Introduction to kriging: The Best Linear Unbiased Estimator (BLUE) for space/time mapping.

Slides:



Advertisements
Similar presentations
Spatial point patterns and Geostatistics an introduction
Advertisements

Spatial point patterns and Geostatistics an introduction
Introduction to Data Assimilation NCEO Data-assimilation training days 5-7 July 2010 Peter Jan van Leeuwen Data Assimilation Research Center (DARC) University.
Introduction to Smoothing and Spatial Regression
Point Estimation Notes of STAT 6205 by Dr. Fan.
Kriging.
1 Regression Models & Loss Reserve Variability Prakash Narayan Ph.D., ACAS 2001 Casualty Loss Reserve Seminar.
Prediction, Correlation, and Lack of Fit in Regression (§11. 4, 11
STAT 497 APPLIED TIME SERIES ANALYSIS
Sampling Distributions
Statistical Inference Chapter 12/13. COMP 5340/6340 Statistical Inference2 Statistical Inference Given a sample of observations from a population, the.
FOUR METHODS OF ESTIMATING PM 2.5 ANNUAL AVERAGES Yan Liu and Amy Nail Department of Statistics North Carolina State University EPA Office of Air Quality,
Applied Geostatistics
G. Cowan Lectures on Statistical Data Analysis 1 Statistical Data Analysis: Lecture 8 1Probability, Bayes’ theorem, random variables, pdfs 2Functions of.
Statistical Background
Deterministic Solutions Geostatistical Solutions
Economics 20 - Prof. Anderson
1 Introduction to Biostatistics (PUBHLTH 540) Sampling.
A) Transformation method (for continuous distributions) U(0,1) : uniform distribution f(x) : arbitrary distribution f(x) dx = U(0,1)(u) du When inverse.
Geostatistics Mike Goodchild. Spatial interpolation n A field –variable is interval/ratio –z = f(x,y) –sampled at a set of points n How to estimate/guess.
Applications in GIS (Kriging Interpolation)
Method of Soil Analysis 1. 5 Geostatistics Introduction 1. 5
Lecture II-2: Probability Review
Prof. SankarReview of Random Process1 Probability Sample Space (S) –Collection of all possible outcomes of a random experiment Sample Point –Each outcome.
Gaussian process modelling
EXPLORING SPATIAL CORRELATION IN RIVERS by Joshua French.
1 LES of Turbulent Flows: Lecture 1 Supplement (ME EN ) Prof. Rob Stoll Department of Mechanical Engineering University of Utah Fall 2014.
Geo479/579: Geostatistics Ch12. Ordinary Kriging (1)
Gridding Daily Climate Variables for use in ENSEMBLES Malcolm Haylock, Climatic Research Unit Nynke Hofstra, Mark New, Phil Jones.
1 G Lect 8b G Lecture 8b Correlation: quantifying linear association between random variables Example: Okazaki’s inferences from a survey.
Geog. 579: GIS and Spatial Analysis - Lecture 21 Overheads 1 Point Estimation: 3. Methods: 3.6 Ordinary Kriging Topics: Lecture 23: Spatial Interpolation.
Geographic Information Science
Geo479/579: Geostatistics Ch16. Modeling the Sample Variogram.
9.3 and 9.4 The Spatial Model And Spatial Prediction and the Kriging Paradigm.
Week 21 Stochastic Process - Introduction Stochastic processes are processes that proceed randomly in time. Rather than consider fixed random variables.
Review of Random Process Theory CWR 6536 Stochastic Subsurface Hydrology.
Chapter 2 Statistical Background. 2.3 Random Variables and Probability Distributions A variable X is said to be a random variable (rv) if for every real.
PROBABILITY AND STATISTICS FOR ENGINEERING Hossein Sameti Department of Computer Engineering Sharif University of Technology Two Random Variables.
Random processes. Matlab What is a random process?
Geology 5670/6670 Inverse Theory 21 Jan 2015 © A.R. Lowry 2015 Read for Fri 23 Jan: Menke Ch 3 (39-68) Last time: Ordinary Least Squares Inversion Ordinary.
Review of Probability. Important Topics 1 Random Variables and Probability Distributions 2 Expected Values, Mean, and Variance 3 Two Random Variables.
Advanced Tutorial on : Global offset and residual covariance ENVR 468 Prahlad Jat and Marc Serre.
ECE-7000: Nonlinear Dynamical Systems 2. Linear tools and general considerations 2.1 Stationarity and sampling - In principle, the more a scientific measurement.
Stochastic Hydrology Random Field Simulation Professor Ke-Sheng Cheng Department of Bioenvironmental Systems Engineering National Taiwan University.
Geostatistics GLY 560: GIS for Earth Scientists. 2/22/2016UB Geology GLY560: GIS Introduction Premise: One cannot obtain error-free estimates of unknowns.
Geo479/579: Geostatistics Ch12. Ordinary Kriging (2)
Spatial Point Processes Eric Feigelson Institut d’Astrophysique April 2014.
CWR 6536 Stochastic Subsurface Hydrology
G. Cowan Lectures on Statistical Data Analysis Lecture 9 page 1 Statistical Data Analysis: Lecture 9 1Probability, Bayes’ theorem 2Random variables and.
Space-time processes NRCSE. Separability Separable covariance structure: Cov(Z(x,t),Z(y,s))=C S (x,y)C T (s,t) Nonseparable alternatives Temporally varying.
Modeling Space/Time Variability with BMEGUI Prahlad Jat (1) and Marc Serre (1) (1) University of North Carolina at Chapel Hill.
Exposure Prediction and Measurement Error in Air Pollution and Health Studies Lianne Sheppard Adam A. Szpiro, Sun-Young Kim University of Washington CMAS.
CWR 6536 Stochastic Subsurface Hydrology Optimal Estimation of Hydrologic Parameters.
STA302/1001 week 11 Regression Models - Introduction In regression models, two types of variables that are studied:  A dependent variable, Y, also called.
Study of PM10 Annual Arithmetic Mean in USA  Particulate matter is the term for solid or liquid particles found in the air  The smaller particles penetrate.
Stochastic Process - Introduction
Multiple Random Variables and Joint Distributions
Introduction to kriging: The Best Linear Unbiased Estimator (BLUE)
Probability Theory and Parameter Estimation I
Ch3: Model Building through Regression
Modeling Space/Time Variability with BMEGUI
STATISTICS Random Variables and Distribution Functions
Presentation of the project of Yasuyuki Akita Temporal GIS Fall 2004
Inference for Geostatistical Data: Kriging for Spatial Interpolation
Statistical Methods For Engineers
Stochastic Hydrology Random Field Simulation
Problem statement Given: a set of unknown parameters
Generalized Spatial Dirichlet Process Models
Interpolation & Contour Maps
STOCHASTIC HYDROLOGY Random Processes
Presentation transcript:

