Get the full ebook with Bonus Features for a Better Reading Experience on ebookgate.

com

Uncertainty Quantification Theory Implementation

and Applications 1st Edition Ralph C. Smith

https://ebookgate.com/product/uncertainty-quantification-
theory-implementation-and-applications-1st-edition-ralph-c-
smith/

OR CLICK HERE

DOWLOAD NOW

Download more ebook instantly today at https://ebookgate.com

Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Memory Management Algorithms and Implementation in C C 1st

Edition Bill Blunden

https://ebookgate.com/product/memory-management-algorithms-and-
implementation-in-c-c-1st-edition-bill-blunden/

ebookgate.com

Earthquake Engineering Theory and Implementation Second

Edition Nazzal Armouti

https://ebookgate.com/product/earthquake-engineering-theory-and-
implementation-second-edition-nazzal-armouti/

ebookgate.com

Quantification 1st Edition Anna Szabolcsi

https://ebookgate.com/product/quantification-1st-edition-anna-
szabolcsi/

ebookgate.com

Envy Theory and Research Richard Smith (Editor)

https://ebookgate.com/product/envy-theory-and-research-richard-smith-
editor/

ebookgate.com
Adaptive Design Theory and Implementation Using SAS and R
1st Edition Mark Chang

https://ebookgate.com/product/adaptive-design-theory-and-
implementation-using-sas-and-r-1st-edition-mark-chang/

ebookgate.com

Cryogenics Theory Processes and Applications Theory

Processes and Applications 1st Edition Allyson E. Hayes

https://ebookgate.com/product/cryogenics-theory-processes-and-
applications-theory-processes-and-applications-1st-edition-allyson-e-
hayes/
ebookgate.com

General Chemistry Principles and Modern Applications 10th

Edition Ralph H. Petrucci

https://ebookgate.com/product/general-chemistry-principles-and-modern-
applications-10th-edition-ralph-h-petrucci/

ebookgate.com

Statistical Shape and Deformation Analysis Methods

Implementation and Applications 1st Edition Edition Guoyan
Zheng
https://ebookgate.com/product/statistical-shape-and-deformation-
analysis-methods-implementation-and-applications-1st-edition-edition-
guoyan-zheng/
ebookgate.com

Software Quality Assurance From Theory to Implementation

1st Edition Daniel Galin

https://ebookgate.com/product/software-quality-assurance-from-theory-
to-implementation-1st-edition-daniel-galin/

ebookgate.com
Uncertainty Quantification
Computational Science & Engineering
The SIAM series on Computational Science and Engineering publishes research monographs, advanced
undergraduate- or graduate-level textbooks, and other volumes of interest to an interdisciplinary CS&E
community of computational mathematicians, computer scientists, scientists, and engineers. The series
includes both introductory volumes aimed at a broad audience of mathematically motivated readers
interested in understanding methods and applications within computational science and engineering and
monographs reporting on the most recent developments in the field. The series also includes volumes
addressed to specific groups of professionals whose work relies extensively on computational science and
engineering.

SIAM created the CS&E series to support access to the rapid and far-ranging advances in computer
modeling and simulation of complex problems in science and engineering, to promote the interdisciplinary
culture required to meet these large-scale challenges, and to provide the means to the next generation of
computational scientists and engineers.

Editor-in-Chief
Donald Estep
Colorado State University

Editorial Board
Omar Ghattas Max D. Morris
University of Texas at Austin Iowa State University

Max Gunzburger Alex Pothen

Florida State University Purdue University

Des Higham Padma Raghavan

University of Strathclyde Pennsylvania State University

Michael Holst Karen Willcox

University of California, San Diego Massachusetts Institute of Technology

David Keyes
Columbia University and KAUST

Series Volumes
Kuzmin, Dmitri and HCimCilCiinen, Jari, Finite Element Methods for Computational Fluid Dynamics:
A Practical Guide
Rostamian, Rouben, Programming Projects in C for Students of Engineering, Science, and
Mathematics
Smith, Ralph C., Uncertainty Quantification: Theory, Implementation, and Applications

Dankowicz, Harry and Schilder, Frank, Recipes for Continuation

Mueller, Jennifer L. and Siltanen, Samuli, Linear and Nonlinear Inverse Problems with Practical
Applications
Shapira, Vair, Solving PDEs in C++: Numerical Methods in a Unified Object-Oriented Approach,
Second Edition
Borzl, Alfio and Schulz, Volker, Computational Optimization of Systems Governed by Partial
Differential Equations
Ascher, Uri M. and Greif, Chen, A First Course in Numerical Methods

Layton, William, Introduction to the Numerical Analysis of Incompressible Viscous Flows

Ascher, Uri M., Numerical Methods for Evolutionary Differential Equations

Zohdi, T. I., An Introduction to Modeling and Simulation of Particulate Flows

Biegler, Lorenz T., Ghattas, Omar, Heinkenschloss, Matthias, Keyes, David, and van Bloemen
Waanders, Bart, Editors, Real-Time PDE-Constrained Optimization

Chen, Zhangxin, Huan, Guanren, and Ma, V uanle, Computational Methods for Multiphase Flows
in Porous Media
Shapira, V air, Solving PDEs in C++: Numerical Methods in a Unified Object-Oriented Approach
 RALPH C. SMITH
North Carolina State University
Raleigh, North Carolina

Uncertainty Quantification
Theory, Implementation, and Applications

5.La.11l.. .
•

Society for Industrial and Applied Mathematics

Philadelphia
Copyright © 2014 by the Society for Industrial and Applied Mathematics

10 9 8 7 6 5 4 3 2

All rights reserved. Printed in the United States of America. No part of this book may be
reproduced, stored, or transmitted in any manner without the written permission of the publisher.
For information, write to the Society for Industrial and Applied Mathematics, 3600 Market Street,
6th Floor, Philadelphia, PA 19104-2688 USA.

Trademarked names may be used in this book without the inclusion of a trademark symbol. These
names are used in an editorial context only; no infringement of trademark is intended.

MATLAB is a registered trademark of The MathWorks, Inc. For MATLAB product information, please
contact The MathWorks, Inc., 3 Apple Hill Drive, Natick, MA 01760-2098 USA, 508-647-7000,
Fax: 508-647-7001, [email protected], www.mathworks.com.

Figures 1.2, 2.1, and 2.4 reprinted courtesy of NOAA.

Figure 2.3 reprinted with permission from ECMWF.

Figures 2.6, 2.9, and 2.11 reprinted with permission from IPCC.

Figures 2.12 and 2.15 reprinted courtesy of Kansas Geological Survey. All rights reserved.

Figure 2.13 reprinted courtesy of USGS.

Figure 2.14 reprinted courtesy of Mike Norton.

Figure 2.16 reprinted courtesy of USNRC.

Figure 2.17 reprinted courtesy of the U.S. Department of Energy.

Figures 2.18 and 2.19 reprinted with kind permission from Springer Science and Business Media.

Figures 6.7 (a), 8.8, and 8.9 (b) reprinted with permission from SAGE.

This research was supported by the Air Force Office of Scientific Research (Dynamics and Control
Program), the Department of Energy Consortium for Advanced Simulation of Light Water Reactors
(CASL), and the National Science Foundation (Research Training Groups in the Mathematical
Sciences).

Library of Congress Cataloging-in-Publication Data

Smith, Ralph C., 1960- author.
Uncertainty quantification : theory, implementation, and applications/ Ralph Smith, North
Carolina State University, Raleigh, North Carolina.
pages cm. - (Computational science and engineering series)
Includes bibliographical references and index.
ISBN 978-1-611973-21-1
1. Measurement uncertainty (Statistics) 2. Estimation theory. I. Title.
QA276.8.S64 2013
519.5'44-dc23
2013034432

SJaJ11. is a registered trademark.

