Water and Me!

FEB 2011 Institute of Hazard, Risk and Resilience E. Raven

Flood clustering, insurance, and a bit of sediment mixed in! Emma RavenWillis Research Fellow in Hazard and Risk,IHRR, Durham University

February 21st 2011


Water and Me! Emma Waterhouse BSc / MSc: fluvial

studies / catchment dynamics

Fellowship: insurance / extreme floods / rainfall /

clustering

JBA: reinsurance / floods / cat modelling

Extra-curricular water interest

PhD: sediment / fluvial geomorphology /

flood risk

http://www.jbaconsulting.co.uk/

http://www.leeds.ac.uk/


Seminar OverviewPart 1: River Flow Clustering for UK Insurance Risk

Analysis• Reinsurance / Willis Research Network;• Background: flood-risk stats (stationary vs. trends vs. cycles);• Methodology: discharge rather than rainfall;• Characterising clusters: visual / statistical;• Potential atmospheric links / future research needs.

Part 2: Interactions Between Sediment, Engineering and Flood Risk in Gravel-Bed Rivers

• The importance of sediment for flood risk;• Background;• Results from fieldwork;• Overview of model;• Model scenarios.


Part 1: Characterising High River Flow Clustering in UK Rivers


Reinsurance

risk transfer

Clients

Willis are a global reinsurance intermediary – need to understand extreme risks

20,000 associates400 offices

100 countries

USD 11 billion in premiums &

USD 5 trillion of exposed global risk

protected every year

Willis Reinsurance Group

Insurers

Reinsurers


The WRN and Durham

Institute of Hazard, Risk and

ResilienceImprove knowledge and understanding of

extreme eventsMaking a difference to how we live with

hazard and risk

funding / steering

research applications

academic outputs: journals / teaching

methods to help clients identify and quantify their risk exposure

Core Flooding Questions• what factors drive floods (characteristics of rainfall patterns from summer 07)?• what problems do short-records cause for analysis? • do floods occur in decadal-length temporal clusters?• can we quantitatively characterise these clusters?


Stats and Short Term TrendsFlood risk analysis / management require quantitative statistics: Return Intervals / Probabilities.

Long-historical data sets are essential.

DATA: UK river flow data is now widely available (National River Archive) - predominantly post 1960.

River Severn at Bewdley (photo & data)

stationary trending cyclic


Problematic ProbabilitiesReturn Interval = years/rankProbability = 1/RI


Problematic ProbabilitiesReturn Interval = years/rankProbability = 1/RI

1990-2010: 19 years / 3 flows RI = 6.3 years Prob. = 16%

1980-2010: 29 years / 3 flowsRI = 9.6 years Prob. = 10.4%

1900-2010: 109 years / 20 flowsRI = 5.5 years Prob. = 18%

As record length increases, probability changes - influences decisions.


Discharge DataRiver Location

Catchment Area (km 2) Start year

Record length (years)

Avon Evesham 2210 1936 72Bedford Ouse Bedford 1460 1933 75

Chelmer Rushes Lock 533.9 1932 75Colne Berrygrove 352.2 1934 73Dee Woodend 1370 1929 79Dee Manley Hall 1019.3 1937 71

Derwent Yorkshire Bridge 126 1933 74Derwent St Marys 1054 1935 72

Elan Elan Village 184 1908 99Harpers Brook Old Mill Bridge 74.3 1938 69

Lee Feildes 1036 1879 129Leven Newby Brdige 247 1939 68

Nene/Brampton St Andrews 232.8 1939 68Nene/Kislingbury Upton 223 1939 68

Severn Bewdley 4325 1921 87Ter Crabbs Bridge 77.8 1932 75

Thames Kingston 9948 1883 124Thames Days Weir 3444.7 1938 69Vyrnwy Vyrnwy Reservoir 94.3 1920 87Weaver Ashbrook 622 1937 70

Whitewater Lodge Farm 44.6 1927 80Williow Brook Fotheringhay 89.6 1938 69

Wye Ddol 174 1937 70

• 22 longest flow records in UK;• created a Peak Over Threshold series for each;• RI of 1 in 1year, 2years, 4years;• 7-days between peaks.