Introduction to kriging: The Best Linear Unbiased Estimator (BLUE) for space/time mapping

Spatiotemporal Continuum  p=(s,t) denotes a location in the space/time domain E=SxT Spatiotemporal Field  A field is the distribution  across space/time of some parameter X Space/Time Random Field (S/TRF)  A S/TRF is a collection of possible realizations  of the field, X(p)={p,  }  The collection of realizations represents the randomness (uncertainty and variability) in X(p) Definition of Space Time Random Fields X(p)X(p) Space s Time t Realization   X(p)X(p) Space s Time t Realization  (2)

Defining a S/TRF at a set of mapping points  We restrict Space/Time to a set of n  mapping points, p map =(p 1,…, p n )  Each field realization reduces to a set of n values,  map =(  1,…,  n )  The S/TRF reduces to set of n random variables, x map = (x 1,…, x n ) The multivariate PDF  The multivariate PDF f X characterizes the joint event x map ≈  map as Prob.[  map < x map <  map + d  map ] = f X (  map ) d  map hence the multivariate PDF provides a complete stochastic description of trends and dependencies of the S/TRF X(p) at its mapping points Marginal PDFs  The marginal PDF for a subset x a of x map = (x a, x b ) is f X (  a ) = ∫ d  b f X (  a,  b ) hence we can define any marginal PDF from f X (  map ) Multivariate PDF for the mapping points

Stochastic Expectation The stochastic expectation of some function g(X(p), X(p’), …) of the S/TRF is E [g(X(p), X(p’), …)] = ∫ d  1 d  2... g(  1,  2,...) f X (  1,  2,...; p ; p’,...) Mean trend and covariance The mean trend m X (p) =E [X(p)] and covariance c X (p, p’) =E [ (X(p)-m(p)) (X(p’)-m(p’)) ] are statistical moments of order 1 and 2, respectively, that characterizes the consistent tendencies and dependencies, respectively, of X(p) Statistical moments

