generating gridded fields of extreme precipitation for ... · given daily data, if we select the...
TRANSCRIPT
Introduction Background BHM Application Simulation References
Generating gridded fields of extreme precipitationfor large domains with a Bayesian hierarchical
model
Cameron BrackenDepartment of Civil, Environmental and Architectural Engineering
University of Colorado at Boulder
Vannes, FranceMay 19, 2016
1 / 24
Introduction Background BHM Application Simulation References
Coauthors
I Balaji Rajagopalan - University of Colorado
I Will Kleiber - University of Colorado
I Linyin Cheng - NOAA
I Subhrendu Gangopadhyay - Bureau of Reclamation
2 / 24
Introduction Background BHM Application Simulation References
This talk
I Motivation - Link to weather generation
I Bayesian hierarchical model (BHM) for spatial extremes
I Application to the Western US
I Simulation procedure using the BHM
3 / 24
Introduction Background BHM Application Simulation References
Motivation - Extremes in the western US?
I From 1983-2000 the WesternStates (WA, OR, CA, ID, NV,UT, AZ, MT, WY, CO, NM,ND, SD, NE, KS, OK, and TX)experienced $24.7 billion inflood damages, $1.5 billionannually.
I California, Washington, andOregon alone accounted for$10.6 billion (46 percent)[Ralph et al., 2014, Pielkeet al., 2002]
Boulder Flood, 2013
Research into hydroclimate extremes can provide moreaccurate and localized estimates of flood risk.
4 / 24
Introduction Background BHM Application Simulation References
More Motivation
I How are simulated fields of extremes linked to weathergeneration?
I May not need a full weather generator – extremes may be theonly quantity of interest
I Preserve spatial dependence of simulated extremes
I Downscaling
5 / 24
Introduction Background BHM Application Simulation References
What are BHM Spatial Extremes models typically used for?
I Return Levels
35
40
45
−120 −115 −110 −105lon
lat
1 2 5 13 30 69
100 year returnlevel [cm]
35
40
45
−120 −115 −110 −105lon
lat
1 2 5 13 30 69
100 year returnlevel 90% CI [cm]
6 / 24
Introduction Background BHM Application Simulation References
Background on Bayesian Statistics
From Bayes’ rule:
p(θ|y, x)︸ ︷︷ ︸Posterior
∝ p(y|θ, x)︸ ︷︷ ︸Likelihood
p(θ, x)︸ ︷︷ ︸Prior
θ Parameters
y Dependent data (response)
x Independent data (covariates/predictors/constants)
Posterior: The answer, probability distributions of parameters
Likelihood: A computable function of the parameters, modelspecific
Prior: Probability distribution, incorporates existingknowledge of the system, model specific
7 / 24
Introduction Background BHM Application Simulation References
Bayesian Hierarchical Model
In a hierarchical Bayesian model, expand terms using conditionaldistributions where θ = (θ1, θ2):
p(θ |y)︸ ︷︷ ︸Posterior
∝ p(y|θ1)︸ ︷︷ ︸Data Likelihood
p(θ1|θ2)︸ ︷︷ ︸Process Liklihood
p(θ2)︸ ︷︷ ︸Prior
Data Likelihood Relates observed data to distributionparameters
Process Likelihood Relates distribution parameters of the toeach other
8 / 24
Introduction Background BHM Application Simulation References
Statistics of Extremes
Given daily data, if we select the maximum value in each year,those data follow a generalized extreme value (GEV) distribution:
GEV(y; µ, σ, ξ) =1σ
b(−1/ξ)−1 exp{−b−1/ξ
}b = 1 + ξ
(x−µ
σ
), µ: Location, σ: Scale, ξ: Shape.
Return Level (quantile function):
zr = µ+σ
ξ[(− log(1− 1/r))−ξ− 1]
Where r is the return period inyears (100 years for example).