 Contents

Preface ix
Notation xiii
Acronyms and Initialisms xvii
1 Introduction 1
1.1 Nature o f Uncertainties and Errors 4
1 .2 Predictive Estimation 8

2 Large-Scale Applications 11
2.1 Weather Models . . . 11
2.2 Climate Models . . . 21
2.3 Subsurface Hydrology and Geology 33
2.4 Nuclear Reactor Design . 36
2.5 Biological Models 44

3 Prototypical Models 51
3.1 Models . . . . . 51
3.2 Evolution, Stationary, and Algebraic Models 61
3.3 Abstract Modeling Framework . . . 63
3.4 Notation for Parameters and Inputs 65
3.5 Exercises . . . . . . . . . . . . . . . 66

4 Fundamentals of Probability, Random Processes, and Statistics 6 7

4.1 Random Variables, Distributions, and Densities . . . . . . . . 67
4.2 Estimators, Estimates, and Sampling Distributions . . . . . . 79
4.3 Ordinary Least Squares and Maximum Likelihood Estimators 82
4.4 Modes of Convergence and Limit Theorems . 85
4.5 Random Processes . . . . . . . . . . . . . . . . . . 87
4.6 Markov Chains . . . . . . . . . . . . . . . . . . . 90
4.7 Random versus Stochastic Differential Equations . 96
4.8 Statistical Inference . 98
4.9 Notes and References 104
4.10 Exercises . . . . . . . 105

v
vi Contents

5 Representation of Random Inputs 107

5.1 Mutually Independent Random Parameters . 107
5.2 Correlated Random Parameters . . . . . . . 108
5.3 Finite-Dimensional Representation of Random Coefficients 109
5.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

6 Parameter Selection Techniques 113

6.1 Linearly Parameterized Problems . . . . . . 115
6.2 Nonlinearly Parameterized Problems . . . . 1 22
6.3 Parameter Correlation versus Identifiability . 125
6.4 Notes and References 127
6.5 Exercises . . . . . . . . . . . . . . . . . . . . 1 28

7 Frequentist Techniques for Parameter Estimation 131

7.1 Parameter Estimation from a Frequentist Perspective 133
7.2 Linear Regression . . . . . . . . . . . . . . 1 34
7.3 Nonlinear Parameter Estimation Problem . 141
7.4 Notes and References 152
7.5 Exercises . . . . . . . . . . . . . . . . . . . 153

8 Bayesian Techniques for Parameter Estimation 155

8.1 Parameter Estimation from a Bayesian Perspective 155
8.2 Markov Chain Monte Carlo (MCMC) Techniques 159
8.3 Metropolis and Metropolis-Hastings Algorithms . 159
8.4 Stationary Distribution and Convergence Criteria 168
8.5 Parameter Identifiability . . . . . . . . . . . . . . 171
8.6 Delayed Rejection Adaptive Metropolis (DRAM) 1 72
8.7 DiffeRential Evolution Adaptive Metropolis (DREAM) 181
8.8 Notes and References 1 84
8.9 Exercises . . . . . . . . . . . . . . 1 84

9 Uncertainty Propagation in Models 187

9.1 Direct Evaluation fo r Linear Models . 1 88
9.2 Sampling Methods . . . 191
9.3 Perturbation Methods . 192
9.4 Prediction Intervals . 197
9.5 Notes and References 203
9.6 Exercises . . . . . . . 204

10 Stochastic Spectral Methods 207

10.1 Spectral Representation of Random Processes . . . . . . . . 207
10.2 Galerkin, Collocation, and Discrete Proj ection Frameworks . 214
10.3 Stochastic Galerkin Method-Examples . 226
10.4 Discrete Proj ection Method-Example 234
10.5 Stochastic Polynomial Packages 235
10.6 Exercises . . . . . . . . . . . . . . . . . 236
Contents vii

11 Sparse Grid Quadrature and Interpolation Techniques 239

11.1 Quadrature Techniques . . . . . . . . . . . 239
1 1 .2 Interpolating Polynomials for Collocation . 250
1 1 .3 Sparse Grid Software 254
1 1 .4 Exercises . . . . . . . . . . . . . . . . . . . 255

12 Prediction in the Presence of Model Discrepancy 257

12.1 Effects of Unaccommodated Model Discrepancy 261
12.2 Incorporation of Missing Physical Mechanisms . . 263
1 2 .3 Techniques to Quantify Model Errors . . . . . . . 265
12.4 Issues Pertaining to Model Discrepancy Representations 267
1 2 .5 Notes and References 269
1 2 .6 Exercises . . 269

13 Surrogate Models 271

13.1 Regression o r Interpolation-Based Models . 273
1 3 .2 Projection-Based Models . . . . . . . . . 280
1 3 .3 Eigenfunction or Modal Expansions . . . . 283
13.4 Snapshot-Based Methods including POD . 284
1 3 .5 High-Dimensional Model Representation (HDMR) Techniques 289
13.6 Surrogate-Based Bayesian Model Calibration . 298
13.7 Notes and References 299
1 3 .8 Exercises . . . . . . . . 300

14 Local Sensitivity Analysis 303

14.1 Motivating Examples-Neutron Diffusion . . . . . . . 306
14.2 Functional Analytic Framework for FSAP and ASAP 312
14.3 Notes and References 318
14.4 Exercises . . . . . . . . 319

15 Global Sensitivity Analysis 321

15.1 Variance-Based Methods 323
1 5 .2 Morris Screening . . . . . . . . . . . . 331
1 5 .3 Time- or Space-Dependent Responses 337
15.4 Notes and References 343
1 5 .5 Exercises . . . . . . . . . . . . . . 344

A Concepts from Functional Analysis 345

A.l Exercises . . . . . . . . . . . . . . 351

Bibliography 353
Index 373
 Preface

Uncertainty quantification is both a new field and one that is as old as the
disciplines of probability and statistics. The present novelty lies in the synthesis
of probability, statistics, model development , mathematical and numerical analysis,
large-scale simulations, experiments, and disciplinary sciences to provide a compu­
tational framework for quantifying input and response uncertainties in a manner
that facilitates predictions with quantified and reduced uncertainty. This is the
topic of this book.
Uncertainty quantification for physical models can be motivated in the context
of weather modeling. Models for complex phenomena, such as dust-induced cloud
formation, are approximate and uncertain, as are the parameters in these models.
Additional errors and uncertainties are introduced by the numerical algorithms and
experimental data used to approximate and calibrate the models. In the first step
of the prediction process, data assimilation or model calibration techniques are used
to determine input parameters and initial conditions so that quantities of interest ,
such as temperature or relative humidity, match current conditions. The second step
entails the prediction of future weather conditions with uncertainties quantified by
probabilistic statements-e.g., 95% change of rain-or uncertainty cones of the type
reported for hurricanes or tropical storms .
Whereas model calibration and uncertainty propagation comprise the primary
aspects of the prediction process , their implementation for large-scale applications
requires a wide range of supporting topics. These include aspects of probability,
statistics, analysis , and numerical analysis as well as the following topics : param­
eter selection, surrogate model construction, local and global sensitivity analysis ,
and quantification of model discrepancies . The interdisciplinary nature of the field
is augmented by the fact that all of these components must be investigated and
implemented in the context of the underlying applications.
The explosive growth of uncertainty quantification as an interdisciplinary field
is due to a number of factors: increasing emphasis on models having quantified
uncertainties for large-scale applications , novel algorithm development , and new
computational architectures that facilitate implementation of these algorithms .
In Chapter 2 , we detail five applications where model predictions with quanti­
fied uncertainties are critical for understanding and predicting scientific phenomena
and making informed decisions and designs based on these predictions. These ap­
plications are weather models, climate models, subsurface hydrology and geology

ix
x Preface

models, nuclear reactor models, and models for biological phenomena. Whereas the
presence and role of uncertainties in these applications has long been recognized,
the development of computational models that quantify and incorporate uncertain­
ties is receiving increased attention. The reliance of scientists and policy makers on
such models is expected to grow rapidly as the field of uncertainty quantification
for predictive sciences matures and computational resources evolve .
The relatively recent development of supporting mathematical and statistical
theory and algorithms is a second factor supporting the growth of the field. For
example, the adaptive DRAM and DREAM algorithms discussed in Chapter 8 for
Bayesian model calibration were developed within the last ten years. These algo­
rithms are presently being investigated in the context of climate and groundwater
models. Similarly, much of the sparse grid theory discussed in Chapter 1 1 was
developed in the last twenty years, although the original concept is much older.
The availability of massively parallel computer architectures and hardware
has further bolstered uncertainty quantification for complex and large-scale appli­
cations. The DREAM algorithms are inherently parallel, and recent versions of
DRAM are being implemented on parallel architectures. It is anticipated that field
programmable gate arrays ( FPGAs ) will be increasingly utilized for uncertainty
quantification as high-level tools are developed to reduce programming overhead.
The fact that we operate in increasingly data-rich environments will also benefit
uncertainty quantification, and we anticipate increased interaction between data
mining, high-dimensional visualization, and uncertainty quantification.
The growth in the field has spawned the introduction of interdisciplinary
courses on uncertainty quantification, and this text owes its genesis to the author's
development of such a course at North Carolina State University in 2008. This text
was written with the goal of introducing advanced undergraduates, graduate stu­
dents, postdocs, and researchers in mathematics, statistics, engineering, and natural
and biological sciences to the various topics comprising uncertainty quantification
for predictive models. To achieve this , we motivate a number of the topics using
very basic examples that should be familiar to most readers . We have included
numerous definitions and significant detail to provide a common footing for a wide
range of readers. Because this is a new and evolving field, we indicate open research
questions at various points in the text and provide research references in the Notes
and References at the end of each chapter.
Various resources will be maintained at the website http ://www.siam.org/
books/cs12 to augment the text and provide a mechanism to update the material.
This includes data employed in exercises as well as a future erratum.
This text has benefited significantly from graduate students, postdocs, and
colleagues whose comments have improved the exposition and reduced the num­
ber of typos by orders of magnitude. Specifically, sincere thanks are extended to
Nate Burch, Amanda Coons, John Crews , John Harlim, Zhengzheng Hu, Dustin
Kapraun, Zack Kenz , Christine Latten, Jerry McMahan Jr. , Keri Rehm, Marni
Wentworth, and Lucus Van Blaircum for their attention to detail and candid feed­
back regarding parts of the manuscript . The author is also extremely grateful to
Brian Adams and Karen Willcox for their feedback during the review process; the
book is significantly improved due to their detailed comments.
Preface xi

