Phase and Amplitude Variation in Montreal Weather Jim Ramsay McGill University.

Slides:



Advertisements
Similar presentations
ADHD Reaction Times: Densities, Mixed Effects, and PCA.
Advertisements

A.S. 3.8 INTERNAL 4 CREDITS Time Series. Time Series Overview Investigate Time Series Data A.S. 3.8 AS91580 Achieve Students need to tell the story of.
Phase-Plane Plotting the Nondurable Goods Index. Nondurable goods last less than two years: Food, clothing, cigarettes, alcohol, but not personal computers!!
ECON 251 Research Methods 11. Time Series Analysis and Forecasting.
Seasonal Position Variations and Regional Reference Frame Realization Jeff Freymueller Geophysical Institute University of Alaska Fairbanks.
Chapter 10 Curve Fitting and Regression Analysis
Regression Analysis Using Excel. Econometrics Econometrics is simply the statistical analysis of economic phenomena Here, we just summarize some of the.
Data mining and statistical learning - lecture 6
Basis Expansion and Regularization Presenter: Hongliang Fei Brian Quanz Brian Quanz Date: July 03, 2008.
Handwriting: Registration and Differential Equations.
An Introduction to Functional Data Analysis Jim Ramsay McGill University.
An Introduction to Functional Data Analysis Jim Ramsay McGill University.
Jim Ramsay McGill University Basis Basics. Overview  What are basis functions?  What properties should they have?  How are they usually constructed?
Long Term Temperature Variability of Santa Barbara Coutny By Courtney Keeney and Leila M.V. Carvalho.
BA 555 Practical Business Analysis
Chapter 5 Time Series Analysis
Functional Data Analysis T Chapters 10,11,12 Markus Kuusisto.
Chapter 3 Forecasting McGraw-Hill/Irwin
1 Time Series Analysis Thanks to Kay Smith for making these slides available to me!
Human Growth: From data to functions. Challenges to measuring growth We need repeated and regular access to subjects for up to 20 years. We need repeated.
Macroeconomic Facts Chapter 3. 2 Introduction Two kinds of regularities in economic data: -Relationships between the growth components in different variables.
Part II – TIME SERIES ANALYSIS C2 Simple Time Series Methods & Moving Averages © Angel A. Juan & Carles Serrat - UPC 2007/2008.
Forecasting McGraw-Hill/Irwin Copyright © 2012 by The McGraw-Hill Companies, Inc. All rights reserved.
SMOOTHING TECHNIQUES TIME SERIES. COMPONENTS OF A TIME SERIES Components of a time series Seasonal effect Long term trend Cyclical effect Irregularity,
Investigating the long- and short-term variations in meteorology in Atlanta Lucas Henneman 25 April, 2013.
From Data to Differential Equations Jim Ramsay McGill University.
What are the dominant frequencies? Fourier transforms decompose a data sequence into a set of discrete spectral estimates – separate the variance of a.
Human Growth: From data to functions. Challenges to measuring growth We need repeated and regular access to subjects for up to 20 years. We need repeated.
Winter’s Exponential smoothing
The Forecast Process Dr. Mohammed Alahmed
Copyright © 2004 by The McGraw-Hill Companies, Inc. All rights reserved. Copyright © 2003 by The McGraw-Hill Companies, Inc. All rights reserved.
Empirical Modeling Dongsup Kim Department of Biosystems, KAIST Fall, 2004.
