11 Day 11 (February 27)
11.1 Announcements
- I have to leave at 9:15 today!
- I will have office hours today in Dickens Hall room 109 from 2-3pm.
11.2 Extreme precipitation in Kansas
On September 3, 2018 there was an extreme precipitation event that resulted in flooding in Manhattan, KS and the surrounding areas. If you would like to know more about this, check out this link and this video.
What we may need to learn
- How to use R as a geographic information system
- New general tools from statistics
- Gaussian process
- How to use the hierarchical modeling framework to describe Kriging
- Hierarchical Bayesian model vs. “empirical” hierarchical model
- Specialized language used in spatial statistics (e.g., range, nugget, variogram)
11.3 Gaussian process
- See bottom of pg. 139 in Wikle et al. (2019)
- A Gaussian process is a probability distribution over functions
- If the function is observed at a finite number of points or “locations,” then the vector of values follows a multivariate normal distribution.
11.3.1 Multivariate normal distribution
Multivariate normal distribution
\(\boldsymbol{\eta}\sim\text{N}(\mathbf{0},\sigma^{2}\mathbf{R})\)
- Definitions
Correlation matrix – A positive semi-definite matrix whose elements are the correlation between observations
Correlation function – A function that describes the correlation between observations
Example correlation matrices
Compound symmetry \[\mathbf{R}=\left[\begin{array}{cccccc} 1 & 1 & 0 & 0 & 0 & 0\\ 1 & 1 & 0 & 0 & 0 & 0\\ 0 & 0 & 1 & 1 & 0 & 0\\ 0 & 0 & 1 & 1 & 0 & 0\\ 0 & 0 & 0 & 0 & 1 & 1\\ 0 & 0 & 0 & 0 & 1 & 1 \end{array}\right]\]
AR(1) \[\mathbf{R(\phi})=\left[\begin{array}{ccccc} 1 & \phi^{1} & \phi^{2} & \cdots & \phi^{n-1}\\ \phi^{1} & 1 & \phi^{1} & \cdots & \phi^{n-2}\\ \phi^{2} & \phi^{1} & 1 & \vdots & \phi^{n-3}\\ \vdots & \vdots & \vdots & \ddots & \ddots\\ \phi^{n-1} & \phi^{n-2} & \phi^{n-3} & \ddots & 1 \end{array}\right]\:\]
- Example simulating from \(\boldsymbol{\eta}\sim\text{N}(\mathbf{0},\sigma^{2}\mathbf{R})\) in R
n <- 200 x <- 1:n I <- diag(1, n) sigma2 <- 1 library(MASS) set.seed(2034) eta <- mvrnorm(1, rep(0, n), sigma2 * I) cor(eta[1:(n - 1)], eta[2:n])
## [1] -0.06408623
par(mfrow = c(2, 1)) par(mar = c(0, 2, 0, 0), oma = c(5.1, 3.1, 2.1, 2.1)) plot(x, eta, typ = "l", xlab = "", xaxt = "n") abline(a = 0, b = 0) n <- 200 x <- 1:n # Must be equally spaced phi <- 0.7 R <- phi^abs(outer(x, x, "-")) sigma2 <- 1 set.seed(1330) eta <- mvrnorm(1, rep(0, n), sigma2 * R) cor(eta[1:(n - 1)], eta[2:n])
## [1] 0.7016481
Example correlation functions
Gaussian correlation function \[r_{ij}(\phi)=e^{-\frac{d_{ij}^{2}}{\phi}}\] where \(d_{ij}\) is the “distance” between locations i and j (note that \(d_{ij}=0\) for \(i=j\)) and \(r_{ij}(\phi)\) is the element in the \(i^{\textrm{th}}\) row and \(j^{\textrm{th}}\) column of \(\mathbf{R}(\phi)\).
library(fields) n <- 200 x <- 1:n phi <- 40 d <- rdist(x) R <- exp(-d^2/phi) sigma2 <- 1 set.seed(4673) eta <- mvrnorm(1, rep(0, n), sigma2 * R) plot(x, eta, typ = "l") abline(a = 0, b = 0)
## [1] 0.9717508
Linear correlation function \[r_{ij}(\phi)=\begin{cases} 1-\frac{d_{ij}}{\phi} &\text{if}\ d_{ij}<0\\ 0 &\text{if}\ d_{ij}>0 \end{cases}\]
library(fields) n <- 200 x <- 1:n phi <- 40 d <- rdist(x) R <- ifelse(d<phi,1-d/phi,0) sigma2 <- 1 set.seed(4803) eta <- mvrnorm(1, rep(0, n), sigma2 * R) plot(x, eta, typ = "l") abline(a = 0, b = 0)
## [1] 0.9778878
11.5 Extreme precipitation in Kansas
- Live demonstration using what we have learned so far (Download R code here)