Statistics, Numerical Models and Ensembles

Similar documents
Distributions, spatial statistics and a Bayesian perspective

Resampling Methods. Chapter 5. Chapter 5 1 / 52

Internal vs. external validity. External validity. This section is based on Stock and Watson s Chapter 9.

Smoothing, penalized least squares and splines

x 1 Outline IAML: Logistic Regression Decision Boundaries Example Data

CAUSAL INFERENCE. Technical Track Session I. Phillippe Leite. The World Bank

What is Statistical Learning?

A New Evaluation Measure. J. Joiner and L. Werner. The problems of evaluation and the needed criteria of evaluation

Probabilistic assessment of regional climate change: a Bayesian approach to combining. predictions from multi-model ensembles

Climate Change: the Uncertainty of Certainty

NUMBERS, MATHEMATICS AND EQUATIONS

How do we solve it and what does the solution look like?

Comparison of hybrid ensemble-4dvar with EnKF and 4DVar for regional-scale data assimilation

Bootstrap Method > # Purpose: understand how bootstrap method works > obs=c(11.96, 5.03, 67.40, 16.07, 31.50, 7.73, 11.10, 22.38) > n=length(obs) >

Resampling Methods. Cross-validation, Bootstrapping. Marek Petrik 2/21/2017

5.4 Measurement Sampling Rates for Daily Maximum and Minimum Temperatures

COMP 551 Applied Machine Learning Lecture 4: Linear classification

NAME TEMPERATURE AND HUMIDITY. I. Introduction

SAMPLING DYNAMICAL SYSTEMS

The general linear model and Statistical Parametric Mapping I: Introduction to the GLM

Hypothesis Tests for One Population Mean

ENSC Discrete Time Systems. Project Outline. Semester

Misc. ArcMap Stuff Andrew Phay

Concept Category 2. Trigonometry & The Unit Circle

Exam #1. A. Answer any 1 of the following 2 questions. CEE 371 October 8, Please grade the following questions: 1 or 2

Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeoff

Modelling of Clock Behaviour. Don Percival. Applied Physics Laboratory University of Washington Seattle, Washington, USA

Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeoff

ABSORPTION OF GAMMA RAYS

AP Statistics Notes Unit Two: The Normal Distributions

Checking the resolved resonance region in EXFOR database

We say that y is a linear function of x if. Chapter 13: The Correlation Coefficient and the Regression Line

Exam #1. A. Answer any 1 of the following 2 questions. CEE 371 March 10, Please grade the following questions: 1 or 2

MATCHING TECHNIQUES. Technical Track Session VI. Emanuela Galasso. The World Bank

Physics 2010 Motion with Constant Acceleration Experiment 1

Five Whys How To Do It Better

Differentiation Applications 1: Related Rates

APPLICATION OF THE BRATSETH SCHEME FOR HIGH LATITUDE INTERMITTENT DATA ASSIMILATION USING THE PSU/NCAR MM5 MESOSCALE MODEL

Simple Linear Regression (single variable)

INTERNATIONAL CENTRE FOR THEORETICAL PHYSICS

Lab #3: Pendulum Period and Proportionalities

4th Indian Institute of Astrophysics - PennState Astrostatistics School July, 2013 Vainu Bappu Observatory, Kavalur. Correlation and Regression

Relationships Between Frequency, Capacitance, Inductance and Reactance.

Methods for Determination of Mean Speckle Size in Simulated Speckle Pattern

Eric Klein and Ning Sa

AP Statistics Practice Test Unit Three Exploring Relationships Between Variables. Name Period Date

Part 3 Introduction to statistical classification techniques

CHAPTER 4 DIAGNOSTICS FOR INFLUENTIAL OBSERVATIONS

SUPPLEMENTARY MATERIAL GaGa: a simple and flexible hierarchical model for microarray data analysis

Web-based GIS Systems for Radionuclides Monitoring. Dr. Todd Pierce Locus Technologies

k-nearest Neighbor How to choose k Average of k points more reliable when: Large k: noise in attributes +o o noise in class labels

Lab 1 The Scientific Method

Phys. 344 Ch 7 Lecture 8 Fri., April. 10 th,

Comparing Several Means: ANOVA. Group Means and Grand Mean

Computational modeling techniques

A Matrix Representation of Panel Data

SNOW AND AVALANCHES FORECASTING OVER THE ANDES MOUNTAlNS. Jose A. Vergara* Departamento de Geoflsica, Universidad de Chile

Questions? Contact the guys below for help. Social Media Information

Probabilities for climate projections

Trigonometric Ratios Unit 5 Tentative TEST date

Regents Chemistry Period Unit 3: Atomic Structure. Unit 3 Vocabulary..Due: Test Day

CS 109 Lecture 23 May 18th, 2016

The Law of Total Probability, Bayes Rule, and Random Variables (Oh My!)

