Development of IDF Under Climate Change

THE UNIVERSITY OF WESTERN ONTARIO
DEPARTMENT OF CIVIL AND

ENVIRONMENTAL ENGINEERING
Water Resources Research Report
Development of Probability Based IntensityDuration-Frequency Curves under Climate

Change
By:
Tarana A. Solaiman
and
Slobodan P. Simonovic
Report No: 072

Date: March 2011
ISSN: (print) 1913-3200; (online) 1913-3219;
ISBN: (print) 978-0-7714-2893-7; (online) 978-0-7714-2900-2;
Development of Probability Based Intensity-Duration-Frequency

Curves under Climate Change
By:
Tarana A. Solaiman
And
Slobodan P. Simonovic
Department of Civil and Environmental Engineering

The University of Western Ontario
London, Ontario, Canada
March 2011
1
Executive Summary
Hydrologic design of storm sewers, culverts, retention/detention basins and other

components of storm water management systems are typically performed based on specified
design storms derived from the rainfall intensity-duration-frequency (IDF) estimates and an
assumed temporal distribution of rainfall. Use of inappropriate data or design storms could lead
to malfunctions of the infrastructure systems: over-estimation may result in costly over-design or
under-estimation may be associated with risk and human safety. One of the expected hydroclimatic impacts of climate change for London is the increase in the magnitude and frequency of
extreme rainfalls which can have serious impact on the design, operation and maintenance of
existing municipal water infrastructure.
This study presents a methodology for updating the rainfall IDF curves for the City of
London incorporating various uncertainties associated with the assessment of climate change
impacts on a local scale. Overall, two objectives have been achieved: first, an extensive
investigation of the possible realizations of future climate from 29 scenarios developed from
Atmosphere-Ocean Global Climate Models (AOGCM) and scenarios are performed using a
downscaling based disaggregation approach. Annual maximum series of rainfall are fitted to
Gumbel distribution to develop IDF curves for 1, 2, 6, 12 and 24 hour durations for 2, 5, 10, 25,
50 and 100 years of return periods. Next, the associated uncertainties are estimated using nonparametric kernel estimation approach and the resultant IDF curve is developed based on a
probabilistic approach.
The results indicate that rainfall patterns in the City of London will most certainly change in
future due to climate change. The use of the multi-model approach, rather than a single scenario
2
is encouraged. Inherent uncertainties associated with different AOGCMs are quantified by a

kernel based plug-in estimation approach. The resultant scenario indicates approximately 2040% changes in different duration rainfalls for all return periods. Use of a probability based
intensity-duration-frequency curve is encouraged in order to apply the updated IDF information
with higher level of confidence.
Table of Contents
Executive Summary ........................................................................................................................ 2

Table of Contents ............................................................................................................................ 4
List of Tables .................................................................................................................................. 6
List of Figures ................................................................................................................................. 7
1. Introduction ................................................................................................................................. 8
1.1 Problem Definition ................................................................................................................ 8
1.2 Uncertainties in Atmosphere-Ocean Global Climate Models ............................................. 10
1.3 Intensity-Duration-Frequency Curves ................................................................................. 11
1.4 Outline of the Report ........................................................................................................... 12
2. Literature Review...................................................................................................................... 14
3. Methodology ............................................................................................................................. 18
3.1 Database .............................................................................................................................. 18
3.1.1 Observations ................................................................................................................. 18
3.1.2 Climate Change Scenarios ............................................................................................ 20
3.2 Development of Methodology ............................................................................................ 22
3.2.1 Downscaling ................................................................................................................. 22
3.2.2 Bias Correction of Downscaled Outputs ...................................................................... 26
3.2.3 Disaggregation .............................................................................................................. 27
3.2.4 Generation of Rainfall for Different Durations ............................................................ 29
3.2.5 Intensity-Duration-Frequency Analysis ....................................................................... 30
3.2.6 Uncertainty Quantification ........................................................................................... 32
4. Results and Discussion ............................................................................................................. 37
4
4.1 Selection of Appropriate Stations ....................................................................................... 37

4.2 Development of Climate Change Scenarios........................................................................ 40
4.3 Verification of the IDF Generation Methods ...................................................................... 41
4.4 IDF Results for Future Climate ........................................................................................... 45
4.5 Uncertainty Quantification of IDF Results ......................................................................... 49
5. Conclusions ............................................................................................................................... 55
Acknowledgments......................................................................................................................... 57
References ..................................................................................................................................... 58
APPENDIX A: Regression Test Results ...................................................................................... 66
APPENDIX B: SRES Emission Scenarios ................................................................................... 70
APPENDIX C: Atmosphere-Ocean General Circulation Models ................................................ 72
APPENDIX D: Comparison of Future IDF Results in terms of Intensities (mm/hr) ................... 81
APEPNDIX E: IDF Plots of Selected Scenarios .......................................................................... 87
APPENDIX F: Previous Reports in the Series ............................................................................. 89
List of Tables
Table 1: Rain Gauge Station Details ............................................................................................. 19
Table 2: List of AOGCM Models and Emission Scenarios .......................................................... 21
Table 3: Groups for Regression Analysis based on Distances ..................................................... 38
Table 4: Cross-Correlation Results for Stations Within 200 km Distance from London ............. 40
Table 5 (a): Monthly Mean Precipitation (mm) from Different AOGCMs for 1965-1990 .......... 41
Table 6 (a): Comparison of Extreme Rainfall in London in terms of Depth (mm) ...................... 44
Table 6 (b): Relative Difference between EC IDF Information and Historic Unperturbed
Scenario......................................................................................................................................... 45
Table 7: Percent Differences between Historic Perturbed, Wet and Dry Scenarios ..................... 46
Table 9: Difference between Historical Perturbed and the Resultant Scenario for 2080s ......... 52
List of Figures
Figure 1: Meteorological Stations Used in the Study ................................................................... 18

Figure 2: Schematic Diagram of Developing IDF Curve ............................................................. 23
Figure 3: Performances of Stations based on Distance ................................................................. 39
Figure 4: Box and Whiskers Plot of Simulated Monthly Rainfall in London .............................. 42
Figure 5: Frequency Plots of Observed (Obs) and Simulated (Sim) Hourly Rainfall .................. 43
Figure 6: IDF Plots of AOGCM Scenarios for Different Durations ............................................. 48
Figure 7: Comparison of IDF Plots for Different Scenarios ......................................................... 51
Figure 8: IDF Plot for Resultant Scenario .................................................................................... 52
Figure 9 (a): Probability based IDF Curve of 1 and 2 Hour Duration .......................................... 53
Figure 9 (b): Probability based IDF Curve of 6, 12, 24 Hour Duration ....................................... 54
1. Introduction
1.1 Problem Definition
The increase of carbon dioxide concentration in the atmosphere due to industrial activities in
the past and recent times has been identified as the major cause of global warming and climate
change. The normal balance of Earths hydrological cycle has been altered due to the changes in
the temperature and precipitation patterns. Projections from climate models suggest that the
probability of occurrence of intense rainfall in future will increase due to the increase in green
house gas emission (Mailhot and Duchesne, 2010). Research related to the analysis of extreme
precipitation indices have projected an increase in the annual total precipitation during the
second half of the past century; the number of days with precipitation is also expected to
increase, with no consistent pattern for extreme wet events (Vincent and Mekis, 2005). Stone et
al. (2000) reported seasonally increasing trends in total precipitation during the 20th century for
southern parts of Canada resulting from increased heavy and intermediate events. Research
related to the Upper Thames River basin (Solaiman and Simonovic, 2011) have also indicated
that there is now higher probability that the occurrence of extreme precipitation events will be
more frequent in future. Such changes in extreme events have enormous ecological, societal and
economic impacts in the form of floods, droughts, heat waves, summer and ice storms and have
great implications for municipalities: a small shift in the climate normals can have large
consequences on the existing infrastructure; climate change will affect any municipalities (big or
small, rural or urban) by damaging existing municipal infrastructure (bridges/roads), natural
systems (watersheds, wetlands and forests) and human system (health and education) (Mehdi et
al, 2006). The design standards at present are based on the historic climate information and
required level of protection from natural phenomena. Under a changing climate, it has become a
8
priority for the municipalities to search for appropriate procedures, planning and management to
deal with and adopt to changing climatic conditions. Decision makers and stakeholders need to
understand the possible effects for developing suitable management decisions for the future.
Possible changes may demand new regulations, guidelines for storm water management studies,
revision and update of design practices and standards, or retrofitting of existing infrastructure or
even constructing additional ones (Prodanovic and Simonovic, 2007).
Global scale climate variables are commonly projected by Coupled Atmosphere-Ocean
Global Climate Models (AOGCMs) to provide a numerical representation of the climate system
based on the physical, chemical and biological properties of their components and feedback
interactions between them (IPCC, 2007). They are, currently the most reliable tools available for
obtaining the physics and chemistry of the atmosphere and oceans and to derive projections of
meteorological variables (temperature, precipitation, wind speed, solar radiation, humidity,
pressure, etc). They are based on various assumptions about the effects of the concentration of
greenhouse gases in the atmosphere coupled with projections of CO2 emission rates (Smith et al.,
2009).
There is a high level of confidence that AOGCMs are able to capture large scale circulation
patterns and correctly model smoothly varying fields such as surface pressure, especially at
continental or larger scales. However, it is extremely unlikely that these models properly
reproduce highly variable fields, precipitation (Hughes and Guttorp, 1994), on a regional scale,
let alone, for small to medium watersheds. Although confidence has increased in the ability of
AOGCMs to simulate extreme events, such as hot and cold spells, the frequency and the amount
of precipitation during intense events are still being underestimated.
Present study aims to provide an insight into the future changes in the intensity of extreme
rainfall events associated with model and scenario uncertainties and suggest methods for
quantifying these uncertainties. The result is presented in the form of probability based intensityduration-frequency (IDF) curves appropriate for the future climatic conditions.
1.2 Uncertainties in Atmosphere-Ocean Global Climate Models

In recent years, quantifying uncertainties from AOGCMs and scenarios for impact
assessment studies has been identified as a critical climate change and adaptation research topic.
Climate change impact studies derived from AOGCM outputs are associated with uncertainties
due to incomplete knowledge originating from insufficient information or understanding of
biophysical processes or a lack of analytical resources. Examples include simplification of
complex processes involved in atmospheric and oceanographic transfers, inaccurate assumptions
about climatic processes, limited spatial and temporal resolution resulting in a disagreement
between AOGCMs over regional climate change, etc. Uncertainties also emerge due to
unknowable knowledge arising from the inherent complexity of the Earth system and from our
inability to forecast future socio-economic and human behavior in a deterministic manner (New
and Hulme, 2000; Allan and Ingram, 2002; Proudhomme et al., 2003; Wilby and Harris, 2006;
Stainforth et al., 2007; IPCC, 2007, Buytaert et al, 2009). It is now established that the accuracy
of AOGCMs decrease at finer spatial and temporal scales; a typical resolution of AOGCMs
ranges from 250 km to 600 km, but the need for impact studies conversely increase at finer
scales. The representation of regional precipitation is distorted due to the coarse resolution and
cannot capture the subgrid-scale processes significant for the formation of site-specific
precipitation conditions. While some models are parameterized, details of the land-water
distribution or topography in others are not represented at all (Widmann et al., 2003). Studies
10
have found that the models failed to predict the high variability in daily precipitation and could
not accurately simulate present-day monthly precipitation amounts (Trigo and Palutikof, 2001;
Brissette et al., 2006).
1.3 Intensity-Duration-Frequency Curves

Reliable rainfall intensity estimates are necessary for hydrologic analyses, planning and
design problems. The rainfall intensity-duration-frequency (IDF) curve is one of the most
common tools for urban drainage designer. Information from IDF curves are used to describe the
frequency of extreme rainfall events of various intensity and durations. According to the
guideline for Development, Interpretation and Use of Rainfall Intensity-Duration-Frequency
(IDF) Information: A Guideline for Canadian Water Resources Practitioners developed by
Canadian Standards Association (CSA, 2010), the major reasons for increased demand for
rainfall IDF information can be summarized as follows:
As the spatial heterogeneity of extreme rainfall patterns becomes better understood and
documented, a stronger case is made for the value of locally relevant IDF information.
As urban areas expand, making watersheds generally less permeable to rainfall and
runoff, many older water systems fall increasingly into deficit, failing to deliver the services for
which they were designed. Understanding the full magnitude of this deficit requires information
on the maximum inputs (extreme rainfall events) with which drainage works must contend.
Climate change will likely result in an increase in the intensity and frequency of extreme
precipitation events in most regions in the future. As a result, IDF values will optimally need to
be updated more frequently than in the past and climate change scenarios might eventually be
drawn upon in order to inform IDF calculations.
11
The typical establishment of rainfall IDF curves involves three steps. First a probability
distribution function (PDF) or Cumulative Distribution Function (CDF) is fitted to each group
comprised of the data value for any specific duration. The maximum rainfall intensity for each
time interval is related with the corresponding return period from the cumulative distribution
function. For a given return period , the cumulative frequency
can be expressed as:
(1.1)
or
(1.2)
If the cumulative frequency is known, the maximum rainfall intensity can be determined
using an appropriate theoretical distribution function (such as Generalized Extreme Value
(GEV), Gumbel, Pearson Type III, etc).
In the presence of climate change, the theoretical distribution based on historical
observations will be different for the future conditions. The issue is further aggravated by the
presence of various uncertainties associated with AOGCM models and emission scenarios.
Therefore, in this study the non-parametric kernel estimator is used to combine uncertainties
generated from different AOGCMs. Probability of occurrences of maximum rainfall generated
for any specific duration is presented in the form of cumulative distribution function for different
return periods.
1.4 Outline of the Report