Circle diameter = catchment size

POT eg: 100 year record: RI of 1year = top 100 flows; RI of 4years = top 25 flows


What about Rainfall Records?

Length of record: UK monthly average rainfall series (Met Office): July data: 1960-2008 vs. 1766-2008

Type of record: July totals only vs. June & July totals

average of 145 mm

2007

average of 132 mm


Linking Rain to River FlowSeasonal totals for Central England: 1920 - 2008

1947 Snowmelt floods

Autumn 2000 floods

Flood-poor period

Rain

River

complex factors

(1) What measure of rain do you use – hourly intensity / weekly totals / seasonal totals?

(2) How do we account for runoff processes?


Concerns Over Discharge DataChanges in catchment land-use?

(1) we are more interested in the timing of peaks rather than their magnitude.

(2) changes in land-use are likely to be manifest as trends rather than cycles.

(3) with only ~20 catchments we can examine each for unusual changes (e.g. step change associated with regulation).


Clusters - Bubble Plots

Methods of capturing clustering: flood frequency counts

No. of peaks within a moving 5-year window.

RI = 1 year

Cycles not Trends

Spatial Correlations

south

north

FLOOD RICH

FLOOD POOR


Changing the Return Interval


Time Between Peaks

Quantitative clustering: individual event timing rather than associated year.


Clustering Statistics

Statistical measure of clustering

over- and under-dispersionDispersion Stat: a measure of the deviation of points in time from equi-dispersion.

http://www.exeter.ac.uk/


Applicability of Dispersion Stat. 10 synthetic POT series

Dispersion stat. tells us that a series has clustering but not about its nature.


Box PlotsPeak counts in moving window (1year – 40years);

POT RI = 1 year;

Red : twice as many floods as we would expect;

Clustering is manifest over different time-scales and at different times-periods.

window length (years)

date

5 10 15 20 25 30 35 40

1900

1910

1920

1930

1940

1950

1960

1970

1980

1990

2000

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2Thames at Kingston

Moving away from individual events.


Peak counts in moving window (1year – 40years);

POT RI = 1 year;

Red : twice as many floods as we would expect;

Clustering is manifest over different time-scales and at different times-periods.

Box Plots

FEB 2011 Institute of Hazard, Risk and Resilience E. RavenClimatic Drivers of

Clustering Summer 2007 Floods and the Jet Stream

1880 1900 1920 1940 1960 1980 2000 202030

40

50

60

70

80

90

100

110

120

Year

Flow

(cum

ecs)

1860 1880 1900 1920 1940 1960 1980 2000 2020-0.4-0.3-0.2-0.1

00.10.20.30.4

year

AMO

ind

ex (3

yr M

A)

Atlantic Multidecadal Oscillation

River Lee

But are catchment processes going to filter out the climate signal?


Challenges to Address• Spatial correlations - catchment size and location;• Advantages and difficulties associated with rainfall

analysis (pluvial flooding);• Climatic influences –teleconnections / climate

change?• Trends on top of clusters; • Clustering of other phenomena – windstorm,

rainfall, droughts, landslides, banking collapse?

• Application to risk management and the insurance industry – is clustering too complex to provide a product?

FEB 2011 Institute of Hazard, Risk and Resilience E. RavenBut what about the sediment?

Part 2: Interactions between sediment, engineering and flood risk in gravel-bed rivers.


Sediment Morphology Interactions

High coarse sediment supply

In-channel deposition Bank erosion Channel capacity

is maintained

Processes in natural, unmanaged, sinuous upland gravel-bed rivers.


Interactions Provoking Management

too high

Loss in capacity

Increased flood-risk

too rapid

Loss of land

CHAN

NEL

MAN

AGEM

ENT

High coarse sediment supply

In-channel deposition Bank erosion Channel capacity

is maintained

Processes leading to channel management: levees, bank protection, gravel traps.

See Raven et al, (2010), PiPG 34, P23 for broad discussion


The Importance of Sediment• Sediment agg/deg can change channel

capacity > flood risk > changing RI;• Sediment moved in large floods can

end up deposited on flood plains: costly clear up;

• Sediment can create problems for infrastructure - bridges, weirs;

• Sediment is important for river aesthetics and habitat.