0.0
0.1
0.2
0.3
0.4
-5.0 -2.5 0.0 2.5 5.0y
f(y)
GEV(0,1,-0.5) GEV(0,1,0) GEV(0,1,0.5)
9 / 24
Introduction Background BHM Application Simulation References
Hierarchical spatial model
(Y(s1, t), . . . , Y(sn, t)) ∼ Cg[Σ; {µ(s), σ(s), ξ(s)}]Y(s, t) ∼ GEV[µ(s), σ(s), ξ(s)]
µ(s) = βµ,0 + xTµ(s)βµ(s) + wµ(s)
σ(s) = βσ,0 + xTσ (s)βσ(s) + wσ(s)
ξ(s) = βξ,0 + xTξ (s)βξ(s) + wξ(s)
wµ(s) ∼ GP(0, C(θµ))
wσ(s) ∼ GP(0, C(θσ))
wξ(s) ∼ GP(0, C(θξ))
Covariates: xT(s) = (Elevation, Mean Seasonal Precip)T
C: Stationary, isotropic, exponential covariance model
10 / 24
Introduction Background BHM Application Simulation References
Hierarchical spatial model
(Y(s1, t), . . . , Y(sn, t)) ∼ Cg[Σ; {µ(s), σ(s), ξ(s)}]Y(s, t) ∼ GEV[µ(s), σ(s), ξ(s)]
µ(s) = βµ,0 + xTµ(s)βµ(s) + wµ(s)
σ(s) = βσ,0 + xTσ (s)βσ(s) + wσ(s)
ξ(s) = βξ,0 + xTξ (s)βξ(s) + wξ(s)
wµ(s) ∼ GP(0, C(θµ))
wσ(s) ∼ GP(0, C(θσ))
wξ(s) ∼ GP(0, C(θξ))
Covariates: xT(s) = (Elevation, Mean Seasonal Precip)T
C: Stationary, isotropic, exponential covariance model
10 / 24
Introduction Background BHM Application Simulation References
Hierarchical spatial model
(Y(s1, t), . . . , Y(sn, t)) ∼ Cg[Σ; {µ(s), σ(s), ξ(s)}]Y(s, t) ∼ GEV[µ(s), σ(s), ξ(s)]
µ(s) = βµ,0 + xTµ(s)βµ(s) + wµ(s)
σ(s) = βσ,0 + xTσ (s)βσ(s) + wσ(s)
ξ(s) = βξ,0 + xTξ (s)βξ(s) + wξ(s)
wµ(s) ∼ GP(0, C(θµ))
wσ(s) ∼ GP(0, C(θσ))
wξ(s) ∼ GP(0, C(θξ))
Covariates: xT(s) = (Elevation, Mean Seasonal Precip)T
C: Stationary, isotropic, exponential covariance model
10 / 24
Introduction Background BHM Application Simulation References
Hierarchical spatial model
(Y(s1, t), . . . , Y(sn, t)) ∼ Cg[Σ; {µ(s), σ(s), ξ(s)}]Y(s, t) ∼ GEV[µ(s), σ(s), ξ(s)]
µ(s) = βµ,0 + xTµ(s)βµ(s) + wµ(s)
σ(s) = βσ,0 + xTσ (s)βσ(s) + wσ(s)
ξ(s) = βξ,0 + xTξ (s)βξ(s) + wξ(s)
wµ(s) ∼ GP(0, C(θµ))
wσ(s) ∼ GP(0, C(θσ))
wξ(s) ∼ GP(0, C(θξ))
Covariates: xT(s) = (Elevation, Mean Seasonal Precip)T
C: Stationary, isotropic, exponential covariance model
10 / 24
Introduction Background BHM Application Simulation References
Spatially varying regression coefficients
βµ(s) =k
∑i=1
cµi η
µi (s)
βσ(s) =k
∑i=1
cσi ησ
i (s)
βξ(s) =k
∑i=1
cξi η
ξi (s)
ηi(s) = exp(di/φi)
11 / 24
Introduction Background BHM Application Simulation References
Elliptical copula for data dependence
The Gaussian elliptical copula constructs the joint cdf of of arandom vector (V1, . . . , Vn) as
FC(v1, . . . , vn) = ΦΣ(u1, . . . , un) (1)
where ΦΣ(·) is the joint cdf of an n-dimensional multivariatenormal distribution with covariance matrix Σ, ui = φ−1(Fi[vi]),φ−1 is the inverse cdf (quantile function) of the standard normaldistribution and Fi(·) is the marginal GEV cdf of variable i.