A homogeneous/stationary S/TRF is defined by  A mean trend that is constant over space (homogeneity) and time (stationarity) m X (p) = m X  A covariance between point p =(s,t) and p’ =(s’,t’) that is only a function of spatial lag r=||s-s’|| and the temporal lag  = |t-t’| c X (p, p’) = c X ( (s,t), (s’,t’) ) = c X ( r=||s-s’||,  =|t-t’|  ) A homogeneous/stationary S/TRFs has the following properties  It’s variance is constant, i.e.  X 2 (p)=  X 2 Proof:  X 2 (p)= E[(X(p)- m X (p)) 2 ] = c X (p, p) = c X ( r=0,  =0 ) is not a function of p  It’s covariance can be written as c X (r,  )= E[X(s,t)X(s’,t’)] ||s-s’|| =r, |t-t’| =  - m X 2,  This is a useful equation to estimate the covariance Homogeneous/Stationary S/TRF

When having site-specific data, and assuming that the S/TRF is homogeneous/stationary, then we obtain experimental values for it’s covariance using the following estimator where N(r,  ) is the number of pairs of points with values (X head, X tail ) separated by a distance of r and a time of . In practice we use a tolerance dr and d , i.e. such that r-dr ≤ ||s head -s tail || ≤ r+dr and  -d  ≤ ||t head -t tail || ≤  +d  Experimental estimation of covariance

Gaussian model: c X (r) = c o exp-(3r 2 /a r 2 )  c o = sill = variance  a r = spatial range  Very smooth processes Exponential model: c X (r) = c o exp-(3r/a r )  more variability Nugget effect model c X (r) = c o  (r)  purely random Nested models c X (r) = c 1 (r) + c 2 (r) + …  where c 1 (r), c 2 (r), etc. are permissible covariance models Example: Arsenic c X (r) = 0.7  X 2 exp-(3r/7Km)  X 2 exp-(3r/40Km)  where the first structure represents variability over short distances (7Km), e.g. geology, the second structure represents variability over longer distances (40Km) e.g. aquifers. Spatial covariance models

c X (r,  ) is a 2D function with spatial component c X (r,  ) and temporal component c X (r=0,  ) Space/time separable covariance model  c X (r,  ) = c Xr (r) c Xt (  ), where c Xr (r) and c Xt (  ) are permissible models Nested space/time separable models  c X (r,  ) = c r1 (r) c t1 (  ) + c r 2 (r)c t 2 (  ) + … Example: Yearly Particulate Matter concentration (ppm) across the US  c X (r,  ) = c 1 exp(-3r/a r1 -3  /a t1 ) + c 2 exp(-3r/a r2 -3  /a t2 )  1 st structure c 1 = (log mg/m 3 ) 2, a r1 =448 Km, a t1 =1years is weather driven  2 nd structure c 1 = (log mg/m 3 ) 2, a r1 =17 Km, a t1 =45years due to human activities Space/time covariance models

Gather the data  hard =[  1,  2,  3, …] T and obtain the experimental covariance Fit a covariance model c X (r) to the experimental covariance Simple kriging (SK) is a linear estimator  x k (SK)  0  T x hard SK is unbiased  E[x k (SK) ] = E[x k ] ═► x k (SK)  m k + T (x hard  m hard ) SK minimizes the estimation variance  SK 2 = E [( x k  x k (SK)  ) 2 ]  ∂  SK 2 / ∂ T  ═► T = C k,hard C hard,hard -1 Hence the SK estimator is given by  x k (SK)  m k + C k,hard C hard,hard -1 (x hard.  m hard ) T And its variance is   SK 2   k 2 - C k,hard C hard,hard -1 C hard,k The simple kriging (SK) estimator

Run Kriging Example introToKrigingExample.m Example of kriging maps

Observations  Only hard data are considered  Exactitude property at the data points  Kriging estimates tend to the (prior) expected value away from the data points  Hence, kriging maps are characterized by “islands” around data points  Kriging variance is only a function to the distance from the data points Limitations of kriging  Kriging does not provide a rigorous framework to integrate hard and soft data  Kriging is a linear combination of data (i.e. it is the “best” only among linear estimators, but it might be a poor estimator compared to non-linear estimators)  The estimation variance does not account for the uncertainty in the data itself  Kriging assumes that the data is Gaussian, whereas in reality uncertainty may be non-Gaussian  Traditionally kriging has been implemented for spatial estimation, and space/time is merely viewed as adding another spatial dimension (this is wrong because it is lacking any explicit space/time metric)