The support provided by several funding agencies has been instrumental both
for related research and the writing of this text . These agencies include the Air Force
Office of Scientific Research ( Dynamics and Control Program ) , the Department of
Energy Consortium for Advanced Simulation of Light Water Reactors ( CASL ) , and
the National Science Foundation ( Research Training Groups in the Mathematical
Sciences ) . Part of this text was written while the author was a Faculty Fellow
in the 20 1 1-12 Statistical and Applied Mathematical Sciences Institute ( SAMSI)
Program on Uncertainty Quantification. Collaboration and interactions during this
year significantly influenced aspects of the book, and the author very gratefully
acknowledges the scientific and financial contributions from this program. Finally,
I would like to thank Elizabeth Greenspan from SIAM for her assistance and en­
couragement throughout the process of writing this book.

Ralph C. Smith
North Carolina State University
Raleigh, NC
July 1 , 2013
 Notation

This compilation does not include all of the symbols used throughout the text ,
and we neglect those that appear one time in a specific context such as those in the
models of Chapter 2. Instead, it is meant to clarify the role of symbols that appear
multiple times throughout the discussion.

Symbol Meaning Page

OD, 80. Boundaries of regions D and 0. 62, 63
a(q * lq k - 1 ) Probability of accepting candidate q * 159
'Yi Normalization factor for (- , · ) P 209
ri , r Range of ith random parameter, random vector 108
8, 8(xi ), 8(ti) Model discrepancy or error 133, 257
€1E Random and realized measurement errors 82, 132
AM1 Lebesgue constant 252
µ Mean 70
µi , µ; Morris sensitivity measures 332
I/ Dimension of model response y(t, q) 61
7ro (q), 7r(qly) Bayesian prior and posterior density 100
7r(ylq) Bayesian likelihood function 100
pq, (qi), pq (q) Density for ith random parameter, random vector 108
u02 Unknown measurement error variance 135
&2 , u 2 Estimator and estimate for u5 135
Uj Singular values of the matrix A 117
E Matrix of singular values of matrix A 117
v Realized measurements of Y 132, 156
y Random variable for measurements 82
<Pi(x) Spatial basis functions 219
x Independent variables X = [x, t] E D x 7 =: 0. 63
1/J k (Q), '1!k (Q) Univariate, multivariate orthogonal polynomials 209, 213
A (q, p) Sparse grid quadrature operator 247
B(u, q), B(q)u Boundary operators 62, 63

xiii
xiv Notation

Symbol Meaning Page

c Observation matrix or vector 61
di (q), df (q) Morris elementary effects for ith input 331
D, Di, D ij Total and partial variances of response Y 324
D Spatial domain in R 1 , R 2 , or R 3 62
f(q), f (t , q)
x, Model response 132
f(q), f( t, q)x, Surrogate model 274
F(q) Source terms 62
Hu , Hq, HF Hilbert spaces for state, parameters, and source 63
Hi (Q) Hermite polynomials 210
1-l (q, p) Sparse quadrature grid 248
i' , j', k' Multi-indices 212
J(P) f Integral operator in RP 240
Il(f) Multi-index sets 246
Ie(p ) u Interpolation operator in RP 254
I(q) Identifiable subspace 113
I(q) Space of influential parameters 1 14
:J(q) Least squares functional 135
J(q * lq k - 1 ) Proposal or j umping distribution 159
f(q l v) Log-likelihood function 83
L(q l v) Likelihood function 83
L(q)u Linear operator 63
L m (q) Lagrange interpolating polynomial 251
L�, (ri), L� (r) Square integrable functions on r i , r 215
M, Me Number of collocation points or samples 253
n Number of measurements or model evaluations 61
N Dimension of state u 61
NI(q) Unidentifiable subspace 113
NI(q) Space of noninfluential parameters 1 14
N(A) Null space of the matrix A 116
N(u, q) Linear or nonlinear differential operator 62
p Number of parameters 100
Pi (Q) Legendre polynomials 211
IP' k Space of polynomials with argument less
than or equal to k 208
IP' k Polynomials in IP' k that are orthogonal to IP' k - l 208
Notation xv

Symbol Meaning Page

qo True but unknown parameter 82
q = [q1 , . . . ' qp ] Realizations of Q 100
q* Proposed Markov chain parameter 159
qk- 1 Parameter at k 1 step in Markov chain
- 159
qr , qm Quadrature , collocation, and sample points 211 , 217
fi.oLs , qOLS Least squares estimator, estimate for q0 82
qM AP Maximum a posteriori estimate 157
qMLE Maximum likelihood estimate 84
Q = [Q 1 , . . . ' Qp ] Random vector of parameters 100
Q Orthogonal matrix in QR factorization 118
Q Admissible parameter space 82
Q Sample space 82
Q( P ) Quadrature operator in RP 240
r Rank of matrix A 117
R, Rt Number of quadrature points 243
R Upper triangular matrix in QR factorization 118
R, R Residual estimator and estimate 136
R Number of sparse grid quadrature points 248
R (u, q) General observation or response 63
R (A) Range of the matrix A 116
Bi Local sensitivity indices 192 , 322
S'!i Sigma-normalized sensitivity indices 322
si , si j , sT, Sobol sensitivity indices 324
SSq Sum of squares error 156
7 Temporal domain 63
u (q), u (t, q)
x, State variable 61
u(q), u(t, q)x, Surrogate state representation 279
V , VJ Spaces of spatial test functions 219
vk Chain covariance matrix 172
wr Quadrature weights 21 1
x Deterministic n x p design matrix 131
X(q) Sensitivity matrix 144
y Realizations of Y 132
y Random variable for model response 321
z, z K Spaces of parameter test functions 219
 Acronyms and lnitialisms

Term Meaning Page

ANOVA Analysis of variance 291
AR Autoregressive (model) 89
ASAP Adj oint sensitivity analysis procedure 306
BWR Boiling water reactor 36
cadlag Continue a droite, limite a gauche 69
CASL Consortium for Advanced Simulation of Light 37
Water Reactors
cdf Cumulative distribution function 68
CESM Community Earth System Model 30
CFCs chlorofluorocarbons 25
CRUD Chalk River unidentified deposit 43
CVTs Centroidal Voronoi tesselations 285
DAKOTA Design Analysis Kit for Optimization and 236
Terascale Applications
DOE Department of Energy 37
DRAM Delayed rej ection adaptive Metropolis 172
DREAM DiffeRential Evolution Adaptive Metropolis 181
ECMWF European Centre for Medium-Range Weather Forecasts 16
FPGAs Field programmable gate arrays x

FSAP Forward sensitivity analysis procedure 306

gcd Greatest common divisor 94
GCR Gas-cooled reactor 36
GP Gaussian process 89
gPC Generalized polynomial chaos 207
HDMR High-dimensional model representation 289
HIV Human immunodeficiency virus 45
iid Independent and identically distributed 79
xvii
xviii Acronyms and l n itialisms

Term Meaning Page

IPCC Intergovernmental Panel on Climate Change 32
kde Kernel density estimation 75
LANL Los Alamos National Laboratory 41
LQR Linear quadratic regulator 50
MAP Maximum a posteriori (estimate) 157
MCMC Markov chain Monte Carlo 159
MLE Maximum likelihood estimate 84
NISP Nonintrusive spectral projection 225
NWP Numerical weather prediction 16
ODE Ordinary differential equation 51
OLS Ordinary least squares 82
ORNL Oak Ridge National Laboratory 41
PC Polynomial chaos 207
PCA Principal component analysis 109
PDE Partial differential equation 51
pdf Probability density function 69
POD Proper orthogonal decomposition 285
PRA Probabilistic risk assessment 44
PWR Pressurized water reactor 36
Q-Q Quantile-quantile 74
Qol Quantity of interest 4
SAMS I Statistical and Applied Mathematical Sciences Institute xi
SDE Stochastic differential equation 97
SIR Susceptible , infected, recovered (model) 55
SVD Singular value decomposition 117
WMO World Meteorological Organization 17
Cha pte r 1