Temperature correction of energy consumption time series Sumit Rahman, Methodology Advisory Service, Office for National Statistics.
Time series Decomposition
Simple Linear Regression Models
Empirical Financial Economics Asset pricing and Mean Variance Efficiency.
© 1998, Geoff Kuenning Linear Regression Models What is a (good) model? Estimating model parameters Allocating variation Confidence intervals for regressions.
Holt’s exponential smoothing
Time Series Analysis and Forecasting
Chapter 6 Business and Economic Forecasting Root-mean-squared Forecast Error zUsed to determine how reliable a forecasting technique is. zE = (Y i -
Time series Decomposition Farideh Dehkordi-Vakil.
Time-Series Forecasting Overview Moving Averages Exponential Smoothing Seasonality.
BIOSYST-MeBioSwww.biw.kuleuven.be The potential of Functional Data Analysis for Chemometrics Dirk De Becker, Wouter Saeys, Bart De Ketelaere and Paul Darius.
1 1 Slide © 2011 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole.
Curve Registration The rigid metric of physical time may not be directly relevant to the internal dynamics of many real-life systems. Rather, there can.
Correlation of temperature with solar activity (SSN) Alexey Poyda and Mikhail Zhizhin Geophysical Center & Space Research Institute, Russian Academy of.
Image Registration with Hierarchical B-Splines Z. Xie and G. Farin.
Application of a North America reference frame to the Pacific Northwest Geodetic Array (PANGA) M M Miller, V M Santillan, Geodesy Laboratory, Central Washington.
Introduction Outdoor air pollution has a negative effect on health. On days of high air pollution, rates of cardiovascular and respiratory events increase.
Economics 173 Business Statistics Lecture 26 © Fall 2001, Professor J. Petry
BUAD306 Chapter 3 – Forecasting.
Copyright © 2013, 2010 and 2007 Pearson Education, Inc. Chapter Describing the Relation between Two Variables 4.
MGS3100_03.ppt/Feb 11, 2016/Page 1 Georgia State University - Confidential MGS 3100 Business Analysis Time Series Forecasting Feb 11, 2016.
Forecasting is the art and science of predicting future events.
Chapter 20 Time Series Analysis and Forecasting. Introduction Any variable that is measured over time in sequential order is called a time series. We.
Principal Components Analysis ( PCA)
Statistics for Business and Economics Module 2: Regression and time series analysis Spring 2010 Lecture 7: Time Series Analysis and Forecasting 1 Priyantha.
Assessing the Impact of Informality on Wages in Tanzania: Is There a Penalty for Women? Pablo Suárez Robles (University Paris-Est Créteil) 1.
Chapter 20 Time Series Analysis and Forecasting. Introduction Any variable that is measured over time in sequential order is called a time series. We.
Forecasting. Model with indicator variables The choice of a forecasting technique depends on the components identified in the time series. The techniques.
3. Data analysis SIS.
John Loucks St. Edward’s University . SLIDES . BY.
Physics 114: Exam 2 Review Material from Weeks 7-11
Phase-Plane Plotting the Nondurable Goods Index
Bradley W. Vines McGill University
Human Growth: From data to functions
Linear regression Fitting a straight line to observations.
3 4 Chapter Describing the Relation between Two Variables
What is the function of the graph? {applet}
Presentation transcript:

Phase and Amplitude Variation in Montreal Weather Jim Ramsay McGill University

The Data 34 years of daily temperatures, inclusive 34 years of daily temperatures, inclusive Values are averages of daily maximum and minimum Values are averages of daily maximum and minimum observations in tenths of a degree Celsius observations in tenths of a degree Celsius Available for Montreal and 34 other Canadian weather stations Available for Montreal and 34 other Canadian weather stations

We know that there are two kinds of variation in these data: 1. Amplitude variation: day-to-day and year-to-year variation in temperature at events such as the depth of winter. 2. Phase variation: the timing of these events -- the seasons arrive early in some years, and late in others.

Goals Separate phase variation from amplitude variation by registering the series to its strictly periodic image. Separate phase variation from amplitude variation by registering the series to its strictly periodic image. Estimate components of variation due to amplitude and phase variation. Estimate components of variation due to amplitude and phase variation.

Smoothing The registration process requires that we smooth the data two ways: 1. With an unconstrained smooth that removes the day-to-day variation, but leaves longer-term variation unchanged. 2. With a strictly periodic smooth that eliminates all but strictly periodic trend.

Unconstrained smooth Raw data are represented by a B-spline expansion using 500 basis functions of order 6. Raw data are represented by a B-spline expansion using 500 basis functions of order 6. Knot about every 25 days. Knot about every 25 days. The standard deviation of the raw data about this smooth, adjusted for degrees of freedom, is 4.30 degrees Celsius. The standard deviation of the raw data about this smooth, adjusted for degrees of freedom, is 4.30 degrees Celsius.

Periodic smooth The basis is Fourier, with 9 basis functions judged to be enough to capture most of the strictly periodic trend for a period of one year. The basis is Fourier, with 9 basis functions judged to be enough to capture most of the strictly periodic trend for a period of one year. The standard deviation of the raw about data about this smooth is 4.74 deg C. The standard deviation of the raw about data about this smooth is 4.74 deg C. Compare this to 4.30 deg C. for the unconstrained smooth. Compare this to 4.30 deg C. for the unconstrained smooth.

Plotting the unconstrained B-spline smooth minus the constrained Fourier smooth reveals some striking discrepancies. Plotting the unconstrained B-spline smooth minus the constrained Fourier smooth reveals some striking discrepancies. We focus on Christmas, The Ramsay’s spent the holidays in a chalet in the Townships, and awoke to –37 deg C. No skiing, car dead, marooned! We focus on Christmas, The Ramsay’s spent the holidays in a chalet in the Townships, and awoke to –37 deg C. No skiing, car dead, marooned! This temperature would still be cold in mid-January, but less unusual. This temperature would still be cold in mid-January, but less unusual.

Registration Let the unconstrained smooth be x(t) and the strictly periodic smooth be x 0 (t). Let the unconstrained smooth be x(t) and the strictly periodic smooth be x 0 (t). We need to estimate a nonlinear strictly increasing smooth transformation of time h(t), called a warping function, such that a fitting criterion is minimized. We need to estimate a nonlinear strictly increasing smooth transformation of time h(t), called a warping function, such that a fitting criterion is minimized.

Fitting criterion The fitting criterion was the smallest eigenvalue of the matrix This criterion measures the extent to which a plot of x[h(t)] against x 0 (t) is linear, and thus whether the two curves are in phase.

The warping function h(t) Every smooth strictly monotone function h(t) such that h(0) = 0 can be represented as We represent unconstrained function w(v) by a B-spline expansion. Constant C is determined by constraint h(T) = T.

The deformation d(t) = h(t) - t Plotting this allows us to see when the seasons come early (negative deformation) or late (positive deformation).

Mid-winter for arrived about 25 days early. Mid-winter for arrived about 25 days early. The next step is to register the temperature data by computing x*(t) = x[h(t)]. The registered curve x*(t) contains only amplitude variation. The next step is to register the temperature data by computing x*(t) = x[h(t)]. The registered curve x*(t) contains only amplitude variation. Registration was done by Matlab function registerfd, available by ftp from Registration was done by Matlab function registerfd, available by ftp fromego.psych.mcgill.ca/pub/ramsay/FDAfuns

Amplitude variation The standard deviation of the difference between the unconstrained smooth and the strictly periodic smooth is 2.15 C. The standard deviation of the difference between the unconstrained smooth and the strictly periodic smooth is 2.15 C. The standard deviation of the difference between the registered smooth and the periodic smooth is 1.73 C. The standard deviation of the difference between the registered smooth and the periodic smooth is 1.73 C. ( – )/ =.35, the proportion of the variation due to phase. ( – )/ =.35, the proportion of the variation due to phase.

The standard deviation of the raw data around the registered smooth is 2.13 C, compared with 2.07 C for the unregistered smooth. The standard deviation of the raw data around the registered smooth is 2.13 C, compared with 2.07 C for the unregistered smooth. About 10% of the total variation is due to phase. About 10% of the total variation is due to phase.

Conclusions Phase variation is an important part of weather behavior. Phase variation is an important part of weather behavior. Statisticians seldom think about phase variation, and classical time series methods ignore it completely. Statisticians seldom think about phase variation, and classical time series methods ignore it completely. Phase variation needs more attention, and registration is an essential tool. Phase variation needs more attention, and registration is an essential tool.