson
,
S.
,
Stott
,
P. A.
&
Allen
,
M. R.
2016
Human influence on climate in the 2014 southern England winter floods and their impacts
.
Nature Climate Change
6
,
627
634
.
doi 10.1038/nclimate2927
.
Serinaldi
,
F.
&
Kilsby
,
C. G.
2015
Stationarity in undead: uncertainty dominates the distribution of extremes
.
Advances in Water Resources
77
,
17
36
.
doi 10.1016/j.advwatres.2014.12.013
.
Spencer
,
P. S.
,
Palmer
,
T.
&
Wass
,
P.
2004a
HiFlows-UK
. In
Defra Flood and Coastal Management Conference
,
York
,
Paper 09.7
.
Spencer
,
P.
,
Higginson
,
N.
,
Palmer
,
T.
&
Wass
,
P.
2004b
HiFlows-UK
. In
Irish National Hydrology Seminar
,
Tullamore
,
Ireland
.
Sraj
,
M.
,
Viglione
,
A.
,
Parajka
,
J.
&
Bloschl
,
G.
2016
The influence of non-stationarity in extreme hydrological events on flood frequency estimation
.
J. Hydrol. Hydromech.
64
.
doi 10.1515/johh-2016-0032
.
Stewart
,
E. J.
,
Morris
,
D. G.
,
Jones
,
D. A.
&
Gibson
,
H. S.
2012
Frequency analysis of extreme rainfall in Cumbria 16–20 November 2009
.
Hydrology Research
43
,
649
662
.
doi: 10.2166/nh.2012.033
.
Stewart
,
E. J.
,
Jones
,
D. A.
,
Svensson
,
C.
,
Morris
,
D. G.
,
Dempsey
,
P.
,
Dent
,
J. E.
,
Collier
,
C. G.
&
Anderson
,
C. W.
2013
Reservoir Safety – Long Return Period Rainfall. R&D Technical Report WS 194/2/39/TR in 2 Volumes
.
DEFRA, HM Government
,
London
,
UK
.
Towe
,
R.
2014
Modelling and estimating the clustering of floods. https://connect.innovateuk.org/documents/2906271/3749832/PA13-020_JBATrust_Lancaster_CaseStudy (accessed 24 November 2017)
.
Vesuviano
,
G.
,
Stewart
,
L.
,
Haxton
,
T.
,
Young
,
A.
,
Hunt
,
T.
,
Spencer
,
P.
&
Whitling
,
M.
2016
Reducing uncertainty in small-catchment flood peak estimation
. In
FLOODrisk 2016 – 3rd European Conference on Flood Risk Management
,
17–21 October
,
Lyon
,
France
.
doi 10.1051/e3sconf/20160701008
.
Wass
,
P.
,
Lindsay
,
D.
&
Faulkner
,
D. S.
2008
Flash flood! A lucky escape for 10,000 bikers
. In
10th British Hydrological National Symposium
,
September 2008
,
University of Exeter
.
Webster
,
P.
&
Ashfaq
,
A.
2003
Comparison of UK flood event characteristics with design guidelines
.
Proc. Institution of Civil Engineers (Water & Maritime Engineering)
156
,
33
40
.