The report is organized as follows: research related to IDF information under climate is
presented in Chapter 2. Chapter 3 details the database and methodology applied in the study. The
12
results and discussion obtained from the analysis are explained in Chapter 4. Finally the report
ends with conclusions based on the research findings.
13
2. Literature Review
Literature related to intensity-duration-frequency (IDF) curves concentrates on developing

appropriate distribution of fit, comparison of sampling techniques and generation of IDF
information under climate change.
Interesting research is emerging on the development of alternative methods, other than the
distribution fit, for developing IDF values. Huard et al (2010) applied a Bayesian analysis to the
estimation of IDF curves. Comparison of the Bayesian and classical approach using GEV
distribution using Peak Over Threshold (POT) method indicated the extent of uncertainties in the
IDF curves. Svensson et al (2007) made an experimental comparison of methods for estimating
rainfall IDF from fragmented records for Eskdalemuir, Scotland. Three different methods were
applied to cope with the missing data in the annual and monthly series: (i) using only
years/months with complete records; (ii) using only years/months with complete records with not
more than 20% missing data; and (iii) using censored data from months where records are
incomplete. The result recommends the use of monthly maxima for calculating return period
rainfall allowing up to 20% of missing data in each month. Despite the fact that over a decade
long research have been investigating for alternate methods for IDF development, studies related
to developing IDF curves incorporating climate change are limited.
Estimations of future modifications in rainfall due to increase in greenhouse gas
concentrations depend on response from global climate models. Studies have related statistical
downscaling with outputs from global and regional climate model outputs. Nguyen et al (2007a,
b) and Desramaut (2008) presented a spatial-temporal downscaling method based on scale
invariance technique for constructing IDF relations using outputs from two GCMs (HadCM3 A2
14
and CGCM2 A2) for future climate. The spatial downscaling methodology based on SDSM was
used to generate daily precipitation data. The temporal scaling was performed for extreme value
distribution factors based on current historical rainfall distribution. The studies found large
differences in future IDF values between two the models.
Prodanovic and Simonovic (2007) developed IDF curves for current and future climate for
city of London using a K-NN based weather generator. Future rainfall derived for the wet
(CCSRNIES B21) scenario projected 30% increase in rainfall magnitude for a range of durations
and return periods. More recently Simonovic and Peck (2009) used all the available
precipitation data for different durations for developing IDF information under the wet climate
change scenario. The 24 hr duration rainfall was modified by applying moving window
procedure to recreate maximum 24 hour rainfall events crossing the calendar day boundary.
Their study indicated 10.7% to 34.9% change in IDF information for 2050s.
Coulibaly and Shi (2005) used outputs from CGCM2 B2 to develop IDF curves for Grand
River and Kenora Rainy River regions in Ontario using statistical SDSM downscaling
methodology. Their study found an increase in the range of 24-35% in the rainfall intensity for
24 hour and sub-daily durations for all stations of interest for 2050s and 2080s with decreases in
2020s.
Mailhot et al. (2006, 2007) used outputs from Regional Climate Models (RCMs) (CRCM A2)
for developing IDF for different durations for May-October over Southern Quebec using regional
frequency analysis. The results were obtained for the RCM grid-box scale ranging over 45 km
distances in between two grids. Projected rainfall showed 50% decrease by 2050s for 2 and 6
hour durations and 32% decrease for 12 and 24 hour durations than the base climate (1961-
15
1990). The results indicated limitation of using grid box scale and acknowledged that the results
may be improved by using point estimates.
Onof and Arnbjeg-Nielsen (2009) used an hourly weather generator approach with
disaggregation to derive IDF values from hourly rainfall data. Future hourly data was obtained
from RCM A2 scenario with a 10 KM x 10 KM resolution for 2050s. The limitation of the study
includes the stationarity assumption that the ratio of areal to the point estimates will remain
unchanged with any changes in the climate.
Literature related to developing IDF values incorporating climate change from AOGCM
models suffer from:
(i) Limitations of statistical downscaling approaches: Downscaling approaches such as
SDSM or most of the weather generators assumed to have stationary climate. One possible way
to overcome such issue is to perturb the model to generate values to achieve outputs beyond the
range of inputs, which can be easily included in the weather generator.
(ii) Application of sub-daily scaling factors to daily precipitation data and uncertainties: Use
of historical hourly data can prevent this issue.
(iii) Use of single AOGCM response: In all the literature listed above, single AOGCMs have
been used for predicting future climate. It is well understood that in the presence of significant
uncertainties, utilization of a single AOGCM may be one of all possible realizations and cannot
be representative of the future. So, for a comprehensive assessment of the future changes, it is
important to use collective information by utilizing all available GCM models, synthesizing the
projections and uncertainties in a probabilistic manner.
(iv) Appropriate distribution fit for future: In presence of human induced warming trends
added to Earths natural variability, it is unlikely that the present precipitation or rainfall pattern
16
will comply with the future. Differences in the initializations and parameterizations of different
climate model responses make it more complex to assume a specific distribution for all possible
realizations.
17
3. Methodology
3.1 Database
3.1.1 Observations
Environment Canada is responsible for collection and distribution of weather data in
Canada. The Environment Canadas hourly database mostly consists of rainfall data; the hourly
gauges often freeze during winter and estimates obtained from them are not accurate; hence for
this part of the study, precipitation could not be used. Hourly rainfall data covering stations
around London for the period of 1965-2003 (Figure 1) has been extracted from the Data Access
Integration Network (DAI, 2009).
Figure 1: Meteorological Stations Used in the Study
18
Daily rainfall data for the same stations and same time period is obtained from Environment
Canadas Weather Office (http://climate.weatheroffice.ec.gc.ca/climateData/canada_e.html).
The station selection process is highly dependent on the availability of hourly data with
adequate lengths. This is an important step in running nearest neighbor based weather generator
used in the present study. The number of stations used in the K-NN algorithm influences
computation of regional means and the Mahalanobis distance (see section 3.2.1 for details),
which affects the choice of nearest neighbor. Data of shorter durations are available only for a
handful of stations. So stations closer to London but with shorter record have not been
considered in this study. At first, all hourly stations within 200 km radius of London are
considered. Next, stations with data going back to 1965 with a record till 2001 are selected.
Figure 1 and Table 1 present the details of stations used initially for IDF analysis.
Table 1: Rain Gauge Station Details
Latitude Longitude Elevation Distance from London
(deg)
(deg)
(m)
(km)
Barrie WPCC
44.3758
-79.6897
221
190
Brantford MOE
43.1333
-80.2333
196
75
Chatham WPCP
42.39
-82.2153
180
113
Delhi
42.8667
-80.55
232
52
Elora
43.65
-80.4167
376
91
Fergus
43.7347
-80.3303
418
102
Hamilton A
43.1717
-79.9342
238
100
Hamilton RBG
43.2833
-79.8833
102
106
London Intl A
43.0331
-81.1511
278
0
Owen Sound MOE 44.5833
-80.9333
179
173
Sarnia
43
-82.3
181
93
St. Thomas
42.7833
-81.1667
236
28
Stratford MOE
43.3689
-81.0047
345
39
Toronto
43.6667
-79.4
113
158
Toronto Intl A
43.6772
-79.6306
173
142
Waterloo A
43.45
-80.3833
317
78
Wiarton A
44.7458
-81.1072
222
190
Woodstock
43.1361
-80.7706
282
33
Climate ID Station Name

6110557
6140954
6131415/6
6131982/3
6142285/6
6142400
6153194
6153300/1
6144475/8
6116132
6127519
6137361/2
6148105
6158350
6158733
6149387
6119500
6149625
19
3.1.2 Climate Change Scenarios

Atmosphere-Ocean Global Climate Models (AOGCMs) represent physical processes in
the atmosphere, ocean, cryosphere and land surface and are the most advanced tools available for
simulating the response of the global climate system to increasing greenhouse gas
concentrations. While simpler models have also been used to provide global or regional average
estimates of the climate response, only AOGCMs, possibly in conjunction with nested regional
models, have the potential to provide geographically and physically consistent estimates of
regional climate change needed for climate change impact studies (IPCC, 2007). It is important
to note that AOGCMs offer only possibilities of future climate pattern in differing socioeconomic conditions depending on continual growth of population, increased carbon dioxide
emission, rate of urbanization, etc. Outputs from AOGCMs, thus, should not be considered as the
forecasts of future climate conditions.
The Canadian Climate Change Scenarios Network (CCCSN) provides access to several
AOGCM models and emission scenarios. The website allows the user to specify the range of
geographical co-ordinates required, as well as the climatic variable and time period of interest.
For the purpose of this study, precipitation data for two time slices: 1960-1990 (baseline) and
2071-2100 (2080s) are collected. It is important to note here that the AOGCMs provide only
precipitation data which is a combination of snow and rainfall during winter. They do not count
for rainfall change information. Hence for this study, change in the precipitation between
different AOGCM scenarios and historical observed precipitation are used to calculate the
change fields and are applied to develop modified rainfall series to be used in the weather
generator.
20
A total of 27 scenarios from 11 AOGCMs, each with two to three emission scenarios
(Nakicenovic et al, 2000) are selected for developing future scenarios. Full descriptions of the
emissions scenarios and AOGCMs can be found in Appendices B and C. Table 2 provides a
complete list of the details of the AOGCM scenarios used in this study.
Table 2: List of AOGCM Models and Emission Scenarios
GCM models
CGCM3T47, 2005
CGCM3T63, 2005
CSIROMK3.5, 2001
ECHAM5AOM,
2005
ECHO-G, 1999
GFDLCM2.1, 2005
GISSAOM, 2004
MIROC3.2HIRES,
2004
Sponsors, Country
Canadian Centre for Climate Modelling

and Analysis, Canada
Commonwealth Scientific and Industrial
Research Organization (CISRO)
Atmospheric Research, Australia
Max Planck Institute for Meteorology,
Germany
Meteorological Institute of the University
of Bonn, Meteorological Research Institute
of the Korea Meteorological
Administration (KMA), and Model and
Data Group, Germany/Korea
U.S. Department of Commerce/ National
Oceanic and Atmospheric Administration
(NOAA)/Geophysical Fluid Dynamics
Laboratory (GFDL), USA
National Aeronautics and Space
Administration (NASA)/ Goddard Institute
for Space Studies (GISS), USA
Centre for Climate System Research

(University of Tokyo), National Institute
for Environmental Studies, and Frontier
MIROC3.2MEDRES,
Research Centre for Global Change
2004
(JAMSTEC), Japan
Centre for Climate System Research,
CCSR/NIES B21,
University of Tokyo and National Institute
1999
for Environmental Studies, Japan
Commonwealth Scientific and Industrial
CSIROMK2b, 1997
Research Organization (CISRO)
Atmospheric Research, Australia
SRES
scenarios
Atmospheric
resolution
Lat
Long
3.75
3.75
2.81
2.81
A1B, A2,
B1
1.875
1.875
A1B, B1,
A2
1.875
1.875
A1B, B1,
A2
3.9
3.9
A1B, B1,
A2
2.5
A1B, B1
A1B, B1
1.125
1.125
A1B, A2,
B1
2.8
2.8
B21
5.6
5.6
B11
5.6
3.2
A1B, A2,
B1
A1B, A2,
B1
21
3.2 Development of Methodology

Figure 2 presents the schematic of the methods adopted for generating IDF curves for
2080s incorporating uncertainties from AOGCM scenario. The step by step procedures are
presented below.
3.2.1 Downscaling
Stochastic weather generators simulate weather data to assist in the formulation of water
resource management policies. The basic assumption for producing synthetic sequences is that
the past would be representative of the future. They are essentially complex random number
generators, which can be used to produce a synthetic series of data. This allows the researcher to
account for natural variability when predicting the effects of climate change.
In order to reduce multi-dimensionality and collinearity associated with the large number of
input variables, principal component analysis has been integrated with the weather generator.
The process requires selecting appropriate principal components (PCs) that will adequately
represent most information of the original dataset.
The WG-PCA algorithm with p variables and q stations works through the following steps:
1) Regional means of p variables for all q stations are calculated for each day of the observed
data:
Xt x1, t, x 2, t,..., xp, t
t 1,2,...,T
(3.1)
22
Observed Rainfall
(Daily)
Station
Selection
Gridded AOGCM
Outputs (27 Scenarios)
Linear
Interpolation
Cross-Correlation Analysis
Regression Analysis
AOGCM Outputs on
Station Scale
Weather
Generator
Selected Observed
Rainfall (Daily)
Unperturbed
Modified Daily Output

(27 Scenarios)
Simulated Output
(Present)
Simulated Hourly
Output (Present)
Maximum Rainfall
for Different
Durations (Present)
Change Field
Analysis
Perturbed
Daily Outputs 2080s

(27 Scenarios)
Disaggregation
Hourly Output 2080s

(27 Scenarios)
AMS
Max. Rainfall for Diff.

Durations 2080s
(27 Scenarios)
IDF Curve for

Present
IDF Curves for 2080s

(27 Scenarios)
Quantifying
Uncertainty
Plug-in
Kernel
IDF Curve 2080s

(Resultant Scenario)
Figure 2: Schematic Diagram of Developing IDF Curve

where x i , t
1 q j
xi,t
q j 1
i 1,2,..., p
(3.2)
23
2) Selection of potential neighbors, L days long where L=(w+1) (N-1) for each of p individual
variable with N years of historic record, and a temporal window of size w which can be set by
the user of the weather generator. The days within the given window are all potential neighbors
to the feature vector. N data which correspond to the current day are deleted from the potential
neighbors so the value of the current day is not repeated.
3) Regional means of the potential neighbors are calculated for each day at all q stations.
4) A covariance matrix, Ct of size L p is computed for day t.
5) The first time step value is randomly selected for each of p variables from all current day
values in the historic record.
6) Next, using the variance explained by the principal component, Mahalanobis distance is
calculated with equation 3.
dk
PCt PCk 2 / Var ( PC )
k 1,2,..., K
(3.3)
Where PCt is the value of the current day and PCk is the nearest neighbor transferred by the
Eigen vector. The variance of the first principle component is Var(PC) for all K nearest
neighbors.
7) The selection of the number of nearest neighbors, K, out of L potential values using K L .
8) The Mahalanobis distance dk is put in order of smallest to largest, and the first K neighbors in
the sorted list are selected (the K Nearest Neighbors). A discrete probability distribution is used
which weights closer neighbors highest in order to resample out of the set of K neighbors. Using
equations 4 and 5, the weights, w, are calculated for each k neighbor.
24
wk
1/ k
K
1 / i
k 1,2,..., K
(3.4)
i 1
Cumulative probabilities, pj, are given by:

j
p j wi
(3.5)
i 1
9) A random number u (0,1) is generated and compared to the cumulative probability

calculated above in order to select the current days nearest neighbor. If p1 < u < pk, then day j
for which u is closest to pj is selected. However, if pi > u, then the day which corresponds to d1
is chosen. If u=pK, then the day which corresponds to day dK is selected. Upon selecting the
nearest neighbor, the K-NN algorithm chooses the weather of the selected day for all stations in
order to preserve spatial correlation in the data (Eum et al, 2009).
10) In order to generate values outside the observed range, perturbation is used. A conditional
standard deviation for K nearest neighbors is estimated. For choosing the optimal bandwidth of
a Gaussian distribution function that minimizes the asymptotic mean integrated square error
(AMISE), Sharma et al. (1997) reduced Silvermans (Silverman 1986, pp. 86-87) equation of
optimal bandwidth into the following form for a univariate case:
1.06K 1 / 5
(3.6)
Using the mean value of the weather variable xij,t obtained in step 9 and variance ( i j ) 2 , a
new value yij,t can be achieved through perturbation (Sharma et al. 1997).
y ij,t xij,t ij z t
(3.7)
25
Where zt is a random variable, distributed normally (zero mean, unit variance) for day t.
Negative values are prevented from being produced for precipitation by employing a largest
acceptable bandwidth: a x*,j t / 1.55 *j where * refers to precipitation. If again a negative value
is returned, a new value for zt is generated (Sharif and Burn, 2006).
3.2.2 Bias Correction of Downscaled Outputs