Fall 2013 Physics 172 Recitation 3 Momentum and Springs

NAME: Prof. Ruiz. 1. [5 points] What is the difference between simple random sampling and stratified random sampling?

BOUNDED UNCERTAINTY AND CLIMATE CHANGE ECONOMICS. Christopher Costello, Andrew Solow, Michael Neubert, and Stephen Polasky

Elements of Machine Intelligence - I

BASD HIGH SCHOOL FORMAL LAB REPORT

Study Group Report: Plate-fin Heat Exchangers: AEA Technology

Chapter 1 Notes Using Geography Skills

Least Squares Optimal Filtering with Multirate Observations

Data mining/machine learning large data sets. STA 302 or 442 (Applied Statistics) :, 1

AP Physics Laboratory #4.1: Projectile Launcher

This section is primarily focused on tools to aid us in finding roots/zeros/ -intercepts of polynomials. Essentially, our focus turns to solving.

2004 AP CHEMISTRY FREE-RESPONSE QUESTIONS

1 Introduction. Jean-Philippe Boulanger Æ Fernando Martinez Enrique C. Segura

Verification of Quality Parameters of a Solar Panel and Modification in Formulae of its Series Resistance

Tree Structured Classifier

1b) =.215 1c).080/.215 =.372

B. Definition of an exponential

Empiricism, objectivity and falsifiability are important scientific tenets. Together they tell us that

CS 477/677 Analysis of Algorithms Fall 2007 Dr. George Bebis Course Project Due Date: 11/29/2007

SticiGui Chapter 4: Measures of Location and Spread Philip Stark (2013)

Module 3: Gaussian Process Parameter Estimation, Prediction Uncertainty, and Diagnostics

Lesson Plan. Recode: They will do a graphic organizer to sequence the steps of scientific method.

, which yields. where z1. and z2

Formal Uncertainty Assessment in Aquarius Salinity Retrieval Algorithm

Math Foundations 20 Work Plan

**DO NOT ONLY RELY ON THIS STUDY GUIDE!!!**

Unit 14 Thermochemistry Notes

Multiple Source Multiple. using Network Coding

COMP 551 Applied Machine Learning Lecture 5: Generative models for linear classification

Pattern Recognition 2014 Support Vector Machines

Analysis on the Stability of Reservoir Soil Slope Based on Fuzzy Artificial Neural Network

IN a recent article, Geary [1972] discussed the merit of taking first differences

MATCHING TECHNIQUES Technical Track Session VI Céline Ferré The World Bank

Lecture 23: Lattice Models of Materials; Modeling Polymer Solutions

Time, Synchronization, and Wireless Sensor Networks

Multiband retardation control using multi-twist retarders

INTERNATIONAL BIRD STRIKE COMMITTEE IBSC27/WP X-3 Athens, May 2005 BIRD AVOIDANCE MODELS VS. REALTIME BIRDSTRIKE WARNING SYSTEMS A COMPARISON

Transcription:

Statistics, Numerical Mdels and Ensembles Duglas Nychka, Reinhard Furrer,, Dan Cley Claudia Tebaldi, Linda Mearns, Jerry Meehl and Richard Smith (UNC). Spatial predictin and data assimilatin Precipitatin extremes Cmbining IPCC climate mdel exp. Supprted by the Natinal Science Fundatin CAS2K5, Annecy, FR, Sep 2005

Anther way f summarizing the talk Part 1: Observatins are in the wrng place! Part 2 Observatins d nt measure what we want! Part 3 Nt sure what we have bserved!