–4 –2 0 2 4–4
–2
0
2
4
Normalized X1
Nor
mal
ized
X2
–4 –2 0 2 4–4
–2
0
2
4
Normalized X1
Nor
mal
ized
X2
0 0.5 10
0.5
1
F1(V1)
F2(V 2
)
–2 0 2 4 6 8
0
500
1000
V1
V 2
Gaussiantransformation
Cumulatedprobability
MultivariateGaussian model
12 / 24
Introduction Background BHM Application Simulation References
Elliptical copula for data dependenceThe corresponding joint pdf is
fC(y1, . . . , ym) =
m
∏i=1
fi[yi]
m
∏i=1
ψ[ui]
ΨΣ(u1, ....um) (2)
where fi is the marginal GEV pdf at site i, ψ is the standardnormal pdf and ΨΣ is the joint pdf of an m-dimensionalmultivariate normal distribution.
Exponential dependence matrix for spatial data:
Σ(i, j) = exp(−|| si− sj ||/a0) (3)
a0 is the copula range parameter. Termed the dependogram sincevalues are not covariances [Renard, 2011].
13 / 24
Introduction Background BHM Application Simulation References
Pros and Cons of Gaussian elliptical copulas
Benefits:
I Easy to implement
I Few parameters
I Flexible (can use any marginal distribution)
Requires checking two conditions in application:
I Asymptotic independence (dependence is hard to find inpractice)
I Multivariate normality after transformation (usually satisfied)
14 / 24
Introduction Background BHM Application Simulation References
Composite Likelihood (CL)
I Evaluating the full GP likelihood for 2500 stations is infeasiblein a Bayesian model
I One inversion (or Cholesky decomposition) of the covariancematrix takes seconds
I CL is an approximation of the full likelihood
I Stations are broken up into G groups each with ng stations
Lc(θ|y1, . . . , yG) =G
∏g=1
Lg(θ|yg) (4)
I CL Requires O(Gn3g) computations as opposed to O(n3).
I This approximation is applied to the copula as well as each ofthe GEV parameter residuals.
15 / 24
Introduction Background BHM Application Simulation References
Precipitation DataGlobal Historical ClimatologyNetwork (GHCN), daily totalprecip data
I ∼2500 stations with nearcomplete data from1948-2013
I 3 day aggregationwindow
I Fall maxima
Very large region/datasetfor typical Bayesian spatialmodel
●●●●●
●
●●●
●● ● ● ●●● ● ●●● ●● ●●● ●● ●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●●●
●
●●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●●
●
●
●
●●
●●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●●●●●
●
●
●
●
●●
●
●
●●
●
●
●
●●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●●●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●●
●
●
●
●
●● ●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●●●
●●
●●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●●
●
●●
●●
●
●
●
●● ●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●●
●
●●
●
●
●
●●
●
●
●●●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
● ●
●●
●
●
●
●●
●
●
●
● ●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●●
●●
●
●
●
●
●
●●
●
●
●
●
● ●●
●
●
●
●●
●
●
●●
●
●●
●
●
●
●
●●
●●
●●
●●
●
●
●
●
●
●●
● ●
●
●●●
●
●
●
●
● ●
●
●
●
●
●
●●
●
●
●●
●
●
●
●●
●
●
●●
●
●
●●
●
●
●
●●
●●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●●
●
●
●
●
●
●
●
●
●
●●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●●