The downscaling process scales down coarse grid outputs of AOGCMs into the scale of
interest. However, significant simulation bias still may exist from the initializations of
atmospheric-oceanic processes. Hence, employing coarse resolution global model output for
regional and local climate studies requires an additional bias correction step based on the ability
of the AOGCMs to reproduce the past climate. In this study, bias from the downscaled outputs is
corrected by the following equation:
Bias in AOGCM,
(3.8)
Where,
bi = bias from different AOGCMs
xi = Monthly mean of observed precipitation for 1965-1990
yi = Monthly mean from different AOGCMs for 1965-1990
the correction factor for the AOGCMs is then calculated using,

(3.9)
So, the treated downscaled rainfall for 2080s,
(3.10)
26
where,
pi = untreated daily downscaled rainfall for 2080s
3.2.3 Disaggregation
The disaggregation scheme works by extracting rainfall event records from the hourly
observed data. A rainfall event can be defined as a period of non-zero rainfall for two or more
days where the total amount of rainfall during the consecutive days is considered as the event
rainfall value. Once the rainfall events are extracted from the historic record, they are
disaggregated by a K-nearest neighbor approach. The algorithm considers daily rainfall produced
by the weather generator for day , for each station. A set of potential events are selected from
the observed record from which once such event is chosen based on which the daily output is
disaggregated into hourly values.
The selection of neighboring events from the observed record follows simple rule:
Only events within a moving window of
days are selected to account for the seasonally
varied temporal distribution of rainfall. Events are selected from the prescribed moving window
from all years in the historic record of events as a potential set of neighbors. The daily totals
from downscaled outputs are compared with the set of neighboring event totals to assure that the
only disaggregation of similar events is considered.
Observed hourly data is used as a template on how the hourly values of the generated
outputs would look like. A specific number of days are considered to compare with the present
day value. The best match is determined by (Mansour and Burn, 2010):
) )
) ) (3.11)
27
Where,
is the daily rainfall output from weather generator,
daily rainfall,
and
is the historical observed
are the events calculated from WG outputs and historical observed data
respectively. The weights
are used to identify the best historical hourly ratio of the data.
An hourly value set within the moving window of days can be chosen for a similar event
for the daily total rainfall
. The combination of the weights, which provide the lowest
or
for
each value within the window, is considered as the daily ratio of historical hourly values used to
disaggregate the WGs daily data into hourly values. The ratio of the hourly values found within
the chosen day is applied to the daily value to create a plausible hourly set-up for the given daily
data. This is done based on the methods of fragments (Svanidze, 1977; Sharif et al., 2007). The
fragments represent the fraction of daily rainfall that occur during each hour of the day summing
to unity and can be expressed as:
where
observations,
(3.12)
represents the fragments calculated for hour ,
is the chosen hourly data from
is the number of hours in a day which is 24. The fragments are then multiplied
with the daily data to produce data for each hour:
(3.13)
where
is the daily rainfall (mm). This program has been sent daily data that already had
known hourly values and the results have been compared in an attempt to verify that the model
works correctly.
28
This approach utilizes locally observed data using a non-parametric method avoiding the
chance of errors that may occur from the parametric methods due to theoretical distribution fits,
parameter estimations and calibration. Additionally, there is high likelihood that the statistical
characteristics of the disggregated rainfall are stored by applying the resampling algorithm.
3.2.4 Generation of Rainfall for Different Durations

Sampling of rainfall data for estimating rainfall extremes are commonly proceeded using
two approaches: the annual maxima series (AMS) or block maxima and peak over threshold
(POT) or partial duration series (PDS) (Coles, 2001). Literatures have identified limitations and
advantages of both methods. Madsen et al. (1997), Buishand et al. (1990), Rasmussen et al.
(1994) found POT to be a better approach than AMS. While Kartz et al. (2002), Smith (2003), de
Michele and Salvadori (2005) suggested use of both methods. By definition, AMS approach
includes the yearly peaks in the observational period while the POT involves all the peak events
that exceed a given threshold value. The AMS method is more straightforward. If the number of
annual maxima is small (<100), the obtained estimates may be sensitive to outliers. It is an
asymptotic method that works well if the number of inputs from which a maximum is
considered, is large. Jeruskova et al., 2006 showed that convergence to limit any distribution fit
can be slow. For determining annual maxima, the maxima of 365 daily values are considered.
The seasonal effect may also play a role. Application of POT is somewhat difficult than the
AMS because of its selection of an appropriate threshold. For a satisfactory stability of the
obtained results, testing of several threshold values such as 90%, 95% and 98% are
recommended. Jeruskova et al. (2006) have shown that the POT method may work well for short
memory series only. For longer data series, the series should be split into several more
homogeneous groups. Both methods however, have their own disadvantages too; the AMS may
29
neglect certain high values, while the POT may suffer from serial correlation problem (Jervis et
al., 1936; Langbein, 1949;, Taesombat and Yevjevich, 1978).
3.2.5 Intensity-Duration-Frequency Analysis

Rainfall intensity-duration-frequency (IDF) curves are derived from the statistical
analysis of rainfall events over a period over time and used to capture important characteristics
of point rainfall for shorter durations. It is considered as a convenient tool for gathering regional
rainfall information required for municipal storm water management works. Site specific curves
represent intensity-time relationship for a specific return period from a series of storms.
Information is summarized by plotting the durations on the horizontal axis, the rate of rainfall
(intensity in depth per unit of time) on the vertical axis and the curves for each design storm
return period. Frequency is expressed in terms of return period, T, the average length of time
between rainfall events that equals or exceed any given magnitude. For each selected duration,
annual maximum rainfall is extracted from the rainfall data and frequency analysis is performed
to the annual maximum rainfall to fit a probability distribution for standardizing the
characteristics of rainfall for each station with varying rainfall record.
In Canada, Environment Canada is responsible (a) for collection and quality control of
rainfall data and (b) for providing the rainfall extreme information in the form of IDF curves.
Gumbels Extreme Value distribution is normally used to fit the annual extremes of rainfall using
AMS method. It is acknowledged here that due to changes in future precipitation extremes, the
future rainfall may not follow the conventionally used Gumbels distribution. It would be
adequate to consider a generalized extreme value (GEV) distribution. But the inherent
uncertainties in the responses of AOGCM outputs do not guarantee GEV as the best fit for all
AOGCMs. Furthermore, fitting of three parameters for the GEV distribution by maximum
30
likelihood method requires considerable computation time, and will be different for different
AOGCM responses. For simplification, and to comply with Environment Canadas procedure,
use of Extreme Value (EV) type 1 which is Gumbel distribution is adopted in this study.
The Gumbel probability distribution is expressed (Watt et al., 1989):
(3.14)
where
represents the magnitude of the
year event,
deviation of the annual maximum series, and

period, . The frequency factor
and
are the mean and standard
is a frequency factor depending on the return
is obtained using the following equation:
( (
))
(3.15)
Meteorological Service of Canada (MSC) uses the above method to calculate rainfall
frequency for durations of 5, 10, 30 minutes and 1, 2, 6. 12, 24 hours. Since most of the stations
do not have observed sub-hourly data, the calculation of the frequencies for periods shorter than
1 hour may be based on the ratios provided by the World Meteorological Organization (MTO,
1997):
Duration (min)
10
15
30
Ratio (n-min to 60-min)
0.29
0.45
0.57
0.79
However, for the present study, durations shorter than 1 hour are not considered.
31
The IDF data is next fitted to a continuous function in order to make the process of IDF data
interpolation more efficient i.e. if the ratio of any duration is not available, the IDF data is fitted
to the following three parameter function:
where
(3.16)
presents the rainfall intensity in mm/hr,
is the duration of rainfall in minute,
are the constants. To obtain optimal values for these three parameters, a reasonable
value of
is assumed and the values of
are estimated by the least square method. The
process is repeated to achieve the closest fit of the data (MTO, 1997). Plots of rainfall intensity
vs. duration for each return period is then produced from the fitted IDF data to equation 3.14.
3.2.6 Uncertainty Quantification

A practical approach to deal with AOGCM and scenario uncertainties originating from
inadequate information and incomplete knowledge should: (i) be robust with respect to model
choice; (ii) be statistically consistent in a uniform application across different spatial scales such
as global, regional or local/watershed scales; (iii) be flexible enough to deal with the variety of
data; (iv) obtain the maximum information from the sample; and (v) lead to consistent results.
Most parametric methods do not meet all these requirements.
Probability Density Function (PDF) is commonly used to describe the nature of data. In
applications an estimate of the unknown
from ( ) is calculate in the form of
( ) based on random sample

( ). Probability distribution functions estimated by
any nonparametric method without prior assumptions can be suitable to quantify AOGCM and
scenario uncertainties. Several approaches such as kernel methods, orthogonal series methods,
32
penalized-likelihood methods, k-nearest neighbor methods, Bayesian-spline methods, and

maximum-likelihood or histogram like methods can be found in the literature (Adamowski,
1985).
Kernel density estimation method has been widely used as a viable and flexible alternative to
parametric methods in hydrology (Sharma et al., 1997; Lall, 1995), flood frequency analysis
(Lall et al., 1993; Adamowski, 1985), and precipitation resampling (Lall et al., 1996) for
estimating probability density function.
A kernel density estimate is formed through the convolution of kernels or weight functions
centered at the empirical frequency distribution of the data. A kernel density estimator involves
the use of kernel function (K(x)) defined by:
( )
(3.17)
A PDF thus, can be used as a kernel function. The Parzen-Rosenbalt kernel density estimate
( ) at x, from a sample of {
( )
where
} of sample size n can is given by:
) and
( ) is a weight or kernel function required to satisfy criteria such as
(3.18)
symmetry, finite variance, and integrates to unity. Successful application of any kernel density
estimation depends more on the choice of the smoothing parameter or bandwidth (h) and the type
of kernel function K(.), to a lesser extent. Bandwidth for kernel estimation may be evaluated by
minimizing the deviation of the estimated PDF from the actual one.
33
The behavior of the estimator (equation 3.18) may be analyzed mathematically under the
assumption that the data sets represent independent realizations from a probability density f(x).
The basic methodology of the theoretical treatment is to discuss the closeness of estimator to
the true density, . Successful application of the estimator depends mostly on the choice of a
kernel and a smoothing parameter or bandwidth. Literatures have found that the choice of
bandwidth is more critical. A change in kernel bandwidth can dramatically change the shape of
the kernel estimate (Efromovich, 1999). For each x, ( ) can be thought as a random variable
. Except otherwise stated, will refer to a sum for
because of its dependence on
and to an integral over the range (
).
The discrepancy of the density estimator from its true density
can be measured by mean
square error (MSE):

( )
( ( )
( )
(3.19)
By standard elementary properties of mean and variance,
( )
( ( )
( ) }
( ),
(3.20)
the sum of the squared bias and the variance at . In many applications a trade-off is applied
between the bias and the variance in equation (3.20); the bias can be reduced by increasing the
variance and vice versa by adjusting the degree of smoothing.
It can be obtained by minimizing the mean integrated square error (MISE), a widely used
measure of global accuracy of as an estimator of
(Rosenblatt, 1956; Adamowski, 1985; Scott
et al., 1981, Jones et al., 1996) defined as:

34
( )
( ( )
( )
(3.21)
or in alternative forms,
( )
( )
( ( )
which gives the
( )
( )
(3.22)
as the sum of the integrated square bias and the integrated variance.
Asymptotic analysis provides a simple way of quantifying how the bandwidth h works as a
smoothing parameter. Under standard assumptions, MISE is approximated by the asymptotic
mean integrated square error (AIMSE) (Jones et al., 1996):
( )
where ( )
( )
( )
( )(
and
)
( )
(3.23)
, n is sample size, h is bandwidth. The
first term (integrated variance) is large when h is too small, and the second term (integrated
squared bias) is large when h is too large.
( ) is easily calculated as:
The minimizer of
( )
( )
( )(
(3.24)
In this study the solve-the-equation-plug-in approach is used to derive the data-driven

bandwidth for estimating the densities.
The main thought behind the solve the equation plug in approach is to plug an estimate of
the unknown ( ) in the equation (3.24). The major challenge is to estimate a pilot bandwidth.
35
The solve the equation approach proposed by Hall (1980), Sheather (1983, 1986) and later
refined by Sheather and Jones (1991) is used in this study. The smallest bandwidth, hSJPI is
considered as the solution of the fixed point equation
( )
( ( ) )(
(3.25)
A dummy minimizer in the form g(h) is used to provide better representation of ( ). It is

done by estimating an analogue of
for estimating ( ) by ( ).
The minimizer of the asymptotic mean square error (AMSE) is expressed as:
{ (
for suitable functional
representation of
( )
)} ( )
and
for
{ ( )
for appropriate functionals
. The expression of
(3.26)
in terms of
comes from solving the
and substituting to get

(
)} ( )
. The unknowns ( ) and (
(3.27)
) are estimated by ( ) and
( ), with bandwidths chosen by reference to a parametric family, as for

Many variations have been tested for treatment of
( ) and ( ).The major contribution
has been to try to reduce the influence of the normal parametric family even further by using
pilot kernel estimates instead of normal interference (Jones et al., 1996). Park and Marron (1990)
has shown the improvements in terms of the asymptotic rate of convergence up to a certain point.
36
4. Results and Discussion
This chapter presents the results of the methodology introduced above for the development of
IDF curve for 2080s. Results are presented for the City of London.
4.1 Selection of Appropriate Stations