Statistical Science What d yu want t knw? e.g. θ What have yu measured? e.g Y Relate them using a prbability distributin. e.g. Data = parameter errr = θ θ Characterize reasnable values fr θ given the data 0

The statistical methd Fr cmplicated prblems Use Bayesian mdels and Mnte Carl methds t generate a statistical ensemble fr θ. The ensemble mean is a gd estimate fr θ. The spread is a gd measure f uncertainty fr θ.

Part 1: Observatins are in the wrng place! Air quality

Spatial Predictin Predict surface zne where it is nt mnitred. 200 150 Ambient daily zne in PPB June 16, 1987, US Midwestern Regin. 100 50 0

A mdel fr the spatial field The zne surface has a mean and variance that can vary ver space. The crrelatin f zne at tw different lcatins has a knwn frm. Ozne fllws a Gaussian distributin

An ensemble apprach Start with a ensemble f fields that are distributed accrding t nes best guess r frecast withut cnsulting the data. Update each member f the ensemble using the bserved data. The sample mean and cvariances amng the ensemble members culd be used fr the update calculatins. This is the same algrithm used in the ensemble Kalman filter fr numerical weather predictin

Sme ensemble members fr zne

Uncertainty f zne at center f regin Predictins acrss 100 members. Frequency 0 5 10 15 20 25 60 80 100 120 140 160 PPB

Spatial Predictin The ensemble mean A Kriging, Bayes, OI,, BLUE slutin 200 150 100 50 0

A real ensemble frecast. Updates dne as in zne example

Part 2 Observatins d nt measure what we want!

Precipitatin extremes Hw will climate change effect extreme precipitatin? Extremes in precipitatin are used t determine fld ptential fr urban areas, fr dam and radway specificatins and als have extensive eclgical imprtance. Hw des ne estimate extremes where n bservatins are made? Hw des ne determine a pssible 25 year event frm 20 years wrth f data? Typically extremes are described by the return perid: A 25 year event = prbability f seeing this value (r higher) in a given year is 1/25 r 4%

The Western US

Clrad Frnt Range

Observed precipitatin fr Bulder, CO Daily precipitatin amunts mm 0 40 100 1950 1960 1970 1980 1990 2000 years

Observed precipitatin fr Bulder, CO Daily precipitatin amunts threshlded at 2.5 cm mm 0 40 100 1950 1960 1970 1980 1990 2000 years Distributin abve threshld: Density 0.0 0.5 1.0 1.5 2.0 0 1 2 3 4 mm

A spatial mdel fr precipitatin extremes Use extreme value statistical thery t apprximate the distributin f large values three parameters. Assume that the parameters f the distributin vary ver space. (see Part 1.) If yu knw the parameters f the distributin this can easily be cnverted t a 25 year return level.

Distributin fit t the Bulder exceedances... and the estimated 25 year event ( 9cm) Density 0.0 0.2 0.4 0.6 0.8 2 4 6 8 10 12 CM

Six ensemble members fr 25 year event 25 year return level based n all daily met statins in the Frnt Range

Ensemble mean f the 25 year return level

Ensemble mean f the 25 year return level Elevatin and return level (cm)

Part 3 Nt sure what yu have bserved!...

Data and the IPCC What will the climate be like in 2100? Hw much data d we have t answer this questin. The mst recent experiments t supprt the furth reprt f the Internatinal Panel n Climate Change amunt t an archive f apprximately 100 Tb. Mre than 20 different climate mdels/mdeling centers represented. Several different future scenaris. Multiply respnses e.g. temperature, precipitatin,

Sme Data Present winter temperatures, 9 AOGCMs Future - Present really a massive data set? Is this

NCEP Nrthern Hemisphere Winter

Standard IPCC regins

A Statistical Mdel Observatins = truth P errr Mdel Present = truth P mdel/regin bias 1 errr. Mdel Future = truth F mdel/regin bias 2 errr. Climate change = truth F - truth P

Sme ensemble members fr reginal change Future Present DJF temperature (A2)

Western Nrth America temp. change Frequency 0 50 150 250 3.0 3.5 4.0 4.5 5.0 C

Ensemble distributins: AR4 A summary fr the 2 3 4 5 6 7

Summary Statistical ensembles are a useful way t estimate spatial fields and characterize uncertainty. Statistical methds can be used t estimate cmplex indirect features. The size f massive data sets may nt be massive. Statistics can be used t gauge the representativeness f a sample.

Thank yu!