●●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●●
● ●
●
●
●
●
●●●
●
●
●
● ●
●
●
●
●
●
●
●● ●
●
●●
● ●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●●
●●
●
●
●
●●
●
●
●
●
●
●
●●
●●
●●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●● ●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●●
●
●●
●
●
●
●
●●
●●●
●●
●
●
● ●●
●
●
●
●●
●●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●●
●
●
●
●
●
●●●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●●●
●
● ●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
● ●
●
●●
●
●●
●
●●
●
●
●
●●●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
● ●
●
●●
●●
●
●
●
●
●
●●
●
●
●●●
●●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
● ● ●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●●
●
●
●
● ●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●●
●
●
●● ●
●●
●
●
●
●
●●
●
●
●●●
●
●
●
●
● ●
●
●●●
●
●
● ●
●
● ●
●
●
●
●
●
●
●●
●●●
●
●
●●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●●
●
●
●
●
●
●
●
●
●●
●
●●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●●
●
●
●
●●
●●
●
●●● ●
●
●
●
●●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●
●
●●
●●●
●●
●
●●
●
●
●
●
●
●
● ●●
●
● ●●
●
●
●●
●
●
●
●●●
●
●
●
●
●
●
●●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
●●●●
●
●
● ●
●
●
●
●
●
●
●●
●
●
●
● ●
●
● ●
●
●
●
● ●
●
●
●
●
●●
●
●●
●●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●●●
●
●
●
●
●●●
●
●
●
● ●
●
●
●
●●
●
●
●
●●
●●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
● ●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●●
●
●●
●
●
●●
●
●
●
●
●●
●
●
●●
●●
●
●
●●
●
●
●●
●
●
●
●●
●
●
●
●●
●●●
●●●●●
●●
●●
●
●
●●
●●●●●●
●● ●●
●●●
●●
●●●
●●
●●●
●
●
●
●
●●
●●
●
●●●●●●●
●●
●
●
● ●
●●
●●●
●●
●
●
●
●
●
●
●
●
●
●●●
●●
●
●●●
●●
● ●
●●●●
●●●
●
●
●
●
●
●●
●● ●●
●●
●● ●●●
●●●
●
●●●●●
●●
●●
●●●●● ●●●
●●
●●
●
●
●●
●●
●●●
●●
●● ●
● ●●●
●●●●
●●
●
●
●●●●
● ●
●
●●●
●●
●●●
●
●●●
●
●●●
●●
●
●●
●
●
●●
●
●
●
●
●●●●
●●●●
●
●
●●
●
●
●
●●
●
●
●
● ●●
●
●
●
●●
●●
●
●●●●
●
●●
●●
●
●●
●●
●
●
●
●●●●●
●●
●●
●●
●●
●
●
●●●
●
●
●
●●●●●
●●●●
●
●●●
●
●●
●●●●
●
●
●●●
●●
●●
●●
●●
●●●●
●
●●●●
●●●
●
●●●●
●●●●●●●
●●●
●●
●
●
●●
●●
●
●●●
●●●●●●●●
●●
●●●
●●●
●
●
●
●
●●
●
●
●●●●●●●●●●
●●
●●● ●●
●
●●
●
●●
●●●●
●●●●●●
●●
●●●●●
●
●
●
●
●
●●
●
●●●
●●●●●
●●
●●●●
●●
●●
●●
●
●●
●
●●●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●
●
●●
●●● ●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
● ●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●●
●
●
●
●
●
35
40
45
−120 −115 −110 −105lon
lat ●
●
CompleteIncompleteKnot
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
● ●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
35
40
45
−120 −115 −110 −105lon
lat ●
●
CompleteIncomplete
16 / 24
Introduction Background BHM Application Simulation References
Simulation procedure
Fit the modelDraw posterior sample
Compute latent regression
coefficient surfaces
Compute latent parameter means
Conditionally simulate latent parameter residuals
Unconditional unit Fréchet copula simulation
Extreme value simulation!