The number of stations used to generated long sequence of rainfall series influence outputs of
weather generator. Stations surrounding the point of interest help to capture the spatial and
temporal characteristics in the region. In cases where only limited data is available, surrounding
stations may help to add spatial and temporal characteristics of the rainfall values. Conversely,
use of too many stations can be computationally expensive and unnecessary; especially for short
duration rainfall where convective storms are highly localized weather patterns, operating on
relatively small spatial scales. Stations located too far may affect the performance of weather
generator. So regression and cross correlation analysis are performed for identifying important
stations for London. For regression analysis, the stations are grouped based on selected distances
from London (Table 3). Regression results for each group are provided in the Appendix A. The
results are expressed in terms of t-test statistics, p values and the coefficient of determination.
Results from the Appendix A show significant t-test statistic for all predictors reducing the
possibility of over-fitting by an insignificant predictor. The term probability value (p) denotes
the results of the testing of hypothesis that the regression coefficient is equal to zero which in
turn quantifies the importance of the regressor. A result of in the probability value column for a
predictor X denotes that with (100) X (1- )% of confidence one can reject the hypothesis that
37
the coefficient of predictor X is zero. Low or near zero value of is desirable as it is inversely
related to the importance of a predictor.
Table 3: Groups for Regression Analysis based on Distances
Groups based on distances (km)
0-200 0-175 0-150 0-125 0-100
Barrie WPCC
Brantford MOE
Chatham WPCP
Delhi
Elora
Fergus
Hamilton A
Hamilton RBG
London Intl A
Owen Sound MOE
Sarnia
St. Thomas WPCP
Stratford MOE
Toronto
Toronto Intl A
Waterloo A
Wiarton A
Woodstock
Total
18
16
14
13
10
Stations
0-75 0-50
The t-statistics for the independent variables are equal to their coefficient estimates divided
by their respective standard errors. In theory, the t-statistic of any one variable may be used to
test the hypothesis that the true value of the coefficient is zero (which is to say, the variable
should not be included in the model). In a standard normal distribution, only 5% of the values
fall outside the range plus-or-minus 2 A low t-statistic (or equivalently, a moderate-to-large
exceedance probability) for a variable suggests that the SEE would not be adversely affected by
its removal. The rule-of-thumb in this regard is to remove the least important variable if its tstatistic is less than 2 in absolute value, and/or the exceedance probability is greater than .05
38
(Minitab help, 2009). From the results it is seen that stations within 100 km distances appear to
be the best option for London. This can be clearly seen from the coefficient of determination plot
in Figure 3 where addition of more stations, beyond 100 km distance apparently cannot improve
the model performance.
80%
20
18
16
14
12
10
8
6
4
2
0
Coef. of Determination (R-sq)

No of Stations
75%
70%
65%
60%
55%
50%
0
50
100
150
200
250
Distance, km
Figure 3: Performances of Stations based on Distance
Next, the cross correlation analysis is performed to identify the correlation between the
stations (Table 4). Results show that stations within 100 km radius are correlated well, with
correlation greater than 60% for all stations but Elora. However, the regression test shows that
inclusion of Elora may provide important information to the spatial and temporal pattern of
London. So it is included for IDF analysis. So finally, nine stations with hourly and daily rainfall
data from 1965-2001, located within 100 km radius of London station have been selected for
further analysis.
Both historical daily and hourly data contains missing values. Inverse distance weighted
methods is applied to fill the missing values.
39
Table 4: Cross-Correlation Results for Stations Within 200 km Distance from London
Lag
Distance
Stations
(km)
-2
London A
-0.004 0.094 1.000 0.094 -0.004
Waterloo A
78
-0.004 0.063 0.729 0.097 0.000
Woodstock
33
0.003
0.273 0.723 0.022 -0.012
Sarnia
93
0.018
0.131 0.676 0.054 -0.014
Hamilton A
100
0.003
0.050 0.670 0.136 -0.008
Delhi CS
52
-0.004 0.236 0.657 0.020 -0.014
Brantford MOE
75
0.000
Stratford MOE
39
-0.005 0.263 0.633 0.028 -0.004
Hamilton RBG
106
0.002
Toronto Intl A
142
-0.002 0.043 0.610 0.125 0.000
-1
0.249 0.645 0.026 -0.019
0.214 0.618 0.037 -0.008
St. Thomas WPCP 28
-0.009 0.344 0.609 0.030 -0.008
Fergus
102
-0.005 0.203 0.564 0.033 -0.009
Toronto
158
-0.005 0.158 0.564 0.028 -0.012
Elora
91
-0.002 0.199 0.550 0.066 0.002
Chatham WPCP
113
-0.006 0.278 0.488 0.008 -0.013
Barrie
190
-0.010 0.122 0.461 0.062 -0.007
Wiarton A
190
-0.021 0.049 0.454 0.101 -0.009
Owen Sound
173
-0.024 0.146 0.373 0.047 -0.005
4.2 Development of Climate Change Scenarios

Climate change scenarios from AOGCM outputs (Table 2) are used to condition the input
data using the weather generator. Outputs from AOGCMs for 1961-1990 represent baseline
climate against which the future climate change scenarios for 2071-2099 (2080s) have been
computed.
Based on the AOGCM data, change fields for each scenario is calculated as the difference
between the monthly mean precipitation from their 1961-1990 mean.
40
This difference is then multiplied with the locally observed station data to generate climate
change scenarios appropriate for the City of London at a daily time scale. As an example, if the
change field for the month of July and August are 10% and -5%, all daily July and August
rainfall values are multiplied by a factor of 1.05 and 0.95, respectively. This newly modified
data is then used with the weather generator to generate daily time series of any preferred length
for different scenarios. For this study, 27 different climate scenarios are developed which
represent different realizations of future. Comparison of 1961-1990 mean historical observed
rainfall with those developed from different scenarios for base climate reveal that significant bias
still exit in the base climate which is used to initialize the future climate; which means the bias
may be carried out in the downscaled output. Table 5 (a) presents a comparison between the
monthly mean precipitation from different AOGCM scenarios and historical observed values.
Mean monthly precipitation vary significantly for between months for all models.
Table 5 (a): Monthly Mean Precipitation (mm) from Different AOGCMs for 1965-1990
Scenarios/month
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Observed
CGCM3T47
2.28
1.99
2.21
1.95
2.51
2.25
2.65
2.71
2.48
2.88
2.80
2.66
2.46
2.12
2.79
2.11
3.04
2.35
2.66
2.29
3.17
2.88
3.16
2.73
CGCM3T63
2.14
1.80
2.45
2.84
3.69
3.42
3.00
2.40
2.30
2.87
2.67
2.93
CSIROMK3
1.94
2.16
2.44
2.97
3.28
2.64
2.42
1.80
1.72
2.31
2.42
2.21
ECHAM5OM
3.01
3.63
3.62
4.11
4.33
4.41
3.58
3.47
3.32
2.47
2.99
3.09
ECHO-G
2.08
2.10
2.49
3.43
4.45
3.66
3.82
3.18
2.59
2.67
2.92
2.21
GFDLCM2.1
2.46
2.83
2.86
2.90
3.54
3.19
3.23
3.04
3.36
2.29
2.83
2.62
GISSAOM
2.04
2.22
2.51
2.79
2.54
2.21
2.59
2.91
3.18
3.04
2.57
2.56
MIROC3.2_HIRES
2.88
2.56
2.97
3.32
3.01
3.23
3.76
3.34
3.40
2.90
3.28
3.14
MIROC3.2_MEDRES
2.21
2.57
2.64
2.86
2.92
3.49
3.71
3.00
2.96
2.47
2.52
2.40
CCSRNIES_B21
1.84
2.24
2.86
3.25
3.63
4.18
4.77
3.52
2.07
1.40
1.73
2.05
CSIROMk2b_B11
1.51
1.53
1.74
2.34
2.50
3.21
3.26
2.20
1.71
1.88
1.79
1.64
4.3 Verification of the IDF Generation Methods

Once preparation of data is complete, daily weather generator described in section 3.2.1 is
41
used to simulate a sequence of rainfall for all stations. For the verification purpose, the
perturbation of the weather generator described in section 3.3.1 is kept off in order to replicate
the exact scenario as the historical, observed, one.
This study uses 10 stations for the period of 1965-2003 (N=39) to simulate different rainfall
scenarios. Employing the temporal window of 14 days (w=14) and 39 years of historic data
(N=39), 584 days are considered as potential neighbors (L=(w+1) x N-1=584). Each case is
simulated three times thus generating 117 years of simulated output. It is expected that such
length of output is sufficient enough to estimate event with return period of 100 years.
In order to test the output of the weather generator, the box and whisker plots for monthly
historical simulated rainfall are created (Figure 4).
200
180
Monthly Rainfall (mm)
160
140
120
100
80
60
40
20
0
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Figure 4: Box and Whiskers Plot of Simulated Monthly Rainfall in London
The boxes show the 25th percentile, 50 percentile and 75th percentiles of data while the
whiskers, plotted with 1.5 times the inter-quartile range from the boxes. For all cases historic
42
observed means are shown in terms of line plot to assess the ability of the weather generator to
reproduce the temporal and spatial character of rainfall for the City of London. From the Figure
4, it is seen that the model has been able to replicate the historic observed pattern adequately.
Next, the daily rainfall is disaggregated into hourly values using the method described in
section 3.2.3. The comparison of the performance of the historic simulated hourly values with
the observed hourly data is presented in terms of frequency plots (Figure 5).
0
12 18 24 30 36
Obs
Sim
900
800
Frequency
700
600
500
400
300
200
100
0
0
12 18
24 30 36
Hourly Precipitation (mm)
Figure 5: Frequency Plots of Observed (Obs) and Simulated (Sim) Hourly Rainfall
The frequency of small range rainfall is slightly over-estimated and the mid range rainfall is
slightly under-estimated by the disaggregation model. Overall, the frequency of the extreme
rainfall is captured well.
Finally, the annual maximum rainfall for 1, 2, 6, 12 and 24 hour durations is generated to fit
Gumbel distribution for calculating return periods. These are then compared with the IDF
information obtained from Environment Canada (EC) (Table 6). It should be noted that the
43
Environment Canada uses rainfall data from 1943-2001 to develop IDF curves for London.
However, hourly data is available only from 1961; data prior to 1961 may exist in paper form
and are not available. For the present study, the hourly rainfall data for London is further reduced
down to 1965 for matching rainfall data from other nearby stations to be used for multi-site
weather generator.
Table 6 (a): Comparison of Extreme Rainfall in London in terms of Depth (mm)
Historic Unperturbed (1965-2003)
Return Period, T years
Duration, hrs
10
25
50
100
21.80 30.38 36.06 43.24 48.56
53.85
28.05 40.11 48.09 58.18 65.66
73.09
36.41 49.90 58.83 70.11 78.49
86.80
12
42.61 56.33 65.41 76.89 85.40
93.86
24
49.70 64.63 74.52 87.01 96.28 105.48
EC (1943-2003)
Return Period, T years
Duration, hrs
10
25
50
100
24.40 35.30 42.50 51.60 58.30
65.00
29.60 41.60 49.50 59.60 67.00
74.40
36.70 48.20 55.80 65.40 72.50
79.60
12
43.00 54.70 62.50 72.40 79.70
87.00
24
51.30 66.80 77.10 90.00 99.60 109.20
Table 6 (a) presents the intensity-duration-frequency data obtained from the historic
unperturbed scenario together with the IDF data generated by EC. The results obtained are
compared in terms of the relative differences using the following relationship:
(3.28)
44
Table 6 (b) presents the relative difference of rainfall intensity between the historic
unperturbed and the EC data. The short duration rainfall (1 hr) is underestimated by the historic
unperturbed scenario, while the intermediate (2, 6, 12 hrs) and longer (24 hrs) duration rainfalls
are able to closely replicate the EC generated intensities for all return periods. Overall, the
performance of the historic unperturbed scenario is satisfactory.
Table 6 (b): Relative Difference between EC IDF Information and Historic Unperturbed
Scenario
Duration, min
Return Period, years

2
10
25
50
100
60
11.25 14.98 16.39 17.63 18.22 18.76
120
5.38
3.66
2.89
2.42
2.02
1.78
360
0.80
3.46
5.29
6.96
7.93
8.65
720
0.91
2.94
4.56
6.02
6.91
7.58
1440
3.18
3.30
3.40
3.37
3.39
3.46
4.4 IDF Results for Future Climate

The perturbation process inside the weather generator is added (described in section 3.2.1) to
generate IDF information using the historical observed rainfall. This scenario called historical
perturbed assumes that the future climate will continue to change as the consequence of already
altered green house gas concentrations in the atmosphere, ignoring any future change in green
house gas emissions.
The daily weather generator output, after being disaggregated into hourly rainfall, is next
used to generate intensity duration frequency data for 27 different scenarios presented in Table 4
45
to create different realizations of future climate using different AOGCM responses. Appendix D
presents the IDF data obtained using climate scenarios in terms of intensity.
The difference in the AOGCM scenarios relative to the historic perturbed scenario is
summarized in Table 7.
Table 7: Percent Differences between Historic Perturbed, Wet and Dry Scenarios
ECHAM5AOM_A1B (Wet Scenario) and Historic Perturbed
Return Period, years
Duration, min
10
25
50
100
60
62.68
69.56
72.42
75.00
76.44
77.60
120
60.64
65.84
67.98
69.91
70.99
71.86
360
65.13
77.09
82.29
87.12
89.88
92.13
720
66.03
77.77
83.09
88.17
91.13
93.57
1440
63.22
72.99
77.42
81.63
84.07
86.09
MIROC3MEDRES_A2 (Dry Scenario) and Historic Perturbed

Return Periods, years
Duration, min
10
25
50
100
60
-6.79
-2.90
-1.28
0.18
0.99
1.65
120
-12.70
-15.09
-16.07
-16.96
-17.45
-17.85
360
-7.06
-6.60
-6.40
-6.21
-6.10
-6.02
720
-0.68
1.66
2.72
3.73
4.32
4.81
1440
-0.44
-0.10
0.05
0.20
0.28
0.35
The model results show variable results, with wide range of increase in extreme rainfall.
ECHAM5AOM A1B appears to be the wettest while MIROC3.2MEDRES A2 being the driest of
all. The wettest ECHAM5AOM A1B model shows more than 60% increase in rainfall compared
to historic perturbed scenario. While the driest MIROC3.2 MEDRES A2 scenario shows slight
decrease in precipitation intensity than the historic perturbed scenario. The difference between
the wettest and the driest scenario ranges from 70% to 92% indicating huge range of uncertainty
46
among the realizations of AOGCMs. A comparison of different AOGCMs for specific duration
is presented in Figure 6.
47
1 hr
120
120
80
100
60
80
140
120
100
80
60
40
60
40
40
20
20
20
40
60
80
12 hr
220
100 120
260
6 hr
200
180
160
140
100
Depth, mm
2 hr
160
20
0
20
40
60
80
100 120
20
40
60
80
100 120
24 hr
220
Depth, mm
180
180
140
140
100
100
60
60
20
0
20
20
40 60 80 100 120 0
20 40 60 80 100 120
Return Period, yrs
Return Period, yrs
Figure 6: IDF Plots of AOGCM Scenarios for Different Durations
48
4.5 Uncertainty Quantification of IDF Results
Because of the inherent uncertainties, the newly developed IDF curves from different
AOGCMs are unable to provide an accurate estimate of future extreme rainfall, but they
establish a significant fact: the future climate will not be the same as the historic one. Previous
studies (Simonovic and Peck, 2009; Prodanovic and Simonovic, 2007) have generated updated
IDF information for the City of London for 2050s (2041-2070) based on a single scenario
(CCSRNIES B21) selected from the upper range of all scenarios presented in this study. In
presence of uncertainties presented in section 4.4, adoption of one single scenario may suffer
from under/over-estimation of the risks, which may have significant implications for the storm
water management and design practice.
So, a kernel estimator based on the data driven plug-in approach described in section 3.2.6 is
applied next to quantify the uncertainty arising from different AOGCM scenarios. Due to the
fact that unlike other uncertainty estimation methods, kernel estimator provides variable weights
at each point of interest, weights are calculated from the mean of all AOGCM data for
presentation purpose. The weight function is calculated by modifying equation 3.18 as follows:
(
where, for any time period ,
represents any data point within the ranges of generated data
for which kernel estimator is applied,