Introduction

The synthesis of modeling, large-scale simulations, and experiments has long been
recognized as critical for understanding and advancing the state of science and
technology. When considered in the broad sense of including requisite theory, these
form the pillars of predictive science, as illustrated in Figure 1 . 1 . In the context of
predictive science, uncertainty quantification can be broadly defined as the science
of identifying, quantifying, and reducing uncertainties associated with models, nu­
merical algorithms , experiments , and predicted outcomes or quantities of interest .
Aspects of this field, such as the quantification of measurement uncertainties and
numerical errors , are well understood and are addressed by classical statistics and
numerical analysis theory. However, the systematic quantification of uncertainties
and errors in models, simulations, and experiments and the analysis of how they
are propagated through complex models to affect predicted outcomes is more recent
and constitutes both an active area of research and the subject of this text .
In Chapter 2 , we detail five large-scale applications where model predictions
with quantified uncertainties are critical for understanding and predicting scientific
phenomena and making informed decisions and designs based upon these predic­
tions. These applications are weather models, climate models, subsurface hydrology
and geology models , nuclear reactor designs, and models for biological phenomena.

�\� el Calibration
d Validation

- I Quantity of Interest (Qol) I

Figure 1 . 1 . Modeling, numerical, and experimental components of predictive sci­

ence with associated uncertainties and errors indicated in gray.

1
2 Chapter 1 . I ntrod uction

In the following example, we summarize aspects of the weather model detailed

in Section 2.1 to motivate sources of uncertainty and indicate issues that must be
addressed when making predictions.

Example 1 . 1 (Weather Prediction).

The physical components o f meteorological o r weather models are constructed
by quantifying the interactions between temperature and pressure gradients, wind,
and precipitation using conservation of energy, mass, and momentum. When com­
bined with conservation of water phases and aerosol concentrations, this yields the
equations of atmospheric physics,
oap +
V' . ( pv ) = o ,
t
av 1 A

-
at = -v · V'v - P -V'p - gk - 2n x v '
8T
pcvat + pV' · v = -V' · F + V' · (kV'T) + pq(T,p, p),
(1 . 1)
p= pRT,
om·
a/ = · Y'mj +Sm; (T, mJ> p) , j = 1 , 2 , 3 ,
-v Xj,

oax.! =-v·Y'xj+Bx; (T,XJ>P) , j=l, . . . ,J,

where p, v, T,p, k, and cv respectively denote the density, velocity, temperature,

m1, m2, m3,

pressure , thermal conductivity, and specific heat of air. The concentration of water
in solid, liquid, and gaseous phases is denoted by and whereas the
concentration of the lh aerosol species is denoted by Xj.

m
For the reasons discussed in Section 2.1 , one typically constructs phenomeno­
logical models for the source terms Sm; (T, j , Xj, p) and Bx; (T, Xj, p). For example,
it is established in (2.8) that Sm2 can be formulated as

(1 .2)
where

81 = j5(m2 - [ ( do(m�"_ m:2)) ]-l (1 .3)

m;) 2 1 .2 x 10 - 4 + 1 .569 x 10 - 1 2

requires the specification of the nonphysical parameters j5, m;, and do. The nr ,
remaining components have similar formulations.
Model Errors or Discrepancies. Both the conservation relations (1 . 1) and phe­
nomenological closure equations (1 .2) and (1 .3) are approximations of the true un­
derlying physics. Furthermore, phenomena such as the conversion of cloud droplets
to rain drops, quantified by (1 .3) , occur on much smaller scales than the numerical
grids employed when solving (1 . 1) . The resulting model errors or discrepancies pro-
Chapter 1 . I ntroduction 3

duce biased or systematic uncertainties that are typically difficult to quantify using
a probabilistic framework.
Input Uncertainties. Parameters such as j5, m2, nr, and do in the phenomenolog­
ical representation (1 .3) for S1 are uncertain, as are initial conditions for the evo­
lution equations (1 . 1) . These comprise input uncertainties that are often amenable
to probabilistic analysis.
Numerical Errors and Uncertainties. As detailed in Section 2 . 1 . 3 , local mete­
orological models are numerically approximated on spatial grids having horizontal
spacing on the order of 5 km and vertical spacing of approximately 200 m. This
introduces numerical discretization or approximation errors . Furthermore, it intro­
duces systematic uncertainties due to the fact that parameterized processes , such
as aerosol-induced cloud formation and atmospheric turbulence, occur on much
smaller, subgrid, scales .
Measurement Errors and Uncertainties. It is noted in Section 2 . 1 .4 that me­
teorological data is comprised of earth-surface and atmospheric measurements. The
latter is obtained from weather balloons, weather satellites, and aircraft . There are
two primary sources of uncertainty: limited accuracy of the sensors and uncertainty
associated with the time and location of measurements.
Predictions for Weather Forecasts. Uncertainty quantification for weather
forecasting takes place in two steps. In the first , data assimilation-often performed
in a Bayesian framework-is used to determine values and quantify uncertainties
for inputs such as initial conditions and phenomenological parameters. This is the
model calibration step . In the second step , the calibrated models are run forward
in time to provide forecasts with quantified uncertainties.
To accommodate the effects of input , model, numerical, and measurement
uncertainties, ensemble forecasts are computed by running multiple simulations
from individual or multiple models with differing initial conditions or parameter
values drawn from probability densities constructed during the calibration phase.
Using the ensemble predictions, one computes statistical quantities of interest , such
as the average temperature, relative humidity, or proj ected rain amounts.
Although uncertainties associated with quantities of interest are computed
during the ensemble computations, they typically are not reported in forecasts.
One exception is the prediction of large storms such as cyclones, tropical storms,
or hurricanes. This is illustrated in Figure 1 .2 by the predicted traj ectory and
uncertainty cones for the post-tropical cyclone Debby.
The following definitions quantify terms, introduced in Example 1 . 1 , that play
a fundamental role throughout the text .

Definition 1 . 2 (Inputs) . The term inputs is used to designate parameters, ini­

tial conditions, boundary conditions, or exogenous forces that exhibit uncertainties
which must be determined and propagated through models to construct predic­
tions with quantified uncertainties . Known or fixed coefficients, independent and
dependent variables, and control signals do not constitute inputs as defined here.
4 Chapter 1 . I ntrod uction

Pottintial Track Are�:

C::::::...o.y1-3COoy4'5

Figure 1.2. NOAA image of the trajectory and cone of uncertainty for the post­
tropical cyclone Debby.

Definition 1 . 3 (Quantity of Interest (Qol} ) . The QoI designates the output

of a simulation model or experiment that provides information necessary to make
conclusions or decisions about the process. In many contexts, we employ the terms
model response or model output in a synonymous manner. As illustrated in Exam­
ple 1 . 1 , we often consider statistical or probabilistic QoI to accommodate uncertain­
ties intrinsic to the modeled process. Examples of probabilistic QoI include average
temperatures, expected precipitation in a viewing region, average performance of
a nuclear reactor , and expected impact of drilling in an environmentally fragile re­
gion. Chapter 2 details QoI for weather, climate, groundwater, nuclear reactor, and
systems biology models.

Definition 1 . 4 (Verification). Verification refers to the process of quantifying the

accuracy of simulation codes used to implement mathematical models.
Definition 1 . 5 (Validation). Validation describes the process of determining the
accuracy with which mathematical models quantify the physical processes of inter­
est. This necessarily involves the simulation code used to implement the model and
experimental data from the process.

1.1 N atu re of U n certa i nties and Errors

In Example 1 . 1 , we illustrated that uncertainties and errors arise in the modeling,
simulation, and experimental components of applications. We detail the sources
and nature of these uncertainties in this section.