Latent GEV parameter fields
GEV Transformation
17 / 24
Introduction Background BHM Application Simulation References
Regression coefficient surfaces - one posterior sampleLocation Scale Shape
35
40
45
50
-125 -120 -115 -110 -105lon
lat
35
40
45
50
-125 -120 -115 -110 -105lon
lat
35
40
45
50
-125 -120 -115 -110 -105lon
lat
35
40
45
50
-125 -120 -115 -110 -105lon
lat
35
40
45
50
-125 -120 -115 -110 -105lon
lat
35
40
45
50
-125 -120 -115 -110 -105lon
lat
Mea
n se
ason
al p
reci
pEl
evat
ion
18 / 24
Introduction Background BHM Application Simulation References
GEV Parameter simulation = conditional sim + mean(covariates + betas)
µ(s) = βµ,0 + xTµ(s)βµ(s) + wµ(s)
35
40
45
50
−125 −120 −115 −110 −105lon
lat
0 10 20Location
35
40
45
50
−125 −120 −115 −110 −105lon
lat
0 1 2 3 4 5Scale
35
40
45
50
−125 −120 −115 −110 −105lon
lat
−2 −1 0Shape
19 / 24
Introduction Background BHM Application Simulation References
Unit Frechet Copula Simulations
35
40
45
50
−125 −120 −115 −110 −105lon
lat
35
40
45
50
−125 −120 −115 −110 −105lon
lat
20 / 24
Introduction Background BHM Application Simulation References
GEV from unit Frechet
If Y is a random variable with a GEV distribution with location µ,scale σ and shape ξ. Then,
Z = [1 + ξ(Y− µ)/σ]1/ξ
is unit Frechet distributed i.e. Z ∼ GEV(1,1,1).
If Z is a unit Frechet random variable. Then,
Y = µ + σ(Zξ − 1)/ξ
is unit GEV distributed with location, scale and shape parametersequal to µ, σ and ξ respectively.
21 / 24
Introduction Background BHM Application Simulation References
Extreme precipitation simulations
sim1 sim2 sim3
35
40
45
50
-125 -120 -115 -110 -105 -125 -120 -115 -110 -105 -125 -120 -115 -110 -105lon
lat
2 5 11 26Precip [mm]
Threshold of 1 cm.
22 / 24
Introduction Background BHM Application Simulation References
Discussion and Contributions
Discussion:
I MCMC sampling takes 3+ days!
I Conditional simulation is expensive
I Gaussian elliptical copula is an alternative to a max stableprocess when some conditions hold
Conclusions:
I Bayesian spatial extremes model for return levels
I Can be used for simulations of extremes on a grid at arbitraryresolution
I Composite likelihood and spatially varying regressioncoefficients make it feasible for larger regions
Thanks!
23 / 24
Introduction Background BHM Application Simulation References
Discussion and Contributions
Discussion:
I MCMC sampling takes 3+ days!
I Conditional simulation is expensive
I Gaussian elliptical copula is an alternative to a max stableprocess when some conditions hold
Conclusions:
I Bayesian spatial extremes model for return levels
I Can be used for simulations of extremes on a grid at arbitraryresolution
I Composite likelihood and spatially varying regressioncoefficients make it feasible for larger regions
Thanks!
23 / 24
Introduction Background BHM Application Simulation References
Discussion and Contributions
Discussion:
I MCMC sampling takes 3+ days!
I Conditional simulation is expensive
I Gaussian elliptical copula is an alternative to a max stableprocess when some conditions hold
Conclusions:
I Bayesian spatial extremes model for return levels
I Can be used for simulations of extremes on a grid at arbitraryresolution
I Composite likelihood and spatially varying regressioncoefficients make it feasible for larger regions
Thanks!
23 / 24
Introduction Background BHM Application Simulation References
References I
R A Pielke, M W Downton, and JZB Miller. Flood damage in the United States, 1926-2000: a reanalysis ofNational Weather Service estimates, 2002.
F M Ralph, M Dettinger, A White, D Reynolds, D Cayan, T Schneider, R Cifelli, K Redmond, M Anderson,F Gherke, J Jones, K Mahoney, L Johnson, S Gutman, V Chandrasekar, J Lundquist, N Molotch, L Brekke,R Pulwarty, J Horel, L Schick, A Edman, P Mote, J Abatzoglou, R Pierce, and G Wick. A Vision for FutureObservations for Western U.S. Extreme Precipitation and Flooding. Journal of Contemporary Water Research& Education, 153(1):16–32, April 2014.
B Renard. A Bayesian hierarchical approach to regional frequency analysis. Water Resources Research, 47(W11513):1–21, 2011.
24 / 24