bandwidth , and
is the AOGCM simulated data,
is the plug-in
= number of AOGCM models and scenarios considered =27.
49
The mean of total sample size

be measured, where
is considered as the data point from which the distance will

years of simulated IDF data= 117.
Figure 7 and Appendix E present the IDF curves incorporating the most useful information
from the AOGCM scenarios. In this case, four scenarios have been selected: the historical
perturbed scenario as future state ignoring climate change, ECHAM5AOM A1B scenario as
the wettest scenario, MIROC3MEDRES A2 as the driest and the resultant scenario as the
scenario incorporating uncertainties from all AOGCM models and scenarios. The IDF curves of
these selected scenarios for all durations for specific durations are presented in Appendix E.
Table 9 presents the percent difference between the historical perturbed and resultant scenarios
for 2080s. From the Table 9 it is seen that due to the changing climate, the intensity of rainfall is
expected to increase by 20-40 % in 2071-2099.
50
2 Year Return Period
100
75
10
Historical Perturbed
ECHAM5AOM_A1B (Wet)
MIROC3MEDRES_A2 (Dry)
Resultant
50
25
Intensity (mm/hr)
25
10
25
10
ECHAM5AOM_A1B (Wet)
Resultant
ECHAM5AOM_A1B (Wet)
Resultant
1
60
120
360
720
1440
Duration (min)
1
60
50
50
25
25
10
10
120
50
100
75
100
75
360
720
1440
360
720
1440
360
720
1440
50
25
10
ECHAM5AOM_A1B (Wet)
Resultant
1
Duration (min)
120
Duration (min)
ECHAM5AOM_A1B (Wet)
Resultant
1
120
60
Duration
(min)
Year
Return
Period
100
75
ECHAM5AOM_A1B (Wet)
Resultant
60
100
75
50
Intensity (mm/hr)
Inte nsity (mm/hr)
50
Intensity (mm/hr)
100
75
60
120
360
720
1440
Duration (min)
60
120
360
720
1440
Duration (min)
Figure 7: Comparison of IDF Plots for Different Scenarios

51
Table 9: Difference between Historical Perturbed and the Resultant Scenario for 2080s
Duration (min)
Return Period, yrs

2
10
25
50
100
60
21.76 25.29 26.76 28.08 28.82 29.42
120
17.40 21.41 23.06 24.55 25.38 26.04
360
20.85 27.32 30.14 32.75 34.25 35.47
720
22.20 28.94 31.99 34.91 36.61 38.01
1440
20.63 25.77 28.10 30.32 31.60 32.67
Finally, the probabilities of extreme rainfall for all return periods are presented in
terms of cumulative distribution plots. First, IDF plot of the resultant scenario is created
(Figure 8).
Resultant
175
150
125
100
Depth (mm)
75
50
2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
25
10
60
120
360
720
1440
Duration (min)
Figure 8: IDF Plot for Resultant Scenario
The cumulative distribution plots using the IDF from the AOGCM scenarios are plotted
for all return periods and are presented in Figures 9 (a) and (b). Information from Figures
52
8 and 9 are combined to gather probabilities for any specific storm for any specific return
period. For example, if the depth of 6 hour (360 min) storm for 5 year return period is 75
mm (Figure 8), the maximum probability of this specific storm can be counted as
approximately 0.66 (Figure 9 and so on. This additional probability information will
allow users to use the updated IDF information with more confidence.
Cumulative Distribution Function 1 hr
1
0.9
0.8
100yr
Probability
0.7
50yr
0.6
25yr
0.5
10yr
0.4
5yr
0.3
2yr
0.2
0.1
0
0
10
20
30
40
50
60
70
80
90 100 110 120 130 140
Depth (mm)
1
0.9
0.8
100yr
Probability
0.7
50yr
0.6
25yr
0.5
10yr
0.4
5yr
0.3
2yr
0.2
0.1
0
0
20
40
60
80
100
120
140
160
180
200
Depth (mm)
Figure 9 (a): Probability based IDF Curve of 1 and 2 Hour Duration
53
0.9
0.9
0.8
0.8
100yr
100yr
Probability
50yr
0.6
25yr
0.5
10yr
0.4
5yr
0.3
10yr
0.5
0.2
0.1
0.1
40
60
5yr
0.4
2yr
0.3
2yr
20
25yr
0.6
0.2
0
0
50yr
0.7
80
100
120 140 160

Depth (mm)
180
200
220
240
0
0
260
20
40
60
80 100 120 140 160 180 200 220 240 260 280 300
Depth (mm)
1
0.9
0.8
0.7
Probability
Probability
0.7
100yr
0.6
50yr
0.5
25yr
0.4
10yr
0.3
5yr
0.2
2yr
0.1
0
0
20
40
60
80 100 120 140 160 180 200 220 240 260 280 300
Depth (mm)
Figure 9 (b): Probability based IDF Curve of 6, 12, 24 Hour Duration
54
5. Conclusions
This study presents the methodology for updating of rainfall IDF curves for the City of London
incorporating uncertainties associated with the use of different AOGCMs. The analysis of the
annual maximum rainfall for developing intensity-duration-frequency plots for the City of
London under climate change has resulted in important findings. Overall, two objectives have
been achieved by this study: first, an extensive investigation of the possible realizations of future
climate from 29 scenarios developed from AOGCM models and scenarios are performed using a
downscaling based disaggregation approach. A nonparametric K-Nearest Neighbor multi-site
weather generator operating on a daily time step is used to produce long sequence of rainfall
data. The use of perturbation scheme has overcome the limitation of assumptions of stationarity
by generating data beyond the range of the input. The downscaled daily outputs are
disaggregated into hourly values by a non-parametric nearest neighbor based disaggregation
scheme. Annual maximum series of rainfall are fitted to Gumbel distribution to develop IDF
curves for 1, 2, 6, 12 and 24 hour durations for 2, 5, 10, 25, 50 and 100 years of return periods.
Next, the associated uncertainties are estimated using non-parametric kernel estimation approach
and the resultant IDF curve is developed based on a probabilistic approach.
The basic findings from the study are presented as follows:
The rainfall patterns in the City of London will most certainly change in future due to
climate change.
55
Generation of future IDF information based on single site is limited. Incorporating a

multi-site weather generator to produce sequences of future rainfall offers a more reliable
approach for providing better spatial and temporal characteristics of rainfall patterns.
Adoption of a single scenario for developing IDF information only provides a single
realization of the future; application of a multi-model approach can provide more
realistic information about the future climate.
Use of the wettest or the driest scenario may be useful to capture the upper and lower
bound scenario of the future climate change; however, single use of any of these
scenarios may suffer from over/underestimation of the rainfall extremes with serious
implications on storm water management practice and the development of design
standards.
Although the scenarios developed from different scenarios indicate large uncertainty
associated with the global climate models, all of them indicate increase in intensity of
future rainfall with a varying degree.
A kernel based plug-in estimation approach is able to incorporate the uncertainties

arising from different AOGCM models and to provide a more acceptable change in
future rainfall extremes. The resultant scenario indicates approximately 20-40% changes
in different duration rainfall for all return period.
Use of a probability based intensity-duration-frequency curve is encouraged in order to

apply the updated IDF information with higher level of confidence.
56
Acknowledgments
The authors wish to thank DAI/CCCSN/ EC for providing observation data for the work.
Financial assistance from the Canadian Foundation for Climate and Atmospheric Sciences is
thankfully acknowledged.
57
References
Adamowski, K. (1985). Nonparametric kernel estimation of flood frequencies, Water Resources
Research, 21 (11), 1585-1590.
Allan, M. R., and W. J. Ingram (2002). Constraints on future changes in climate and the
hydrologic cycle, Nature, 419, 224 232.
Brissette, F., R. Leconte, M. Khalili (2007). Efficient stochastic generation of multi-site synthetic
precipitation data. Journal of Hydrology, 345,121-133.
Buishand, T. A. and Demare, G. R. (1990). Estimation of the Annual Maximum Distribution
from Samples of Maxima n Separate Seasons. Stochastic Hydrol. Hydraul. 4, 89-103.
Buytaert, W., R. Celleri, and L. Timbe (2009). Predicting climate change impacts on water
resources in the tropical Andes: Effects of GCM uncertainty, Geophysical Research Letters,36,
L07406, doi:10.1029/2008GL037048.
Coles S. (2001). An Introduction to Statistical Modeling of Extreme Values. London: Springer;
2001.
Coulibaly, P. and X. Shi. (2005). Identification of the Effect of Climate Change on Future
Design Standards of Drainage Infrastructure in Ontario. Report prepared by McMaster
University with funding from the Ministry of Transportation of Ontario. 82 pp.
Cubasch, U., G. A. Meehl, G. J. Boer, R. J. Stouffer, M. Dix, A. Noda, C. A. Senior, S. Raper,
and K. S. Yap (2001). The Scientific Basic. Contribution of working group 1 to the Third
Assessment Report of the Intergovernmental Panel of Climate Change, Cambridge University
Press, Cambridge, UK and New York, USA, 881 pp.
58
CSA (2010). Development, Interpretation and Use of Rainfall Intensity- Duration-requency

(IDF) Information: A Guideline for Canadian Water Resources Practitioners. 1st Edn., Canadian
Standards Association, CSA PLUS 4013-10.
DAI Catalogue, (2009). Catalogue available datasets through DAI (Data Access and Integration),
version 1.0, April 2009, Montreal, QC, Canada, 24 pp.
De Michele, C., Salvadori, G., (2005). Some hydrological applications of small sample
estimators of generalized Pareto and extreme value distributions. Journal of Hydrology 301, 37
53.
Desramault, N. (2008). Estimation of Intensity Duration Frequency Curves for Current and
Future Climates. Thesis submitted to the Graduate and Postdoctoral Studies Office in partial
fulfilments of requirements of the degree of Master of Engineering Department, McGill
University, Montreal, Qubec, Canada. 75 pp.
Efromovich, S. (1999). Nonparametric Curve Estimation: Methods, Theory, and Applications,
Springer, New York.
Eum, H. I., Arunachalam, V. and Simonovic, S. P. (2009). Integrated Reservoir Management
System for Adaptation to Climate Change Impacts in the Upper Thames River Basin. Water
Resources Research Report no. 062, Facility for Intelligent Decision Support, Department of
Civil and Environmental Engineering, London, Ontario, Canada, 81 pages. ISBN: (print) 9780-7714-2710-7; (online) 978-0-7714-2711-4.
Hall, P., and Marron, J. S. (1991). Lower Bounds for Bandwidth Selection in Density Estimation, Probability Theory and Related Fields, 90, 149-173.
59
Huard, D., Mailhot, A., and Duchesne, S. (2010). Bayesian Estimation of Intensity-DurationFrequency Curves and of the Return Period Associated to a Given Rainfall Event. Stoch.
Environ. Res. Risk. Assess., 2, 337-347.
Hughes, J. P., and P. Guttorp (1994). A class of stochastic models for relating synoptic
atmospheric patterns to regional hydrologic phenomena, Water Resources Research, 30(5),
15351546.
IPCC (2007). Climate Change 2007: The Physical Science Basis. Contribution of working group
1 to the Fourth Assessment Report of the Intergovernmental Panel of Climate Change,
Annexes [Baede, A.P.M. (ed.)]. Cambridge University Press, Cambridge UK and New York,
USA, 48pp.
Jaruskova, D. and Hanek, M. (2006). Peak Over Threshold method in Comparison with Block
Maxima Method for Estimating High Return Levels of Several Northern Moravia Precipitation
and Discharges Series. J. Hydrol. Hydromech., 4, 309-319.
Jarvis, C.S. et al., (1936). Floods in the United States, Magnitudes and Frequency. USGS Water
Supply Paper 771, Washington, DC.
Jones, M.C., J. S. Marron, and S. J. Sheather (1996). A brief survey of bandwidth selection for
density estimation, Journal of the American Statistical Association, 91(433), 401407.
Katz, R. W., Parlange, M. B. and Naveau, P. (2002). Statistics of Extremes in Hydrology.
Advances in Water Resources, 25, 1287-1304.
Lall, U. (1995). Nonparametric Function Estimation: Recent Hydrologic Contributions. Reviews
of Geophysics, Contributions in Hydrology, U.S. National Report to the IUGG 1991-1994,
1093-1099.
Lall, U., B. Rajagopalan and D. G. Tarboton (1996). A Nonparametric Wet/Dry Spell Model
60
for Daily Precipitation, Water Resources Research, 32(9), 2803-2823.