Combined Methodology(1) Fieldwork: monitor

channel change;

monitor driving processes; explore interactions.

(2) Modelling: develop, apply and test a model of channel change.

DATA repeat cross-sectional surveys

bank erosion monitoring sediment impact sensors

pebble counts / bulk samples field surveys

bend velocity paths


The Upper Wharfe Study Reachmanaged

sinuoussingle thread

bank erosion

12-30m wide

coarse gravel

exposed barsmeandersannual floods flashy

upland-rural

Yorkshire Dales, Northern England

FEB 2011 Institute of Hazard, Risk and Resilience E. RavenSediment and Overbank

Flows

Raven et al. (2009) “The spatial and temporal patterns of aggradation…” , ESPL, 34, p23-45.

4-years of sediment accumulation =

flood frequency, 2.6 times greater and

overbank flow time increased by 12.8 hours.

channel capacity, 02

channel capacity, 06


Initial and boundary conditions

Modelling Framework

Lateral channel change

Output / results

time stepupdates

Sediment transport

Flow hydraulics

Coupling a SRM model with a lateral channel change component;

Three sub-models;

Iterative scheme;

Novel approach – lateral change using a split channel approach.

Raven et al. (online, early view, Jan 11), Hydrological Processes.


Coupling SRM with Lateral Change

SRM: width-averaged

Q

LEFT SIDEhydraulics,shear stress,sediment transport, bed level change.

RIGHT SIDEhydraulics,

shear stress,sediment transport,

bed level change.

Left bed elevationRight bed elevation

Splitting the cross-sectional geometrical representation in the model


Curvature and Lateral Change

Curvature: shifts average shear

Curvature and deeper flow = higher τ

τ > critical erosion τ = bank erosion

τ < critical narrowing τ = bank narrowing

Bank erosion feeds back to lower flow depth and reduce shear stress

Excess shear stress drives bank erosion


Curvature: shifts average shear

Curvature and deeper flow = higher τ

Hard EngineeringLow critical shear = erodible banksHigh critical shear = protected banks

Preventing Lateral Change

Excess shear stress drives bank erosion


Model CalibrationDownstream fining

Steeper channel slope

Model performance vs. field data


Scenarios: Benchmark Comparison

> 0 bend is right turning, high shear on left

τ high

τ low

Bank protection

> 0 for bank erosion

< 0 for bank narrowing

Cross-sectional node

LR

2-Years of Simulating the Actual River Conditions


Max BE: 0.4 m

Max BE: 1.45 m

Scenario 1: width change with protection


Further implications changes in flow depth; changes in shear stress distribution; changes in the locations of sedimentation. wider channel promoting in-channel deposition

Raises caution to restoration schemes

Max BE: 0.4 m

Max BE: 1.45 m

Scenario 1: width change with NO protection


Model Scenarios2: engineering a problematic reach

Scenario 2: Implementing Engineering

sediment accumulation

severe bank erosion

straighten a 350m reach (loss of 75m); removes high curvature; increases slope; narrow and fix banks.


Scenario 2: Engineering

0.3 0.2 0.4 0 0 00.6 1.3 removed 0 0

Normal reach

Engineered reach

Bank erosion (m)

Engineering simply shifts the problems (and makes them worse) up and downstream.


Part 2: SummarySediment is important for flood risk and river management and also insurance;

In-channel deposition can change RI / probabilities of flood events;

The model’s split-channel approach and lateral change component was effective and allowed asymmetrical channel adjustment;

Limitations remain – simplified geometry, fixed curvature, data. Scenarios raise caution to river management schemes that interfere with the sediment transfer system;

This research supports a recommendation that managing sediment sources may be better than managing the after affects.


Emma [email protected]

IHRR Room 256

mailto:[email protected]

Water and Me!

Documents

water floods rivers

floodrisk stats stationary

exposed global risk

institute of hazard

flood clustering work

flood riskbackgroundresults

catastrophic floods

floods rivershydrology