1.1.1 Experimental U ncertainties and Lim itations

Experimental results are believed by everyone, except for the person who ran the
experiment, quoted by Max Gunzburger, Florida State University; original source
unknown.
There are two fundamental sources of uncertainty and errors in experiments:
limited or incomplete data and limited accuracy or resolution of sensors. The first
1 . 1 . N atu re of Uncertainties a nd Errors 5

can be broadly interpreted as due to the fact that experiments are often surrogates
or provide only partial measurements when we cannot fully observe the underlying
application due to physical infeasibility or expense. Examples include the following.
• The meteorological data noted in Example 1 . 1 is obtained at discrete locations
that can be uncertain.
• Wind tunnel tests are used as surrogates for flight tests. The limitations of
using a scale model in lieu of a full aircraft must be incorporated in designs.
• Pharmaceutical and disease treatment strategies are often too dangerous or
expensive for human tests or large segments of the population. For example,
HIV trials are conducted with test subj ects rather than a full population.
• Climate scenarios cannot be experimentally tested at the planet scale. Instead,
forcing mechanisms such as those due to volcanic eruptions are tested using
measurements such as the 1991 Mount Pinatubo data-see Section 2.2.1 .
• In materials experiments, difficulties obtaining nano- and molecular-level time
and spatial scale data limit multiscale testing of novel material designs.
• Subsurface hydrology data is very limited due to the expense and infeasibility
of drilling large numbers of wells. As a result , there is significant uncertainty
regarding specific subsurface structures-see Section 2.3.
• The harsh radioactive, thermal, and chemical environments i n a nuclear reac­
tor core limit the availability of measurements for performance improvement ,
nondestructive evaluation, and safety regulation-see Section 2.4.
Whereas several of these examples illustrate limited rather than statistically uncer­
tain data, the associated deficiencies increase the reliance on models and augment
model uncertainties due to lack of data.
The limited accuracy or resolution of sensors contributes statistical uncer­
tainties that can produce parameter uncertainties during model calibration. These
sensor uncertainties can occasionally be specified by sensor manufacturers and are
often amenable to statistical analysis.

1.1.2 M o del and I n put U n certainties

Essentially, all models are wrong, but some are useful, George E.P. Box, page 424
of [38] .
Model uncertainties arise from two sources: model errors or discrepancies and
input uncertainties due to uncertain parameters, forcing functions, and initial and
boundary conditions.

Model Errors and Discrepancies

Modeling errors or discrepancies are due to approximate or imprecise rep­
resentation of underlying physical, biological, economic, or social processes. The
following examples illustrate sources of model error .
6 Chapter 1 . I ntrod uction

• Numerous components of weather and climate models-e .g. , aerosol-induced

cloud formation, greenhouse gas processes, turbulence-occur on scales that
are much smaller than the numerical grids used to solve the atmospheric
equations of physics. Moreover, many of these processes represent highly
complex physics that is only partially understood. Subsequently, the processes
are represented by phenomenological models with nonphysical parameters .
Both the model form and the parameters exhibit uncertainty.
• Many biological applications are coupled, complex, highly nonlinear, and
driven by poorly understood or stochastic processes. Moreover, they presently
do not admit an encompassing set of governing relations analogous to those
in physics. Hence, associated models are subj ect to significant uncertainty.
• The predicted production of greenhouse gases is highly dependent on proj ected
economic and technological growth of nations. These processes are highly
uncertain and difficult to model. This uncertainty is addressed in climate
models by considering various economic and technology growth scenarios.

The quantification of model errors is typically problem-dependent since it neces­

sitates obtaining additional knowledge about the problem. The development of
general statistical techniques, such as construction of Gaussian processes for model
discrepancies, constitutes a current research topic.

Input Uncertainties
All models contain parameters that must be specified before the model can
be used to represent or predict the behavior of the process. Moreover, differential
equation models have initial or boundary conditions that must be designated in
addition to potential exogenous forces. As noted in Definition 1 .2 , these compo­
nents introduce input uncertainties that must be quantified and propagated through
models.
• As indicated in Example 1 . 1 and shown in Sections 2.1-2.5 , the phenomeno­
logical models used to represent processes such as turbulence in weather, cli­
mate, and nuclear reactor models have nonphysical parameters whose values
and uncertainties must be determined using measured data.
• It is shown in Section 2.2 that forcing and feedback mechanisms in climate
models serve as boundary inputs. These parameterized phenomenological
relations introduce both model and parameter uncertainties .

The process of estimating model inputs based on measured data is typically

termed model calibration or simply parameter estimation if inputs consist solely of
parameters . The estimation of input uncertainties for a model using measured data
is often referred to as inverse uncertainty quantification.

Coupled Systems

The quantification of model discrepancies and input uncertainties is typically

challenging for individual components of a system model. The difficulty grows
1 . 1 . N atu re of Uncertainties a nd Errors 7

substantially as components are bidirectionally or tightly coupled to quantify mul­

tiscale or multiphysics phenomena. For such problems, the tight coupling generally
prohibits a unidirectional propagation of discrepancies or uncertainties and instead
necessitates a more global accommodation of these terms . Furthermore , the nature
of inputs or parameters can change if they are actually states at another level in
the process. The applications discussed in Sections 2.1-2.5 yield system models
that are coupled to varying degrees, and the development of techniques to quantify
model discrepancies and input uncertainties and construct prediction intervals for
quantities of interest in such applications constitutes an active research area.

1.1.3 N umerica l E rrors and U n certainties

Computational results are believed by no one, except the person who wrote the code,
quoted by Max Gunzburger , Florida State University; original source unknown.
The characterization and regulation of numerical or algorithm errors is a topic
in numerical analysis , and this is the least uncertain component of predictive sci­
ences. Numerical errors and uncertainties include the following.
• Roundoff, discretization, or approximation errors.
• Bugs or coding errors .
• Bit-flipping, hardware failures, and uncertainty associated with future exa­
scale and quantum computing.

Additionally, the 5-50 km grids required for numerical solution of field equa­
tions, in applications such as the weather model outlined in Example 1 . 1 , are
much larger than the scale of physics being modeled ( e.g. , turbulence or aerosol­
induced cloud formation ) . This numerical requirement introduces uncertainty in
phenomenological model relations.

1.1.4 Types of U ncertainty

The previous examples illustrate that modeling, experimental and numerical un­
certainties, and errors can have various forms . The following definitions categorize
uncertainties based on the degree to which they are inherent to the application or
reflect lack of knowledge.

Definition 1 . 6 (Aleatoric Uncertainty) . Also known as statistical, stochastic,

or irreducible uncertainty, this is uncertainty inherent to a problem or experiment
that in principle cannot be reduced by additional physical or experimental knowl­
edge . Examples include uncertainties associated with nonphysical model parame­
ters, subgrid atmospheric conditions such as wind gusts, subsurface microbe levels
between wells, and initial conditions for weather models. Aleatoric uncertainties
are typically unbiased and are often naturally defined in a probabilistic framework.
Hence additional experiments or knowledge serve to better categorize the uncer­
tainty.
8 Chapter 1 . I ntrod uction

Definition 1. 7 (Epistemic or Systematic Uncertainties). Epistemic uncer­

tainties are those that are due to simplifying model assumptions, missing physics,
or basic lack of knowledge . Many of the previously mentioned modeling errors-e.g.,
phenomenological expressions for input or closure relations-and numerical errors
are epistemic in nature. These uncertainties are often biased, and they are typically
less naturally defined in a probabilistic framework.
If numerical errors are negligible, epistemic uncertainties are often termed
model errors, model discrepancies, or model inadequacies. As detailed in Chapter 1 2 ,
the quantification of these components fo r extrapolatory predictions outside the
calibration domain constitutes an active research area.
The distinction between aleatoric and epistemic uncertainties is not always
clear since lack of knowledge is relative and depends on current theory and ex­
perimental capabilities. One goal of uncertainty quantification is to reformulate
epistemic uncertainties as aleatoric uncertainties where probabilistic analysis is ap­
plicable.

1.2 P red ictive Estimation

A broad obj ective of predictive science i s t o use models, simulation codes , and ex­
periments to predict system responses with quantified and reduced uncertainties.
The probabilistic quantification of predicted experimental and computational out­
comes with identified and quantified uncertainties is sometimes termed predictive
estimation. As detailed in [52] , predictive estimation is comprised of three compo­
nents.
• Model Calibration: This involves the assimilation or integration of data to
quantify and update input uncertainties associated with parameters , forcing
functions, initial conditions, or boundary conditions.
• Model Prediction: Here one computes the response, or Qol, along with statis­
tics , error bounds, or the probability density function (pdf} for the Qol.
• Estimation of the Validation Regime: This entails estimating contours of con­
stant probability for the Qol to establish a domain for predictions with spec­
ified uncertainties.