Langbein, W.B., (1949). Annual floods and the partial duration flood series. Transactions of
AGU 30, 879881.
Madsen, H., Rosbjerg, D., (1997). Generalized least squares and empirical Bayes estimation in
regional partial duration series index-flood modeling. Water Resources Research 33 (4), 771
781.
Mailhot, A. and Duchesne, S. (2010). Design criteria of urban drainage infrastructures under
climate change, J Water Resour Plann Manage 136, 201208.
Mailhot, A., Duchesne,S., .Rivard, G., Nantel, E., Caya, D., and Villeneuve, J.-P. (2006).
Climate change impacts on the performance of urban drainage systems for Southern Qubec.
Proceedings of EIC Climate Change Technology Conference, Ottawa, Ontario, Canada.
Mansour, R. and Burn, D. H. (2010). Weather Generator and Hourly Disaggregation Model.
Department of Civil Engineering, University of Waterloo, Waterloo, Ontario, Canada.
Mailhot, A., Duchesne,S., Caya, D., and Talbot, G. (2007). Assessment of future change in
intensitydurationfrequency (IDF) curves for Southern Quebec using the Canadian Regional
Climate Model (CRCM). Journal of Hydrology, 347 (1-2), 197-210.
Mehdi, B., Mrena, C., and Douglas, A. (2006). Adapting to climate change: An introduction for
Canadian Municipalities. Canadian Climate Impacts and Adaptation Research Network (CCIARN), http://www.c-ciarn.ca/.
MTO (1997). Ministry of Transportation of Ontario Drainage Management Manual. Drainage
and Hydrology Section, Transportation Engineering Branch, and Quality Standards Division,
Ministry of Transportation of Ontario, Ottawa, Ontario, Canada.
61
Nakicenovic, N., Alcamo, J., Davis, G., de Vries, B., Fenhann, J., and co-authors. (2000). IPCC
Special Report on Emissions Scenarios. UNEP/GRID-Ardenal Publications.
New, M., and M. Hulme, (2000), Representing uncertainty in climate change scenarios: a MonteCarlo approach, Integrated Assessment, 1, 203-213.
Nguyen, T-D., V-T-V. Nguyen, and P. Gachon. (2007a). A spatial-temporal downscaling
approach for construction of intensity-duration-frequency curves in consideration of GCMbased climate change scenarios. Advances in Geosciences, 6, 11-21.
Nguyen, V-T-V., T-D. Nguyen and A. Cung. (2007b). A statistical approach to downscaling of
sub-daily extreme rainfall processes for climate-related impacts studies in urban areas. Water
Science and Technology: Water Supply, 7 (2), 183-192.
Onof, C., Arnbjerg-Nielsen, K., (2009). Quantification of anticipated future changes in high
resolution
design
rainfall
for
urban
areas.
Atmospheric
Research
92,
350363.
doi:10.1016/j.atmosres.2009.01.014.
Park, B. U. and Marron, J. S. (1990) Comparison of data-driven bandwidth selectors. J. Am.
Statist. Ass., 85, 66-72.
Predrag Prodanovic and Slobodan P. Simonovic (2007). Development of rainfall intensity
duration frequency curves for the City of London under the changing climate. Water Resources
Research Report no. 058, Facility for Intelligent Decision Support, Department of Civil and
Environmental Engineering, London, Ontario, Canada, 51 pages. ISBN: (print) 978-0-77142667-4; (online) 978-0-7714-2668-1.
Prudhomme, C., D. Jakob, and C. Svensson (2003). Uncertainty and climate change impact on
the flood regime of small UK catchments, Journal of Hydrology, 277, 1 23.
62
Rasmussen, P.F., Ashkar, F., Rosbjerg, D., Bobee, B., (1994). The POT method for flood
estimation: a review. In: Hipel, K.W. (Ed.), Stochastic and Statistical Methods in Hydrology
and Environmental Engineering, Extreme Values: Floods and Droughts, vol. 1. Kluwer,
Dordrecht, NL, pp. 1526.
Rosenblatt (1956). Remarks on some nonparametric estimates of a density function. Ann. Math.
Statist. 27, 832-837.
Scott, D. W., and G. R. Terrell, (1987). Biased and Unbiased Cross- Validation in Density
Estimation, Journal of the American Statistical Association, 82, 1131-1146.
Sharif, M., Burn, D. H. and Wey, K.M. (2007). Daily and Hourly weather generation using a KNearest Neighbor Approach, 18th CSCE Hydrotechnical Conference, August 2007, Winnipeg,
MB, 1 - 10.
Sharif, M., and D. H. Burn (2006). Simulating climate change scenarios using an improved Knearest neighbor model, Journal of hydrology, 325, 179-196.
Sharma, A., D. G. Tarbaton, and U. Lall, (1997). Streamflow simulation a nonparametric
approach, Water Resources Research, 33 (2), 291-308.
Sheather, S. J. (1983). A Data-Based Algorithm for Choosing the Window Width When
Estimating the Density at a Point, Computational Statistics and Data Analysis, 1, 229-238.
Sheather, S. J. (1986). An Improved Data-Based Algorithm for Choosing the Window Width
When Estimating the Density at a Point, Computational Statistics and Data Analysis, 4, 61-65.
Sheather, S. J., and M. C. Jones (1991). A Reliable Data-Based Bandwidth Selection Method for
Kernel Density Estimation, Journal of the Royal Statistical Society, Ser. B, 53, 683-690.
Silverman, B. W. (1986). Density estimation for statistics and data analysis. Monographs on
Statistics and Applied Probability, Chapman & Hall/ CRC., Washington, D.C.
63
Slobodan P. Simonovic and Angela Peck (2009). Updated rainfall intensity duration frequency
curves for the City of London under the changing climate. Water Resources Research Report
no. 065, Facility for Intelligent Decision Support, Department of Civil and Environmental
Engineering, London, Ontario, Canada, 64 pages. ISBN: (print) 978-0-7714-2819-7; (online)
978-0-7714-2820-3.
Smith, R. L., Tebaldi, C., Nychka, D., and Mearns, L (2009). Bayesian modeling of uncertainty
in ensembles of climate models, Journal of the American Statistical Association, 104 (485),
97-116., doi:10.1198/jasa.2009.0007.
Smith, R.L., (2003). Statistics of extremes, with applications in environment, insurance and
finance. <http://www.stat.unc.edu/postscript/rs/semstatrls.ps>.
Solaiman and Simonovic (2011). Quantifying Uncertainties in the Modelled Estimates of
Extreme Precipitation Events at Upper Thames River Basin. Water Resources Research Report
Engineering, London, Ontario, Canada, 64 pages.
Stainforth, D. A., T. E. Downing, R. W. A. Lopez, and M. New (2007). Issues in the
interpretation of climate model ensembles to inform decisions, Philos. Trans. R. Soc., Ser. A,
365, 21632177.
Stone, D.A., Weaver, A.J., Zwiers, F.W., (2000). Trend in Canadian precipitation intensity.
Atmos. Ocean 38 (2), 321347.
Svensson, C., Clarke, R., Jones, D. (2007). An Experimental Comparison of Methods for
Estimating Rainfall Intensity-Duration-Frequency Relations from Fragmantory Methods.
Journal of Hydrology, 341, 79-89.
64
Taesombat, V., Yevjevich, V., 1978. Use of partial flood series for estimating distribution of
maximum annual flood peak. Hydrology Papers #97, Colorado State University, Fort Collins,
CO.
Trigo, R.M. and J.P. Palutikof (2001). Precipitation scenarios over Iberia: a comparison between
direct GCM output and different downscaling techniques. J. Climate 14, 4422-4446.
Vincent, L.A., and E. Mekis (2006). Changes in daily and extreme temperature and precipitation
indices for Canada over the twentieth century. Atmosphere-Ocean, 44(2), 177- 193.
Widmann M, Bretherton CS, Salathe-Jr EP. (2003). Statistical precipitation downscaling over
the North-western United States using numerically simulated precipitation as a predictor.
Journal of Climate 16(5), 799-816.
Wilby, R. L., and I. Harris (2006). A framework for assessing uncertainties in climate change
impacts: Low-flow scenarios for the River Thames, UK, Water Resources Research, 42,
W02419, doi: 10.1029/2005WR004065.
65
APPENDIX A: Regression Test Results
Regression Results for Stations Within 0-200 km Radius Distance

Predictor
t-Statistic Probability (p)
Hamilton RBG
-8.52
0.000
Hamilton A
20.54
0.000
Fergus
-3.91
0.000
Elora
-4.17
0.000
Delhi
10.71
0.000
Chatham WPCP
0.27
0.791
Brantford MOE
-6.15
0.000
Woodstock
26.58
0.000
Waterloo A
28.64
0.000
St. Thomas WPCP 10.53
0.000
Stratford MOE
8.06
0.000
Sarnia
25.15
0.000
Barrie
3.92
0.000
Owensound
-2.82
0.005
Wiarton
7.49
0.000
Toronto City
-4.75
0.000
Toronto Intl A
4.69
0.000
66

Predictor
t-Statistic
Probability (p)
Woodstock
26.43
0.000
0.000
Stratford MOE
8.34
0.000
Delhi
10.81
0.000
Brantford MOE
-6.42
0.000
Waterloo A
29.85
0.000
Sarnia
25.80
0.000
Elora
-4.05
0.000
Hamilton A
21.40
0.000
Hamilton RBG
-11.24
0.000
Fergus
-3.96
0.000
Chatham WPCP
0.22
0.829
Toronto Intl A
4.29
0.000
Owensound
2.38
0.017

Predictor
Woodstock
26.36
0.000
0.000
Stratford MOE
8.92
0.000
Delhi
10.83
0.000
Brantford MOE
-6.50
0.000
Waterloo A
29.82
0.000
Sarnia
25.97
0.000
Elora
-4.01
0.000
Hamilton A
21.43
0.000
Hamilton RBG
-11.26
0.000
Fergus
-3.62
0.000
Chatham WPCP
0.33
0.743
Toronto Intl A
4.36
0.000
67

Predictor
Woodstock
26.30
0.000
0.000
Stratford MOE
8.77
0.000
Delhi
10.63
0.000
Brantford MOE
-6.76
0.000
Waterloo A
34.25
0.000
Sarnia
26.30
0.000
Elora
-3.68
0.000
Hamilton A
23.70
0.000
Hamilton RBG
-10.81
0.000
Fergus
-3.30
0.001
Chatham WPCP
0.19
0.853

Predictor
Woodstock
25.76
0.000
0.000
Stratford MOE
7.79
0.000
Delhi
9.41
0.000
Brantford MOE
-10.98
0.000
Waterloo A
34.39
0.000
Sarnia
25.85
0.000
Elora
-8.26
0.000
Hamilton A
21.60
0.000
68

Predictor
Woodstock
30.19
0.000
St. Thomas WPCP
10.22
0.000
Stratford MOE
13.99
0.000
Delhi
15.71
0.000
Brantford MOE
2.27
0.023
Predictor
Woodstock
St. Thomas WPCP
Stratford MOE
t-Statistic
40.28
19.86
16.68
Probability (p)
0.000
0.000
0.000
69
APPENDIX B: SRES Emission Scenarios
Figure A1: SRES Emission Scenarios (Nakicenovic et al, 2000)

A1B: In scenario A1B, the storyline includes rapid economic expansion and globalization, a
population peaking at 9 billion in 2050, and a balanced emphasis on a wide range of energy
sources (Nakicenovic et al, 2000).
B1: The storyline for the B1 scenario is much like A1B in terms of population and globalization;
however there are changes toward a service and information economy with more resource
efficient and clean technologies. Emphasis is put on finding global solutions for
sustainability (Nakicenovic et al, 2000).
A2: For scenario A2, the storyline consists of a world of independently operating nations with a
constantly increasing population and economic development on a regional level.
70
Technological advances in this storyline occur more slowly due to the divisions between
nations (Nakicenovic et al, 2000).
71
APPENDIX C: Atmosphere-Ocean General Circulation Models
Criteria for Selecting Climate Scenarios

Five criteria that should be met by climate scenarios if they are to be useful for impact
researchers and policy makers are suggested by IPCC and are quoted here:
Criterion 1: Consistency with global projections. They should be consistent with a broad
range of global warming projections based on increased concentrations of greenhouse gases.

This range is variously cited as 1.4C to 5.8C by 2100, or 1.5C to 4.5C for a doubling of
atmospheric CO2 concentration (otherwise known as the "equilibrium climate sensitivity").
Criterion 2: Physical plausibility. They should be physically plausible; that is, they
should not violate the basic laws of physics. Hence, changes in one region should be physically
consistent with those in another region and globally. In addition, the combination of changes in
different variables (which are often correlated with each other) should be physically consistent.
Criterion 3: Applicability in impact assessments. They should describe changes in a
sufficient number of variables on a spatial and temporal scale that allows for impact assessment.
For example, impact models may require input data on variables such as precipitation, solar
radiation, temperature, humidity and wind speed at spatial scales ranging from global to site
and at temporal scales ranging from annual means to daily or hourly values.
Criterion 4: Representative. They should be representative of the potential range of
future regional climate change. Only in this way can a realistic range of possible impacts be
estimated.
Criterion 5: Accessibility. They should be straightforward to obtain, interpret and apply
for impact assessment. Many impact assessment projects include a separate scenario
72
development component which specifically aims to address this last point. The DDC and this
guidance document are also designed to help meet this need.
Challenges in using AOGCMs

GCMs depict the climate using a three dimensional grid over the globe (Figure), typically
having a horizontal resolution of between 250 and 600 km, 10 to 20 vertical layers in the
atmosphere and sometimes as many as 30 layers in the oceans. Their resolution is thus quite
coarse relative to the scale of exposure units in most impact assessments, hence only partially
fulfilling criterion 3. Moreover, many physical processes, such as those related to clouds, also
occur at smaller scales and cannot be properly modeled. Instead, their known properties must be
averaged over the larger scale in a technique known as parameterization. This is one source of
uncertainty in GCM-based simulations of future climate. Others relate to the simulation of
various feedback mechanisms in models concerning, for example, water vapor and warming,
clouds and radiation, ocean circulation and ice and snow albedo. For this reason, GCMs may
simulate quite different responses to the same forcing, simply because of the way certain
processes and feedbacks are modeled.
However, while these differences in response are usually consistent with the climate
sensitivity range described in criterion 1, they are unlikely to satisfy criterion 4 concerning the
uncertainty range of regional projections. Even the selection of all the available GCM
experiments would not guarantee a representative range, due to other uncertainties that GCMs
do not fully address, especially the range in estimates of future atmospheric composition.
73
Figure: 3-Dimensional Representation of Climate Models (Climate Research Unit website, 2011)
(from http://www.cru.uea.ac.uk/cru/info/modelcc/ retrieved on 3/01/2011)
74
Canadian Coupled Global Climate Model

The third generation Coupled Global Climate Model (CGCM3) was created in 2005 by the
Canadian Centre for Climate Modelling and Analysis (CCCma) in Victoria, BC for use in the
IPCC 4th assessment report to run complex mathematical equations which describe the earths
atmospheric and oceanic processes. The CGCM3 climate model includes four major
components: an atmospheric general circulation model, an ocean general circulation model, a
thermodynamic sea-ice model, and a land surface model (Hengeveld, 2000) and consists of two
resolutions, T47 and T63. The T47 version has a surface grid whose spatial resolution is roughly
3.75 degrees lat/lon and 31 levels in the vertical. The ocean grid shares the same land mask as
the atmosphere, but has four ocean grid cells underlying every atmospheric grid cell. The ocean
resolution in this case is roughly 1.85 degrees, with 29 levels in the vertical.
The T63 version has a surface grid whose spatial resolution is roughly 2.8 degrees
latitude/longitude and 31 levels in the vertical. As before the ocean grid shares the same land
mask as the atmosphere, but in this case there are 6 ocean grids underlying every atmospheric
grid cell. The ocean resolution is therefore approximately 1.4 degrees in longitude and 0.94
degrees in latitude. This provides slightly better resolution of zonal currents in the Tropics, more
nearly isotropic resolution at mid latitudes, and somewhat reduced problems with converging
meridians in the Arctic.
(Compiled from http://www.ec.gc.ca/ccmac-cccma/default.asp?lang=En&n=1299529F-1)
75
Commonwealth Scientific and Industrial Research Organizations Mk3.5 Climate Systems

