Spatial-Temporal Modelling of Extreme Rainfall

Slides:



Advertisements
Similar presentations
Hydrology Rainfall Analysis (1)
Advertisements

Introduction to modelling extremes
Introduction to modelling extremes Marian Scott (with thanks to Clive Anderson, Trevor Hoey) NERC August 2009.
Chapter 3 Properties of Random Variables
1 McGill University Department of Civil Engineering and Applied Mechanics Montreal, Quebec, Canada.
Mean, Proportion, CLT Bootstrap
Copyright © 2007 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Chapter 18 Sampling Distribution Models.
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Chapter 18 Sampling Distribution Models.
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Chapter 18 Sampling Distribution Models.
Copyright © 2007 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Chapter 18 Sampling Distribution Models.
Inference Sampling distributions Hypothesis testing.
Budapest May 27, 2008 Unifying mixed linear models and the MASH algorithm for breakpoint detection and correction Anders Grimvall, Sackmone Sirisack, Agne.
1 SSS II Lecture 1: Correlation and Regression Graduate School 2008/2009 Social Science Statistics II Gwilym Pryce
Prediction, Correlation, and Lack of Fit in Regression (§11. 4, 11
Sampling Distributions (§ )
Chapter 18 Sampling Distribution Models
Copyright © 2010 Pearson Education, Inc. Chapter 18 Sampling Distribution Models.
Estimation of Rainfall Areal Reduction Factors Using NEXRAD Data Francisco Olivera, Janghwoan Choi and Dongkyun Kim Texas A&M University – Department of.
Analysis of Extremes in Climate Science Francis Zwiers Climate Research Division, Environment Canada. Photo: F. Zwiers.
Multiple Linear Regression Model
Climate Change and Extreme Wave Heights in the North Atlantic Peter Challenor, Werenfrid Wimmer and Ian Ashton Southampton Oceanography Centre.
Evaluating Hypotheses
Extreme Value Analysis, August 15-19, Bayesian analysis of extremes in hydrology A powerful tool for knowledge integration and uncertainties assessment.
Statistics for Managers Using Microsoft Excel, 4e © 2004 Prentice-Hall, Inc. Chap 6-1 Chapter 6 The Normal Distribution and Other Continuous Distributions.
Lehrstuhl für Informatik 2 Gabriella Kókai: Maschine Learning 1 Evaluating Hypotheses.
Analysis of Individual Variables Descriptive – –Measures of Central Tendency Mean – Average score of distribution (1 st moment) Median – Middle score (50.
Role and Place of Statistical Data Analysis and very simple applications Simplified diagram of a scientific research When you know the system: Estimation.
1 BA 555 Practical Business Analysis Review of Statistics Confidence Interval Estimation Hypothesis Testing Linear Regression Analysis Introduction Case.
Sampling Distributions & Point Estimation. Questions What is a sampling distribution? What is the standard error? What is the principle of maximum likelihood?
Analyses of Rainfall Hydrology and Water Resources RG744
Hydrologic Statistics
Concepts and Notions for Econometrics Probability and Statistics.
© Copyright McGraw-Hill CHAPTER 6 The Normal Distribution.
1 DATA DESCRIPTION. 2 Units l Unit: entity we are studying, subject if human being l Each unit/subject has certain parameters, e.g., a student (subject)
Bayesian Spatial Modeling of Extreme Precipitation Return Levels Daniel COOLEY, Douglas NYCHKA, and Philippe NAVEAU (2007, JASA)
Extreme Value Analysis What is extreme value analysis?  Different statistical distributions that are used to more accurately describe the extremes of.
BPS - 3rd Ed. Chapter 211 Inference for Regression.
Chapter Twelve Census: Population canvass - not really a “sample” Asking the entire population Budget Available: A valid factor – how much can we.
University of Ottawa - Bio 4118 – Applied Biostatistics © Antoine Morin and Scott Findlay 08/10/ :23 PM 1 Some basic statistical concepts, statistics.
Sampling Distributions & Standard Error Lesson 7.
Copyright © 2009 Pearson Education, Inc. Chapter 18 Sampling Distribution Models.
Recent Advances in Climate Extremes Science AVOID 2 FCO-Roshydromet workshop, Moscow, 19 th March 2015 Simon Brown, Met Office Hadley Centre.
The Dirichlet Labeling Process for Functional Data Analysis XuanLong Nguyen & Alan E. Gelfand Duke University Machine Learning Group Presented by Lu Ren.
Copyright © 2008 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Chapter 18 Sampling Distribution Models.
Sampling Distribution Models Chapter 18. Toss a penny 20 times and record the number of heads. Calculate the proportion of heads & mark it on the dot.
Extreme values and risk Adam Butler Biomathematics & Statistics Scotland CCTC meeting, September 2007.
© Department of Statistics 2012 STATS 330 Lecture 20: Slide 1 Stats 330: Lecture 20.
Chapter 7 Point Estimation of Parameters. Learning Objectives Explain the general concepts of estimating Explain important properties of point estimators.
Inferential Statistics Introduction. If both variables are categorical, build tables... Convention: Each value of the independent (causal) variable has.
1 Sampling Distribution of Arithmetic Mean Dr. T. T. Kachwala.
New approaches in extreme-value modeling A.Zempléni, A. Beke, V. Csiszár (Eötvös Loránd University, Budapest) Flood Risk Workshop,
ES 07 These slides can be found at optimized for Windows)
Chris Ferro Climate Analysis Group Department of Meteorology University of Reading Extremes in a Varied Climate 1.Significance of distributional changes.
CHAPTER – 1 UNCERTAINTIES IN MEASUREMENTS. 1.3 PARENT AND SAMPLE DISTRIBUTIONS  If we make a measurement x i in of a quantity x, we expect our observation.
Hydrological Forecasting. Introduction: How to use knowledge to predict from existing data, what will happen in future?. This is a fundamental problem.
1 Statistical Analysis - Graphical Techniques Dr. Jerrell T. Stracener, SAE Fellow Leadership in Engineering EMIS 7370/5370 STAT 5340 : PROBABILITY AND.
Lecture 7: Bivariate Statistics. 2 Properties of Standard Deviation Variance is just the square of the S.D. If a constant is added to all scores, it has.
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Chapter 18 Sampling Distribution Models.
Statistical Inference for the Mean Objectives: (Chapter 8&9, DeCoursey) -To understand the terms variance and standard error of a sample mean, Null Hypothesis,
BPS - 5th Ed. Chapter 231 Inference for Regression.
Biostatistics Class 3 Probability Distributions 2/15/2000.
Sampling Distribution Models
Modeling and Simulation CS 313
Sampling Distribution Models
Simple Linear Regression - Introduction
Precipitation Analysis
Simple Linear Regression
Sampling Distributions (§ )
CHAPTER – 1.2 UNCERTAINTIES IN MEASUREMENTS.
CHAPTER – 1.2 UNCERTAINTIES IN MEASUREMENTS.
Presentation transcript:

Spatial-Temporal Modelling of Extreme Rainfall Climate Adaptation Flagship Mark Palmer and Carmen Chan 11th IMSC July 2010

Outline of talk What we wanted to do What data did we have What data did we use How we got the most out of it Borrowing strength BHM’s Combining data Squeezing data Colleagues: Aloke Phatak, Eddy Campbell, Bryson Bates, Santosh Aryal, Neil Viney, Carmen Chan, Yun Li 11th IMSC

What did we want to do Describe the characteristics of extreme rainfall over a relatively large area, which includes both gauged and ungauged sites Model the tails of the rainfall distribution Be able to estimate return levels for various periods Be able to calculate Intensity-Duration-Functions (IDF) curves Be able to calculate Depth-Area (DA) and Areal-Reduction-Factors (ARF) curves 11th IMSC

Return Levels, Return Periods The return period T, for a given duration(say 24 hours) and intensity i(d), is the average time interval between exceedance of the value i(d) ie the Return Period is the reciprocal of the probability of exceedance of that event. Corresponding to the return period T is the return level i(d) “there is no universally agreed definition of this quantity, but one definition is that it is the level exceeded in any one year with probability 1/N. The less precise definition is the level which is exceeded ‘once in N years’ is problematical if there is a trend in the process (for example)’” Smith (2001) 11th IMSC

Intensity Duration Frequency (IDF) Curves X-axis duration Y-axis intensity Each line corresponds to a fixed return period 11th IMSC

What data do we have Commonly Issues Daily data (measured from >9am to 9am the following day) Pluvio data essentially continuous measurements), though generally discretised to small (say 5 minute) intervals can then aggregate this data for any duration of interest, using a sliding window Issues Quality Spatial sparsity Temporal sparsity, gaps Untagged accumulations homogenization 11th IMSC