To illustrate the predictive estimation process, we consider the mathematical

model
Y = f( x, q) (1 .4}
and statistical model
Yi= J( xi, q ) + 8( xi) + ci, ( 1 .5}
where q = [ q1, . . . , qp ] is a vector of inputs, y E R11 is the model output or Qol, and
x denotes independent variables such as time t or space x. In the statistical model,
Yi and ci are random variables representing measurements and measurement errors
and 8( xi) denotes biases due to epistemic or systematic uncertainties, as defined in
1.2. Pred ictive Estimation 9

Definition 1 .7. If numerical errors are negligible, 8(Xi ) quantifies the model error or
model discrepancy.
In the following description, we indicate associated chapters in parentheses.
The complete predictive estimation process is depicted in Figure 1 .3.

Predictive Estimation Process

• Input Representation (Chapter 5): One must first represent inputs in a prob­
abilistic framework that facilitates model calibration and uncertainty propa­
gation. This includes the construction of finite-dimensional representations
for distributed or spatially varying parameters or initial conditions.
Parameter Selection (Chapter 6): For many applications, the parameter di­
mension can be very large, e.g . , p = 100 - 10 6 , which prohibits direct model
•

calibration. Furthermore, parameters in many models are unidentifiable in the

sense that they cannot be uniquely identified from the measured response . For
such applications, one must employ parameter selection techniques to isolate
a subset of most influential parameters q E JR'.P, p < p, that are employed for
subsequent analysis . This relies on the contents of Chapters 14 and 15.
Local Sensitivity Analysis (Chapter 14 ) : Local sensitivity analysis fo­
cuses on the variability of the response as inputs are varied about a
nominal value as quantified by the derivative f,t (q * ) .
Global Sensitivity Analysis (Chapter 15 ) : Global sensitivity analysis
quantifies how uncertainty in model responses can be apportioned to

Input Representation Local Sensitivity Analysis Global Sensitivity Analysis

(Chapter 5) (Chapter 14) (Chapter 15)

!
Parameter Selection
! Surrogate Models
- -

l
(Chapter 6) Model Discrepancy (Chapter 13)
(Chapter 12)

Stochastic Spectral Methods

-
(Chapters 7 and 8)
Sparse Grids Model Calibration

!
(Chapter 11) (Chapter 10)

t
! Sparse Grids
(Chapter 11)
Uncertainty Propagation
(Chapter 9)

Figure 1 . 3 . Components of the predictive estimation process and relevant chap­

ters. Model calibration and uncertainty propagation are the driving objectives. The
remaining topics are required to achieve these objectives for large-scale applications.
10 Chapter 1 . I ntroduction

uncertainties in inputs. We consider variance-based Sobol methods and

screening algorithms based on approximation of the local index M, (q * )
evaluated at random values q * in the admissible parameter space .

• Surrogate Models (Chapter 13) : For the large-scale applications detailed in

Chapter 2 , the complexity of simulation models generally precludes their di­
rect use for model calibration and uncertainty quantification. This is ad­
dressed by constructing surrogate models y = f(Xi, ij_) that encapsulate the
primary behavior of the modeled process but are sufficiently efficient for model
calibration, uncertainty propagation, and control implementation.
Stochastic Spectral Methods (Chapter 10) : Stochastic Galerkin, colloca­
tion, or discrete proj ection methods provide one option for computing
surrogate models.
Sparse Grid Methods (Chapter 1 1 ) : Sparse grid methods are required
to implement stochastic spectral methods and for direct implementation
of Bayesian model calibration methods when parameter dimensions are
moderate; e.g . , p = 8 - 50 .

• Model Discrepancy (Chapter 12) : For applications that exhibit epistemic or

systematic uncertainties due to model discrepancy or numerical errors, one
must quantify the bias term 8(Xi) in (1 .5) using physical, mathematical, or
statistical analysis.
• Model Calibration (Chapters 7 and 8): Frequentist or Bayesian techniques
are used to quantify the uncertainties associated with inputs q based on mea­
sured data y . In the Bayesian framework, inputs are formulated as random
variables with associated pdf. For moderate input dimensions p, the sparse
grid techniques of Chapter 1 1 can be used to directly evaluate Bayes' relation
to avoid sampling-based Metropolis algorithms.
• Uncertainty Propagation (Chapter 9): The final obj ective of predictive esti­
mation is to propagate input uncertainties through models to construct pre­
diction intervals or pdf for Qol. For linearly parameterized problems, one
can establish analytic relations for statistical moments. For mildly nonlinear
problems , linearization using Taylor expansions can achieve the same objec­
tive. More generally, one can employ stochastic polynomial or sampling-based
methods, such as those employed for model calibration, to compute moments
or construct prediction intervals for Qol.

For large-scale applications, one must typically employ model selection tech­
niques, address model discrepancies, and construct surrogate models before model
calibration and uncertainty propagation can be achieved. For applications with
identifiable parameter sets, negligible epistemic uncertainties, and highly efficient
simulation codes, however, one can focus immediately on the model calibration
and uncertainty propagation techniques detailed in Chapters 7 , 8 , and 9. We refer
readers to [52] for details regarding the estimation of the validation regime.
Cha pte r 2

Large-Scale Applications

In this chapter, we illustrate five applications where model predictions with quanti­
fied uncertainties are critical for understanding and predicting scientific phenomena
and making informed decisions and designs based on these predictions. These ap­
plications are weather models, climate models, subsurface hydrology and geology
models, nuclear reactor designs, and models for biological phenomena.

2.1 Weather M o d e ls
If asked to list areas of science in which uncertainty quantification is critical for pre­
dictive estimation, most people would likely include weather forecasting. Moreover,
some would do so with a sense of derision while noting the occasional inaccuracy
of forecasts. This negativity is due in part to the fact that the role of uncertainty
in scientific predictions receives little attention in secondary and undergraduate
curricula, which results in a potentially false sense of security regarding scientific
predictions. This leads to a poor understanding of weather forecasts-e.g. , sur­
veys reveal that many people interpret a 50% chance of rain as meaning that half
the viewing region will receive precipitation-and a general distrust of the science
associated with weather forecasting. When one considers the complexity of the
underlying phenomena and associated models, the inherently chaotic or unstable
nature of the prediction process, and the uncertainties associated with the models,
simulation codes, and data, the accuracy of forecasts with quantified uncertainties
represents a major tour de force of physical modeling and scientific computation.
To motivate the complexity of modeled phenomena, one need only consider
factors required to predict temperatures, precipitation, and winds. As illustrated in
Figure 2.1 , temperature in the atmosphere depends on the absorption and emission
of radiation, latent heat release, advection due to winds, and convective heating or
cooling at the earth's surface . The temperature field thus depends on the horizontal
and vertical distribution of small particulates and liquid droplets (excluding cloud
droplets and precipitation) that are collectively termed aerosols. Additionally, it
depends on wind patterns that produce warm and cold air advection, and the surface

11
12 Cha pter 2. Large-Scale Applications

1H o n izo nta:1 G r1i d

(Latitude-Longitude)

Vertical G ridl
(Height or Pressure)

A.'rMOSF>HEl'lE

±
Figure 2 . 1 . Physical processes that must be incorporated in weather and climate
models and the associated 3-D grid. Image courtesy of NOAA.

topography and heat profile. Furthermore , phase transitions between liquid, solid,
and vapor phases for atmospheric moisture add or remove latent heat from the
atmosphere, thus requiring that these effects be coupled with temperature models.
Aerosols are associated with atmospheric conditions , such as dust or smog
levels, and are catalysts for phenomena such as cloud formation since they serve as
condensation nuclei around which cloud droplets form. The smallest aerosols have
radii on the order of 0.1 µm and are typically attributed to chemical conversion of
sulfate gases to liquids or solids. First principles models for these processes thus
occur on relatively small space and time scales and require quantification of the
associated chemical processes. Larger aerosols include wind-driven dust particles,
particulates from volcanic reactions, and combustion byproducts.
Changes in temperature produce pressure gradients that in turn generate air
movement and wind. Near the earth's surface, wind flow is significantly influenced
2 . 1 . Weather Models 13

b y the terrain and surface conditions such a s temperature. This produces highly
complex frictional and boundary layer effects. At higher altitudes, the situation is
slightly less complicated but is still fully coupled with all of the previously mentioned
phenomena. Hence high altitude flow patterns, such as the jet stream, tend to be
somewhat more stable than surface wind patterns, but they still exhibit highly
turbulent dynamics , instabilities, and bifurcations due to the highly nonlinear and
complex coupling with atmospheric conditions.