Model
Australias Commonwealth Scientific and Industrial Research Organization created the
AOGCM CSIROMK3.5, which is an improved version of the MK climate systems model. The
model consists of several components: atmosphere, land surface, ocean and polar ice. The
dynamic framework of the atmospheric model is based upon the spectral method with the
equations cast in the flux form that conservs predicted variables. The atmospheric moisture
variables (vapour, water and ice) are advected by a Semi-Lagrangian Transport (SLT) algorithm
(McGregor 1993). The most recent version (MK3.5) has included a representation of the Great
Lakes and changes in land surface scheme and its representation of surface albedo under
freezing than its previous versions. The MK3.5 version provides improved information by
including the spatially varying eddy transfer coefficients (Visbeck et al, 1997) and the KrausTurner mixed layer (1967) scheme. Improvements have also been done in its oceanic behavior
in the high latitude Southern ocean, where the stratification and circulation are generally more
realistic than the prior models. The spatial resolution of the model is 1.875 1.875.
Compiled from (http://www.cawcr.gov.au/publications/technicalreports/CTR_021.pdf)
76
Max Planck Institute for Meteorologys ECHAM5AOM Model

ECHAM5 is the 5th generation of the ECHAM general circulation model. Depending on the
configuration the model resolves the atmosphere up to 10 hPa for tropospheric studies, or up to
0.01 hPa for middle atmosphere studies. The current version differ in the vertical extent of the
atmosphere as well as the relevant processes than its earlier versions. It is capable of hosting
sub-models (chemistry, aerosol and vegetation) going beyond the meteorological processes of a
AOGCM. The model can be used as a part of a coupled ocean GCM, in assimilation by linear
relaxation and as a standalone column model.
For integrations to start, the model requires several files. These file contain information for
the description of the initial or re-start state of the atmosphere (boundary conditions at the
surface, the ozone distribution and tables of constants of LW radiation schemes), the description
of assumed conditions during the integration, e.g. sea surface temperature, or the initialization of
parameterizations.
(Compiled from http://www.mpimet.mpg.de/en/science/models/echam/echam5.html)
77
Meteorological Institute, University of Bonn Meteorological Research Institute of KMA

Model and Data Groupe at MPI-Ms ECHO-G Model
The climate model ECHO-G (Legutke and Voss, 1999) is a coupled climate model consisting
of the atmospheric model ECHAM4 (Roeckner et al., 1996) and the ocean model HOPE (Wolff
et al., 1997).
The ECHAM4-model is based on primitive equations. The prognostic variables are vorticity,
divergence, logarithm of surface pressure, temperature, specific humidity, mixing ratio of total
cloud water and optionally a number of trace gases and aerosols.The vertical extension is up to a
pressure level of 10 hPa, which corresponds to a height of approximately 30km. A hybrid sigmapressure coordinate system is used with 19 irregularly ordered levels and with highest resolution
in the atmospheric boundary layer. The bottom level is placed at a height of about 30m above the
surface corresponding approximately to the surface layer. In this study the ECHAM4 model has
a horizontal resolution of about 3.75lat x 3.75lon.
The ocean model HOPE (Hamburg Ocean Primitive Equation) is an ocean general circulation
model (OGCM) based on primitive equations with the representation of thermodynamic
processes. It is a non-eddy resolving circulation model. HOPE-G has a horizontal resolution of
approximately 2.8lat x 2.8lon with a grid refinement in the tropical regions over a band from
10N to 10S. This meridional grid refinement reaches a value of 0.5 at the equator allowing for a
more realistic representation of ENSO variability in the tropical Pacific Ocean . The ocean model
has 20 vertical, irregularly ordered layers.
The coupling as well as the interpolation between the atmosphere and the ocean model is
controlled by the coupling software OASIS (Terray et al., 1998). Concerning the coupling
78
dynamics, at a distinct frequency the atmospheric component of the model passes heat, fresh
water and momentum to the ocean and gets information about surface conditions of the ocean.
This frequency is equal for all exchange fields and describes a 'coupled time step'. The fields that
are exchanged are averaged over the last coupled time step. Further aspects of the exchange
processes are flux corrections due to the interactive coupling between ocean and atmosphere in
order to prevent climate drift. These heat- and freshwater fluxes were diagnosed in a coupled
spin-up integration. Accordingly, the sea-surface-temperature and sea-surface salinity were
restored to their climatological observed values. This flux adjustment is constant in time and its
global average vanishes.
Quoted from (http://coast.gkss.de/staff/wagner/midhol/model/model_des.html)
79
Goddard Institute for Space Studies Atmospheric Ocean Model

The North American Space Association and the Goddard Institute for Space Studies
developed the GISS-AOM climate model, first in 1995 and then a revised version was created
with smaller grids in 2004 for the IPCC 4th assessment report. The model requires two kinds of
input, specified parameters and prognostic variables, and generates two kinds of output, climate
diagnostics and prognostic variables. The specified input parameters include physical constants,
the Earth's orbital parameters, the Earth's atmospheric constituents, the Earth's topography, the
Earth's surface distribution of ocean, glacial ice, or vegetation, and many others. The time
varying prognostic variables include fluid mass, horizontal velocity, heat, water vapor, salt, and
subsurface mass and energy fields. The resolution for the model is 4 longitude by 3 latitude
(PCMDI, 2005). The atmospheric grid has 12 vertical layers (PCMDI, 2005).
Model for Interdisciplinary Research on Climate version 3.2

The Japanese Model for Interdisciplinary Research on Climate version 3.2 (MIROC3.2) was
developed in two resolutions: the high resolution (MIROC3.2HIRES) in 1.125 1.125 grid
and the medium resolution (MIROC3.2MEDRES) in 2.8 2.8 grid. For present study, two
emissions scenarios from MIROC3.2HIRES (A1B and B1) and three scenarios (A1B, A2 and
B1) from MIROC3.2MEDRES were used.
80
APPENDIX D: Comparison of Future IDF Results in terms of

Intensities (mm/hr)
Historic Perturbed
Duration (min)
Return Periods, T yrs

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
25.87
35.16
41.32
49.09
54.85
60.57
120
17.39
23.73
27.92
33.22
37.16
41.06
360
7.26
9.54
11.05
12.96
14.38
15.78
720
4.31
5.49
6.28
7.28
8.01
8.75
1440
2.50
3.20
3.66
4.25
4.68
5.11
CGCM3T47_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
25.53
34.84
41.00
48.79
54.57
60.31
120
16.95
24.17
28.96
35.00
39.48
43.94
360
7.28
10.21
12.15
14.60
16.42
18.23
720
4.43
5.98
7.02
8.32
9.29
10.25
1440
2.54
3.42
4.00
4.73
5.28
5.82
CGCM3T47_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
26.38
36.86
43.80
52.56
59.06
65.52
120
16.08
21.28
24.71
29.05
32.28
35.47
360
7.06
9.20
10.62
12.41
13.73
15.05
720
4.34
5.61
6.46
7.52
8.31
9.10
1440
2.46
3.20
3.69
4.31
4.76
5.22
CGCM3T47_A2
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
25.77
36.90
44.26
53.57
60.47
67.32
120
16.25
22.00
25.80
30.61
34.17
37.71
360
7.21
9.98
11.81
14.12
15.84
17.54
720
4.51
6.22
7.35
8.79
9.85
10.91
1440
2.61
3.58
4.22
5.03
5.63
6.22
81
CGCM3T63_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
31.57
44.95
53.80
64.99
73.30
81.53
120
21.14
32.40
39.85
49.27
56.25
63.19
360
9.07
13.56
16.53
20.29
23.08
25.84
720
5.70
7.93
9.41
11.27
12.65
14.02
1440
3.35
4.57
5.38
6.41
7.16
7.92
CGCM3T63_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
28.35
38.69
45.53
54.18
60.60
66.96
120
18.81
27.89
33.90
41.49
47.12
52.71
360
7.96
11.56
13.94
16.95
19.19
21.40
720
4.95
6.75
7.94
9.44
10.56
11.67
1440
2.85
3.86
4.53
5.37
6.00
6.62
CGCM3T63_A2
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
29.92
39.66
46.11
54.26
60.31
66.31
120
19.95
27.44
32.40
38.66
43.31
47.92
360
8.69
11.68
13.66
16.16
18.02
19.86
720
5.27
6.82
7.86
9.16
10.13
11.09
1440
3.04
3.91
4.48
5.21
5.75
6.29
CSIROMK3.5_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
24.45
33.08
38.80
46.02
51.38
56.70
120
15.40
20.81
24.38
28.91
32.26
35.59
360
6.72
9.11
10.69
12.69
14.17
15.64
720
4.12
5.55
6.50
7.70
8.59
9.47
1440
2.42
3.29
3.86
4.59
5.13
5.67
CSIROMK3.5_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
27.45
39.40
47.30
57.29
64.71
72.06
120
16.86
23.46
27.82
33.34
37.44
41.50
360
7.30
9.91
11.64
13.82
15.44
17.05
720
4.43
5.97
6.99
8.28
9.24
10.19
1440
2.53
3.38
3.94
4.65
5.17
5.70
82
CSIROMK3.5_A2
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
28.73
41.01
49.13
59.40
67.02
74.59
120
18.13
24.83
29.28
34.89
39.05
43.18
360
7.76
10.42
12.19
14.42
16.07
17.72
720
4.70
6.29
7.34
8.66
9.64
10.62
1440
2.64
3.53
4.12
4.86
5.41
5.95
ECHAM5AOM_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
42.09
59.63
71.23
85.90
96.78
107.58
120
27.94
39.35
46.91
56.45
63.53
70.56
360
11.99
16.90
20.15
24.26
27.30
30.33
720
7.15
9.77
11.50
13.69
15.32
16.93
1440
4.09
5.54
6.50
7.71
8.61
9.51
ECHAM5AOM_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
39.63
52.45
60.94
71.66
79.62
87.52
120
25.10
33.75
39.48
46.72
52.09
57.42
360
10.91
14.13
16.26
18.95
20.95
22.93
720
6.58
8.26
9.37
10.78
11.82
12.85
1440
3.78
4.74
5.37
6.17
6.76
7.35
ECHAM5AOM_A2
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
39.65
52.29
60.65
71.22
79.06
86.85
120
26.54
36.07
42.38
50.36
56.27
62.15
360
11.59
15.46
18.03
21.27
23.67
26.05
720
6.79
8.77
10.08
11.73
12.96
14.17
1440
3.88
5.03
5.79
6.76
7.47
8.18
ECHO-G_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
34.85
48.38
57.33
68.65
77.05
85.39
120
22.18
30.94
36.74
44.06
49.49
54.89
360
9.48
13.10
15.49
18.52
20.77
23.00
720
5.53
7.56
8.90
10.60
11.86
13.11
1440
3.16
4.26
4.99
5.91
6.60
7.28
83
ECHO-G_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
34.35
49.95
60.28
73.33
83.02
92.63
120
21.54
30.52
36.47
43.98
49.55
55.08
360
9.29
12.90
15.29
18.30
20.54
22.76
720
5.36
7.48
8.89
10.67
11.99
13.30
1440
3.04
4.15
4.89
5.82
6.52
7.20
ECHO-G_A2
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
34.49
49.16
58.87
71.14
80.25
89.28
120
22.29
31.94
38.34
46.41
52.41
58.36
360
9.58
13.19
15.58
18.60
20.84
23.06
720
5.59
7.46
8.69
10.25
11.41
12.55
1440
3.24
4.21
4.86
5.68
6.28
6.89
GFDLCM2.1_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
32.50
44.43
52.33
62.31
69.72
77.07
120
20.75
27.68
32.27
38.07
42.38
46.65
360
9.14
11.98
13.85
16.23
17.99
19.74
720
5.60
7.23
8.32
9.69
10.70
11.71
1440
3.20
4.10
4.69
5.45
6.00
6.56
GFDLCM2.1_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
33.63
48.66
58.61
71.18
80.51
89.77
120
21.32
30.52
36.61
44.31
50.02
55.68
360
9.22
13.30
15.99
19.40
21.92
24.43
720
5.61
7.92
9.45
11.38
12.82
14.24
1440
3.20
4.45
5.28
6.33
7.11
7.88
GFDLCM2.1_A2
Duration

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
31.88
44.60
53.02
63.65
71.54
79.38
120
20.29
28.65
34.17
41.16
46.34
51.48
360
8.98
12.54
14.90
17.89
20.10
22.29
720
5.37
7.33
8.63
10.26
11.48
12.68
1440
3.08
4.20
4.94
5.88
6.57
7.26
84
GISSAOM_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
32.64
45.34
53.75
64.37
72.25
80.08
120
21.91
32.53
39.55
48.43
55.02
61.56
360
9.23
13.52
16.36
19.95
22.61
25.25
720
5.40
7.56
9.00
10.81
12.16
13.49
1440
3.08
4.24
5.01
5.98
6.70
7.41
GISSAOM_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
29.39
39.40
46.03
54.41
60.62
66.78
120
19.40
27.11
32.21
38.66
43.45
48.20
360
8.13
11.05
12.99
15.44
17.25
19.05
720
4.91
6.35
7.30
8.50
9.40
10.28
1440
2.80
3.61
4.15
4.83
5.34
5.84
MIROC3HIRES_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
28.29
39.68
47.22
56.74
63.81
70.82
120
18.04
24.66
29.05
34.59
38.70
42.78
360
7.86
10.31
11.92
13.97
15.48
16.99
720
4.78
6.20
7.13
8.32
9.19
10.06
1440
2.72
3.54
4.07
4.75
5.26
5.76
MIROC3HIRES_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
32.24
44.09
51.93
61.84
69.20
76.49
120
20.89
30.13
36.26
43.99
49.73
55.43
360
8.77
12.33
14.69
17.67
19.88
22.07
720
5.17
6.88
8.01
9.44
10.50
11.56
1440
2.98
3.91
4.53
5.31
5.89
6.47
MIROC3MEDRES_A1B
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
27.16
39.84
48.24
58.85
66.72
74.54
120
17.30
25.17
30.38
36.97
41.86
46.71
360
7.75
11.26
13.58
16.51
18.69
20.84
720
4.65
6.70
8.06
9.77
11.04
12.31
1440
2.70
3.75
4.45
5.34
5.99
6.64
85
MIROC3MEDRES_B1
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
31.13
43.16
51.13
61.20
68.67
76.09
120
19.90
27.66
32.79
39.28
44.09
48.87
360
8.58
11.64
13.66
16.21
18.11
19.99
720
5.15
6.77
7.83
9.18
10.18
11.17
1440
2.94
3.78
4.35
5.05
5.58
6.10
MIROC3MEDRES_A2
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
24.12
34.15
40.79
49.17
55.40
61.57
120
15.18
20.15
23.44
27.59
30.67
33.73
360
6.75
8.91
10.35
12.16
13.50
14.83
720
1440
4.28
2.49
5.59
3.20
6.45
3.66
7.55
4.25
8.36
4.69
9.17
5.13
CCSRNIES_B21
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
40.08
57.19
68.52
82.84
93.46
104.01
120
25.63
38.11
46.37
56.81
64.55
72.24
360
10.93
15.56
18.62
22.49
25.36
28.21
720
6.36
8.70
10.26
12.22
13.67
15.11
1440
3.62
4.91
5.75
6.83
7.62
8.41
CSIROMK2b_B11
Duration (min)