Apparent embarrassment of riches 11th IMSC

What data do we keep The BoM high quality data set is too small (particularly spatially) The minimal requirement is for seasonal data (eg summer winter) that we know the maximal value of! Can have short time series Can have temporal gaps If there are problems with summer data, might still be able to use winter data for that year (and vice versa) Any bit of ‘information’ in the data is potentially useful (better than completely omitting it) 11th IMSC

How to get the most out of it Develop a BHM with both spatial and temporal components Allows us to borrow strength (ie sites that are “close” together should have similar characteristics Anticipate that small proportions of ‘unusual’ sites or rainfall measurements will be ‘smoothed’ over, rather than having an undue influence on the analysis 11th IMSC

Based on a remarkable distributional result: Central to the statistical modelling of extreme values is the generalized extreme value (GEV) distribution. If represents the maximum of a sequence of independent random variables (say daily rainfall over a year), then the distribution of (the yearly maximum rainfall), is given by a very general result Its relationship to modelling extremes is analogous to the use of the Normal distribution for modelling means or sums of random variables, regardless of their parent distribution (CLT). the parameters the location, scale and shape parameters characterise the distribution of the extremes we describe changes in extremes by changes in these parameters 11th IMSC

At-site model Generally Doesnt this waste data? Use a Generalised Extreme Value (GEV) distribution to characterise extreme rainfall at a site Assume that changes in climate are reflected by changes in the GEV parameters at a site Assume these GEV’s vary smoothly spatially (Convolution kernel approach, Higdon, Sanso etc) Take block maxima approach (ie just use the largest value from each season, year by year) Doesnt this waste data? What about Generalised Pareto (GPD) approach (why not) So, what can we do to maximise information 11th IMSC

Getting most out of available data Build a BHM to borrow strength Use r-order statistics (smaller standard errors, better fit) Reparameterise location to dispersion coefficient (Koutsiyannis, Buishand) Combine over durations (Koutsiyannis et al., 1998) Constant shape parameter (Nadarajah, Anderson, and J.A. Tawn (1998) Constant dispersion coefficient (empirically observed) Model scale parameter over durations (ie now have 5 parameters per site) ‘correct’ daily GEV to 24 hour basis. The maxima from 24 hour aggregated data will be larger than the maxima of daily data [Robinson and Tawn, 2000], so estimate the extremal index, which is used to adjust the location and scale parameters of the GEV for the daily data to make it comparable with GEVs estimated from 24 hour data (the shape parameter remains unchanged 11th IMSC

Example of various durations:modelled scale parameter 11th IMSC

Example: BHM,fitted by MCMC Spatial model ‘between’ covariates eg elevation ‘within’ covariates eg ocean heat GEV parameters ‘within’ covariates coefficients Rainfall Di-graph representation of the Spatial-Temporal model for extreme rainfall 11th IMSC

Results: Spatial plots of parameters Model the scale parameter as a linear function of OHC Fitted surfaces of GEV parameters from MCMC sampling; (a) dispersion coefficient, (b) scale (intercept) parameter, (c) linear ocean heat anomaly coefficient for modelling scale parameter, (d) log(shape) parameter, (e) theta parameter, (f) eta parameter 11th IMSC

Results: Return Level plots Differenced return levels surfaces (2003 – 1953) for a fifty year return period (a), and associated standard error surface (b). 11th IMSC

Results: IDF Curves IDF curves Sample of IDF curves for the pluviograph site 566038, for a 50 year return period (Ocean heat anomaly = 0, i.e. effectively the historical long term average), drawn from the MCMC procedure, median value indicated by solid black line, broken lines indicate 0.025 and 0.975 quantiles. IDF curves 11th IMSC

Conclusions: Build a BHM to borrow strength, and combine many sources of data Use r-order statistics Combine over durations Model scale parameter over durations ‘correct’ daily GEV to 24 hour basis Reparameterise location to dispersion coefficient is useful Would like to say something about the extremes of areal rainfall using this approach, but: Warning: Assumption of conditional independence between sites after spatial modelling of GEV parameters is wrong, try sampling from it need to model this, eg Copulas, max-stable approach (composite likelihoods) 11th IMSC

Thank you Mark Palmer Phone: +61 8 9333 6293 Email: mark.palmer@csiro.au Web: www.csiro.au/cmis Thank you Contact Us Phone: 1300 363 400 or +61 3 9545 2176 Email: Enquiries@csiro.au Web: www.csiro.au

Koutsiyannis reparameterization Reparameterise location to dispersion coefficient properties of the GEV might give some help 11th IMSC