2.1.1 Conservation Relations

The physical components of meteorological or weather models are based on conser­
vation of mass, momentum, and energy in combination with conservation of water
phases and constituent chemical spaces in aerosol models.
As detailed in [193, 213] , conservation of energy using the first law of thermo­
dynamics yields the partial differential equation
8T p \l v = - \!
pcv at + · · F + \l · (k\JT) + p<j(T, p, p), (2.1)
where T and v are the temperature and velocity at a point (x, y, z) E R 3 and p , cv, p,
and k respectively denote the density, specific heat at constant volume, pressure,
and thermal conductivity for an infinitesimal atmospheric volume V . Furthermore,
F is the net radiative flux and <j(T, p, p) is the rate of internal heating or cooling
associated with processes such as latent heat release due to phase changes in atmo­
spheric moisture . The pressure , temperature , and density are typically related by
the ideal gas law
p = pRT, (2.2)
where R is the specific gas constant for air.
Atmospheric flow dynamics, such as the j et stream and local wind patterns,
are quantified using conservation of mass and momentum , which yields the relations
op
at + \l (pv) = 0
- · (2.3)
and
av = -v · \Iv - -\Ip
1 - gk - 2n x v. A

(2.4)
-
at P
Here g is the force due to gravity and -20 x v is the Coriolis force due to the earth's
rotation. The expansion of the nonlinear term v · \Iv and individual equations in
spherical coordinates can be found in [196] .
It is shown in [193] that the concentration of water, in the solid, liquid, and
gaseous phases, and aerosols can also be quantified using conservation relations.
If we let m 1 , m2 , and m3 denote the mass of the solid, liquid, and gaseous water
phases relative to the mass of air in the same volume , the concentration of each
phase can be represented by the relation

a/ = -v
om ·
· \lm j + Sm; (T, m3 ' XJ i p) , j = 1, 2, 3. (2.5)
14 Cha pter 2. Large-Scale Applications

The source or sink terms Sm; quantify the processes that govern phase transitions.
These can be extremely complex due to their dependence on aerosol concentrations
Xj and coupled atmospheric processes. The physics associated with the terms Sm;
is often difficult or impossible to establish on the grid scales illustrated in Figure 2.1 ,
thus necessitating phenomenological parameterizations.
Aerosol concentrations are quantified in a similar manner . If we consider J
species with concentrations Xj , conservation principles yield the relations

a:
ox . = -v . "V (2.6)
xj + SXi (T, Xj , p) ' j = 1 , . . . ' J,
where Sx; incorporate changes in state, chemical reactions , and sedimentation. Like
Sm; , parameterizations are typically required to construct these terms.
The set of equations
op "il
at + (pv) = o ,
·

8v 1 - gk - 2n x v
at = -v '\lv - -'\lp
A

- ·
P
'
8T p'\l
pcv at + v = -'\l F + '\l (k'\lT) + p<j(T, p, p),
· · ·
(2.7)
p = pRT,
om/· = -v '\lm S (T, m , , = 1, 2, 3
a j + m;· j , Xj p) j ,
ox .
a! = -v xj + Sx; (T, xj , p) , j = 1 , . . . , J,
"V ·

are often referred to as the equations of atmospheric physics. If one neglects the
species relations for mj and Xj and employs hydrostatic approximations to the mo­
mentum equation, one obtains what are often termed the primitive equations for
Eulerian fluid motion. Various meteorological and climate models are constructed
by employing simplified forms of these relations in combination with parameteriza­
tions for F, <j, Sm; , and Sxr
We note that meteorological models for tropical dynamics are significantly
more complex than those for midlatitude or extratropical regions, e.g. , poleward
from about 30° latitude. In the middle latitudes, the primary source of energy
driving wind patterns is temperature-induced pressure gradients in balance with
Coriolis forces, and latent heat release and radiative heating are secondary con­
tributors to atmospheric dynamics. Here geostrophic or quasi-geostrophic theory,
based on a balance of pressure gradients and Coriolis forces, provides simplified
momentum relations for meteorological models.
In the tropics, however, temperature gradients are smaller and latent heat re­
lease associated with convective cloud systems is a more significant source of energy.
Moreover , Coriolis forces are also smaller and there is a more significant coupling
between atmospheric and ocean temperatures. Meteorological and regional climate
models for these regions must thus incorporate the interaction between cumulus
2 . 1 . Weather Models 15

convection and mesoscale and large-scale circulations as well as equatorial wave

dynamics and air-ocean interactions . Resulting weather phenomena include mon­
soons , the trade winds , hurricanes, and El Niiio . Readers are referred to Chapter 1 1
of [1 1 3] and the included references for more details regarding tropical weather phe­
nomena and associated modeling issues.

2.1.2 P henomenological Models

To numerically solve the coupled relations (2.7) , expressions for the net radiative flux
F, rate <j (T, p, p) due to latent heat release , and source terms Sm; (T, mj , Xj , p) and
Bx; (T, Xj , p) must be specified. For phenomena such as radiation transfer , modeling
relations can be based on physical principles. However , the terms Sm; and Bx; are
highly complex and occur on scales that are much smaller than computational grids.
This necessitates the construction of phenomenological models having nonphysical
parameters that are determined through model calibration via data assimilation or
from independent experiments.
To illustrate, it is established in [1 93] that if m 2 represents the concentration
of water in liquid phase, then a simplified form of Sm2 is

(2.8)
where 81 represents the conversion of cloud droplets to form raindrops, 82 represents
the accretion of cloud water by raindrops, 83 represents the melting of snow or ice
to rain, and 84 quantifies the evaporation of rain. An accepted relation for 81 is

81 = j5 (m 2 - [
m;) 2 i .2 X 1 0 -4 + ( 1 . 569 X 10-
12 nr
*)) ]-l ,
do ( m 2 - m 2 (2.9)
where j5, m 2 , nr, and do are constants that must be specified.
Parameterized phenomenological models are required for a range of atmo­
spheric and terrestrial phenomena, including aerosol-induced cloud formulations,
reactions that produce aerosols, turbulence at various levels in the atmosphere, and
drag and surface effects due to mountains and attributes of the terrain. Quantifi­
cation of uncertainties in parameters and phenomenological models is necessary for
quantifying uncertainties in predictions of Qol.

2.1.3 S i m u lation M o dels

Essentially every weather person refers to predictions resulting from computer mod­
els. Care must be exercised when interpreting this phrase since it really implies
simulations obtained using discretized physical models.
All approximation techniques yield atmospheric state values on a 3- D grid
such as that depicted in Figure 2.1 . The horizontal and vertical grid spacing is
determined by a number of factors, including the spatial scales of modeled physics
and available computing resources . Local meteorological models employ horizontal
grids on the order of 5 km with vertical spacing of approximately 200 m. Global
16 Cha pter 2. Large-Scale Applications

meteorological and climate models necessarily employ larger horizontal grids that
are on the order of 50 - 100 km .
The gridsizes required for weather and climate models constitute a signifi­
cant source of uncertainty since parameterized processes such as aerosol-induced
cloud formation, latent heat generation due to cloud formation, and atmospheric
turbulence occur on much smaller, subgrid, scales.
Various discretization techniques can be employed to approximate the rela­
tions (2.7) . Finite difference techniques are typically employed to discretize vertical
spatial derivatives, whereas finite difference, or occasionally finite element , tech­
niques are employed for the horizontal components of regional and some global
models. Other global models exploit periodicity by employing spectral approxima­
tion methods.
Semidiscretization in space yields very large , coupled, vector-valued systems
of differential equations that must be numerically integrated forward in time to
provide predictions. The use of explicit methods introduces stringent limits on
temporal stepsizes due to stability or Courant-Friedrichs-Lewy (CFL) conditions
that can limit the utility of algorithms. For example , to quantify gravity waves in
jet streams that have maximum velocities of 200 m/s, horizontal grids of 100 km
yield maximum time steps of approximately 8 minutes . To address this, numerical
weather prediction (NWP) models commonly employ semi-Lagrangian integration
techniques that approximate the path of air parcels (Lagrangian perspective) while
predicting values on a fixed (Eulerian) grid. The advantage is that large temporal
stepsizes can be employed without loss of stability. Moreover, semi-Lagrangian
schemes can be constructed to ensure that species concentrations are conserved
during advection.
The length of simulations varies for differing models. The UK Met Office Uni­
fied Model is run six days into the future, whereas the European Centre for Medium­
Range Weather Forecasts' (ECMWF) Integrated Forecast System provides 10 day
predictions. The decline of accuracy for longer predictions is further discussed in
the context of uncertainty quantification for weather forecasts.