2 yr
5 yr
10 yr
25 yr
50 yr
100 yr
60
26.68
39.56
48.09
58.86
66.86
74.79
120
17.63
26.21
31.89
39.08
44.40
49.69
360
7.29
10.59
12.78
15.53
17.58
19.61
720
4.19
5.89
7.02
8.45
9.51
10.56
1440
2.35
3.25
3.85
4.60
5.16
5.71
86
APEPNDIX E: IDF Plots of Selected Scenarios

Intensity (mm/hr)
100
2 yr
5 yr
10 y r
25 y r
50 y r
100 y r
10
1
60
360
Duration (min)
720
ECHAM5AOM_A1B (Wet)
100
Intensity (mm/hr)
120
1440
2 yr
5 yr
10 y r
25 y r
50 y r
100 y r
10
1
60
120
360
Duration (min)
720
1440
87
MIROC3.2 MEDRES_A2 (Dry)
Intensity (mm/hr)
100
2 yr
5 yr
10 y r
25 y r
50 y r
100 y r
10
1
60
120
720
1440
Resultant
100
Intensity (mm/hr)
360
Duration (min)
2 yr
5 yr
10 y r
25 y r
50 y r
100 y r
10
1
60
120
360
Duration (min)
720
1440
88
APPENDIX F: Previous Reports in the Series

ISSN: (print) 1913-3200; (online) 1913-3219
(1) Slobodan P. Simonovic (2001). Assessment of the Impact of Climate Variability and Change
on the Reliability, Resiliency and Vulnerability of Complex Flood Protection Systems. Water
(2) Predrag Prodanovic (2001). Fuzzy Set Ranking Methods and Multiple Expert Decision
Making. Water Resources Research Report no. 039, Facility for Intelligent Decision Support,
Department of Civil and Environmental Engineering, London, Ontario, Canada, 68 pages. ISBN:
(print) 978-0-7714-2608-7; (online) 978-0-7714-2609-4.
(3) Nirupama and Slobodan P. Simonovic (2002). Role of Remote Sensing in Disaster
Management. Water Resources Research Report no. 040, Facility for Intelligent Decision
Support, Department of Civil and Environmental Engineering, London, Ontario, Canada, 107
pages. ISBN: (print) 978-0-7714-2610-0; (online) 978-0-7714- 2611-7.
(4) Taslima Akter and Slobodan P. Simonovic (2002). A General Overview of Multiobjective
Multiple-Participant Decision Making for Flood Management. Water Resources Research Report
978-0-7714-2613-1.
(5) Nirupama and Slobodan P. Simonovic (2002). A Spatial Fuzzy Compromise Approach for
Flood Disaster Management. Water Resources Research Report no. 042, Facility for Intelligent
Decision Support, Department of Civil and Environmental Engineering, London, Ontario,
Canada, 138 pages. ISBN: (print) 978-0-7714-2614-8; (online) 978-0-7714-2615-5.
(6) K. D. W. Nandalal and Slobodan P. Simonovic (2002). State-of-the-Art Report on Systems
Analysis Methods for Resolution of Conflicts in Water Resources Management. Water
(7) K. D. W. Nandalal and Slobodan P. Simonovic (2003). Conflict Resolution Support System
A Software for the Resolution of Conflicts in Water Resource Management. Water Resources
89
(8) Ibrahim El-Baroudy and Slobodan P. Simonovic (2003). New Fuzzy Performance Indices for
Reliability Analysis of Water Supply Systems. Water Resources Research Report no. 045,
Facility for Intelligent Decision Support, Department of Civil and Environmental Engineering,
London, Ontario, Canada, 90 pages. ISBN: (print) 978-0-7714-2620-9; (online) 978-0-77142621-6.
(9) Juraj Cunderlik (2003). Hydrologic Model Selection for the CFCAS Project: Assessment of
Water Resources Risk and Vulnerability to Changing Climatic Conditions. Water Resources
Environmental Engineering, London, Ontario, Canada, 40 pages. ISBN: (print) 978-0-77142622-3; (online) 978-0-7714- 2623-0.
(10) Juraj Cunderlik and Slobodan P. Simonovic (2004). Selection of Calibration and
Verification Data for the HEC-HMS Hydrologic Model. Water Resources Research Report no.
047, Facility for Intelligent Decision Support, Department of Civil and Environmental
978-0-7714-2625-4.
(11) Juraj Cunderlik and Slobodan P. Simonovic (2004). Calibration, Verification and Sensitivity
Analysis of the HEC-HMS Hydrologic Model. Water Resources Research Report no. 048,
London, Ontario, Canada, 113 pages. ISBN: (print) 978- 0-7714-2626-1; (online) 978-0-77142627-8.
(12) Predrag Prodanovic and Slobodan P. Simonovic (2004). Generation of Synthetic Design
Storms for the Upper Thames River basin. Water Resources Research Report no. 049, Facility
for Intelligent Decision Support, Department of Civil and Environmental Engineering, London,
Ontario, Canada, 20 pages. ISBN: (print) 978- 0-7714-2628-5; (online) 978-0-7714-2629-2.
(13) Ibrahim El-Baroudy and Slobodan P. Simonovic (2005). Application of the Fuzzy
Performance Indices to the City of London Water Supply System. Water Resources Research
Report no. 050, Facility for Intelligent Decision Support, Department of Civil and Environmental
978-0-7714-2631-5.
(14) Ibrahim El-Baroudy and Slobodan P. Simonovic (2006). A Decision Support System for
Integrated Risk Management. Water Resources Research Report no. 051, Facility for Intelligent
(15) Predrag Prodanovic and Slobodan P. Simonovic (2006). Inverse Flood Risk Modelling of
The Upper Thames River Basin. Water Resources Research Report no. 052, Facility for
Intelligent Decision Support, Department of Civil and Environmental Engineering, London,
Ontario, Canada, 163 pages. ISBN: (print) 978-0-7714-2634-6; (online) 978-0-7714-2635-3.
(16) Predrag Prodanovic and Slobodan P. Simonovic (2006). Inverse Drought Risk Modelling of
90
The Upper Thames River Basin. Water Resources Research Report no. 053, Facility for
(17) Predrag Prodanovic and Slobodan P. Simonovic (2007). Dynamic Feedback Coupling of
Continuous Hydrologic and Socio-Economic Model Components of the Upper Thames River
Basin. Water Resources Research Report no. 054, Facility for Intelligent Decision Support,
Department of Civil and Environmental Engineering, London, Ontario, Canada, 437 pages.
ISBN: (print) 978-0-7714-2638-4; (online) 978-0-7714-2639-1.
(18) Subhankar Karmakar and Slobodan P. Simonovic (2007). Flood Frequency Analysis Using
Copula with Mixed Marginal Distributions. Water Resources Research Report no. 055, Facility
for Intelligent Decision Support, Department of Civil and Environmental Engineering, London,
(19) Jordan Black, Subhankar Karmakar and Slobodan P. Simonovic (2007). A Web-Based
Flood Information System. Water Resources Research Report no. 056, Facility for Intelligent
(20) Angela Peck, Subhankar Karmakar and Slobodan P. Simonovic (2007). Physical,
Economical, Infrastructural and Social Flood Risk Vulnerability Analyses in GIS. Water
(21) Predrag Prodanovic and Slobodan P. Simonovic (2007). Development of Rainfall Intensity
Duration Frequency Curves for the City of London Under the Changing Climate. Water
(22) Evan G. R. Davies and Slobodan P. Simonovic (2008). An integrated system dynamics
model for analyzing behaviour of the social-economic-climatic system: Model description and
model use guide. Water Resources Research Report no. 059, Facility for Intelligent Decision
Support, Department of Civil and Environmental Engineering, London, Ontario, Canada, 233
pages. ISBN: (print) 978-0-7714-2679-7; (online) 978-0-7714-2680-3.
(23) Vasan Arunachalam (2008). Optimization Using Differential Evolution. Water Resources
(24) Rajesh Shrestha and Slobodan P. Simonovic (2009). A Fuzzy Set Theory Based
Methodology for Analysis of Uncertainties in Stage-Discharge Measurements and Rating Curve.
Water Resources Research Report no. 061, Facility for Intelligent Decision Support, Department
91
of Civil and Environmental Engineering, London, Ontario, Canada, 104 pages. ISBN: (print)
978-0-7714-2707-7; (online) 978-0-7714-2708-4.
(25) Hyung-Il Eum, Vasan Arunachalam and Slobodan P. Simonovic (2009). Integrated
Reservoir Management System for Adaptation to Climate Change Impacts in the Upper Thames
River Basin. Water Resources Research Report no. 062, Facility for Intelligent Decision Support,
Department of Civil and Environmental Engineering, London, Ontario, Canada, 81 pages. ISBN:
(print) 978-0-7714-2710-7; (online) 978-0-7714-2711-4.
(26) Evan G. R. Davies and Slobodan P. Simonovic (2009). Energy Sector for the Integrated
System Dynamics Model for Analyzing Behaviour of the Social- Economic-Climatic Model.
Water Resources Research Report no. 063. Facility for Intelligent Decision Support, Department
of Civil and Environmental Engineering, London, Ontario, Canada. 191 pages. ISBN: (print)
978-0-7714-2712-1; (online) 978-0-7714-2713-8.
(27) Leanna King, Tarana Solaiman, and Slobodan P. Simonovic (2009). Assessment of Climatic
Vulnerability in the Upper Thames River Basin. Water Resources Research Report no. 064,
London, Ontario, Canada, 61pages. ISBN: (print) 978-0-7714-2816-6; (online) 978-0-77142817-3.
(28) Slobodan P. Simonovic and Angela Peck (2009). Updated Rainfall Intensity Duration
Frequency Curves for the City of London under Changing Climate. Water Resources Research
Report no. 065, Facility for Intelligent Decision Support, Department of Civil and Environmental
Engineering, London, Ontario, Canada, 64pages. ISBN: (print) 978-0-7714-2819-7; (online)
987-0-7714-2820-3.
(29) Leanna King, Tarana Solaiman, and Slobodan P. Simonovic (2010). Assessment of Climatic
Vulnerability in the Upper Thames River Basin: Part 2. Water Resources Research Report no.
066, Facility for Intelligent Decision Support, Department of Civil and Environmental
Engineering, London, Ontario, Canada, 72pages. ISBN: (print) 978-0-7714-2834-0; (online)
978-0-7714-2835-7.
(30) Christopher J. Popovich, Slobodan P. Simonovic and Gordon A. McBean (2010).Use of an
Integrated System Dynamics Model for Analyzing Behaviour of the Social-Economic-Climatic
System in Policy Development. Water Resources Research Report no. 067, Facility for
(31) Hyung-Il Eum and Slobodan P. Simonovic (2009). City of London: Vulnerability of
Infrastructure to Climate Change; Background Report 1 Climate and Hydrologic Modeling.
of Civil and Environmental Engineering, London, Ontario, Canada, 102pages. ISBN: (print) 9780-7714-2844-9; (online) 978-0-7714-2845-6.
92
(32) Dragan Sredojevic and Slobodan P. Simonovic (2009). City of London: Vulnerability of
Infrastructure to Climate Change; Background Report 2 Hydraulic Modeling and Floodplain
Mapping. Water Resources Research Report no. 069, Facility for Intelligent Decision Support,
Department of Civil and Environmental Engineering, London, Ontario, Canada, 147 pages.
ISBN: (print) 978-0-7714-2846-3; (online) 978-0-7714-2847-0.
(33) Tarana A. Solaiman and Slobodan P. Simonovic (February 2011). Quantifying Uncertainties
in the Modelled Estimates of Extreme Precipitation Events at the Upper Thames River Basin.
of Civil and Environmental Engineering, London, Ontario, Canada, 167 pages. ISBN: (print)
978-0-7714-2878-4; (online) 978-0-7714-2880-7.
(34) Tarana A. Solaiman and Slobodan P. Simonovic (January 2011). Assessment of Global and
Regional Reanalyses Data for Hydro-Climatic Impact Studies in the Upper Thames River Basin.
of Civil and Environmental Engineering, London, Ontario, Canada, 74 pages. ISBN: (print) 9780-7714-2892-0; (online) 978-0-7714-2899-9.
93

Development of IDF Under Climate Change

Uploaded by

Copyright:

Available Formats

Development of IDF Under Climate Change

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Development of IDF Under Climate Change

Uploaded by

Copyright:

Available Formats

THE UNIVERSITY OF WESTERN ONTARIO

DEPARTMENT OF CIVIL AND

Water Resources Research Report

Development of Probability Based IntensityDuration-Frequency Curves under Climate

Report No: 072

Development of Probability Based Intensity-Duration-Frequency

Department of Civil and Environmental Engineering

Hydrologic design of storm sewers, culverts, retention/detention basins and other

is encouraged. Inherent uncertainties associated with different AOGCMs are quantified by a

Executive Summary ........................................................................................................................ 2

4.1 Selection of Appropriate Stations ....................................................................................... 37

Figure 1: Meteorological Stations Used in the Study ................................................................... 18

1.2 Uncertainties in Atmosphere-Ocean Global Climate Models

1.3 Intensity-Duration-Frequency Curves

can be expressed as:

1.4 Outline of the Report

Literature related to intensity-duration-frequency (IDF) curves concentrates on developing

Figure 1: Meteorological Stations Used in the Study

Climate ID Station Name

3.1.2 Climate Change Scenarios

Table 2: List of AOGCM Models and Emission Scenarios

Canadian Centre for Climate Modelling

Centre for Climate System Research

3.2 Development of Methodology

Modified Daily Output

Daily Outputs 2080s

Hourly Output 2080s

Max. Rainfall for Diff.

IDF Curve for

IDF Curves for 2080s

IDF Curve 2080s

Figure 2: Schematic Diagram of Developing IDF Curve

PCt PCk 2 / Var ( PC )

Cumulative probabilities, pj, are given by:

9) A random number u (0,1) is generated and compared to the cumulative probability

3.2.2 Bias Correction of Downscaled Outputs

the correction factor for the AOGCMs is then calculated using,

days are selected to account for the seasonally

is the daily rainfall output from weather generator,

is the historical observed

respectively. The weights

. The combination of the weights, which provide the lowest

represents the fragments calculated for hour ,

is the chosen hourly data from

with the daily data to produce data for each hour:

3.2.4 Generation of Rainfall for Different Durations

3.2.5 Intensity-Duration-Frequency Analysis

represents the magnitude of the

deviation of the annual maximum series, and

are the mean and standard

is a frequency factor depending on the return

is obtained using the following equation:

Ratio (n-min to 60-min)

presents the rainfall intensity in mm/hr,

is the duration of rainfall in minute,

is assumed and the values of

are estimated by the least square method. The

3.2.6 Uncertainty Quantification

( ) based on random sample

penalized-likelihood methods, k-nearest neighbor methods, Bayesian-spline methods, and