2.1.4 M o del Calibratio n , Data , and Data Assim ilation

Meteorological models used for weather forecasting are essentially initial value prob­
lems. Based on measurements, one attempts to determine initial conditions and
parameters so that models match present atmospheric conditions, as closely as pos­
sible , as illustrated in Figure 2.2. Data assimilation techniques are used to calibrate
models by determining inputs (initial conditions and parameters) based on a wide
range of measurements. Models are then numerically integrated forward in time
to provide forecasts. Due to the nonlinear nature of the models, they are chaotic
in the sense that uncertainties in initial conditions grow in an unstable manner,
thus significantly diminishing the accuracy of forecasts for increasingly long future
periods. This highlights the necessity of considering ensembles and statistical Qol,
such as average temperatures, and quantifying the uncertainty in predictions.
A critical component of model calibration via data assimilation is globally dis­
tributed and frequently obtained earth-surface and atmospheric data. In the United
2 . 1 . Weather Models 17

-"=' A s s i m i lation Period Forecast Period

:
""E

�-
a

H--j
Cl)
:J

J5
Q)
·
' I - -} ' .1 · ·
i::
Cl)

Q) I I

0
(/)
.0 I I I

Figure 2.2. Assimilation period to estimate initial conditions and parameters so

the model best fits data with measurement uncertainties. Forecast with quantified
uncertainties.
States, there are approximately 1000 surface measurement locations, whereas the
World Meteorological Organization ( WMO ) maintains approximately 10,000 land
sites worldwide. Surface measurements on the ocean are obtained by moored and
drifting buoys as well as ships on an array of routes. Atmospheric conditions are
provided by radiosondes in weather balloons that rise into the stratosphere, weather
satellites, commercial aircraft on prescribed routes, and reconnaissance aircraft that
can be sent to regions yielding high impact data, e.g. , regions where there is signif­
icant uncertainty or storms such as hurricanes.
In combination, this yields a fairly rich data environment . From the perspec­
tive of model construction, prediction, and uncertainty quantification, however,
three properties of the data are important: it is measured on highly irregular grids
and it is frequently distributed in time, the observations are not direct measure­
ments of state variables , and there is uncertainty associated with all data. The
limited accuracy of measuring devices constitutes one source of uncertainty. Sec­
ond, several are moving, so there are varying degrees of uncertainty associated with
the position and time of measurements.
To illustrate issues pertaining to data assimilation, we assume a grid of size
432 x 320 x 50 with a minimum of 8 variables comprising the wind speed, temper­
ature, pressure , and moisture phases. This yields 5.53 x 10 7 states x. Phenomeno­
logical components of the models can easily require in excess of 50 parameters in
addition to initial conditions. Since observations are on the order of thousands , the
problem is highly underdetermined.
To construct a functional to be minimized, we let Yi denote the vector of ob­
servations at times ti where we assume n measurement times. States x are mapped
to observations by the operator H so that y = H (x ) . Prior information x 8 , ob­
tained from a previous model forecast , is often termed the background. The error
and background error covariance matrices are denoted by Hi and B. Predictions
X i+ I provided by the NWP model, based on current states xi , can be represented
as
(2.10)
where Mi+ l , i represents the discretization of the nonlinear model from time ti to
time ti+ I
·
18 Cha pter 2. Large-Scale Applications

The data assimilation algorithm 4D-VAR is employed at several weather pre­

diction centers, including the ECMWF and the UK Met Office as well as stations
in Japan and Canada. In this algorithm, one minimizes the functional
n

J (x o ) = (xo - x � f B - 1 (xo - x � ) + L (Yi - H ( xi )f Ri 1 (Yi - H ( xi )) (2.11)

i =O
subject to the model dynamics

Xi = Mi,o (x o ) .
A primary difference between 4D-VAR and the previous algorithm 3D-VAR is the
use of adj oints to incorporate the times at which observations are made. Whereas
ensemble Kalman filters are being investigated for NWP models, 4D-VAR is presently
considered the state of the art .
We emphasize the fact that determination of initial conditions, often termed
the analysis, and model parameters that provide a best fit to subsequent observa­
tions is central to NWP. This is in contrast to climate models which are essentially
forced boundary value problems that are run until the transient effects of initial
conditions are mitigated.

2.1.5 Sou rces and N ature of U n certainties

There are four primary sources of uncertainty or errors in NWP models: model
limitations, input uncertainties due to initial conditions, boundary conditions or
parameters, numerical errors , and uncertainties in measurements. When combined
with the highly nonlinear nature of models, these produce response uncertainties
that grow as a function of time and must be quantified to provide a meaningful
context in which to interpret predictions or forecasts.
Although they are based on physical conservation laws, the continuity, mo­
mentum, temperature , and species equations (2. 7) are still approximations of the
underlying physical phenomena. Moreover, the phenomenological relations used to
model the flux and rate terms F and q and source terms Sm; and Bx; are ap­
proximate representations for highly complex and often only partially understood
physical phenomena such as turbulence , aerosol-induced cloud formation, and re­
sulting latent heat release as precipitation forms. These uncertainties are primarily
epistemic, as defined in Definition 1 .7.
As illustrated in (2.9) , the phenomenological components of the models con­
tain parameters that are often nonphysical and hence cannot be correlated with
measured data. Thus both their values and variability must be inferred through
model calibration or data assimilation techniques. Even physical parameters such
as the thermal conductivity k will exhibit some variability due to varying atmo­
spheric conditions and the fact that they may partially accommodate unmodeled
physics. In meteorological models, the initial atmospheric conditions constitute a
second critical source of uncertainty that must be inferred from later observations.
These uncertainties are aleatoric or stochastic in the sense defined in Definition 1 .6.
2 . 1 . Weather Models 19

The numerical discretization o f the models introduces uncertainty and errors in

two fundamental ways: the solution on a 3-D grid introduces significant uncertainty
for parameterizations of subgrid scale physics , and the approximation techniques
introduce discretization errors. The first is critical for a number of phenomena
such as turbulence and aerosol-induced cloud formation which occur at the level of
meters, whereas horizontal grids are on the order of tens to hundreds of kilometers.
This uncertainty is obviously related to the previously mentioned model uncertainty.
Discretization errors can often be asymptotically quantified using theory associated
with finite difference, finite element , spectral, or semi-Lagrangian techniques. These
are errors that are typically epistemic in nature .
Uncertainty in the data can arise from two sources : limited accuracy of mea­
surement devices and uncertainty associated with variability in the location and
time of moving sensors. The first is often categorized as aleatoric, whereas the
second is primarily epistemic.

2.1.6 Role of U n certainty Q uantification in Weather Forecasts

As illustrated in Figure 2.2, uncertainty quantification for weather forecasting takes
place in two steps. In the model calibration step , values and uncertainties associated
with inputs, such as initial conditions and parameters, are determined using data
assimilation techniques such as 4D-VAR. As detailed in Chapter 8, this is often
performed in a Bayesian framework, as evidenced by the priors x � in (2. 1 1 ) . In
the second step , the calibrated models are run forward in time to provide forecasts
with quantified uncertainties .
In the 1970 's and 1980's, it was recognized that forecasts obtained with a single
model simulation or realization had limited utility due to the inherent uncertainty
and chaos induced by the highly nonlinear initial value models . This led to the use
of ensemble forecasts, which have been standard since the 1 990's. In single model
approaches, ensemble forecasts are obtained by running multiple simulations from
an individual model with differing initial conditions or parameter values drawn from
probability densities constructed during the calibration phase. Using the ensemble
predictions, one constructs statistical QoI, such as the average temperature, relative
humidity, or proj ected rain amounts . Using ensemble forecasts, a 50% chance of
rain two days in the future means that given the present atmospheric conditions,
half of the simulations predict measurable rain amounts at some random point in
the specified area. Improved forecasts can be obtained using multimodel ensemble
forecasts in which ensemble predictions from multiple models are used to construct
QoI and uncertainty bounds. The reduction in variability of ensemble forecasts for
the hurricane Katrina, obtained with an additional 12 hours of data, is illustrated
in Figure 2.3. The hurricane position was very near the mode of the ensemble
predictions when it made landfall near New Orleans .
Whereas uncertainty bounds or probability densities for QoI are constructed
during the ensemble computations, they typically are not reported in forecasts.
One exception is the prediction of large storms such as cyclones, tropical storms, or
hurricanes . In these cases, forecasts usually include both the predicted traj ectory
and cones of uncertainty, as illustrated in Figure 2.4.
20 Cha pter 2. Large-Scale Applications
a b '
� -i-· · ·-· ·�· -·-
i ' '

.1..-···-- · -·· -···-·-

i i
40" t-i • •• 40"N ••

j
i i
35• r-1 -·· �
i
· -- -· • •. 35"N ·- L-··--····- -
i

--L-- -
l

.J.. t
· lO"N

I �-� 1
30·� ___ _ ·-- .

25"" ---+- 25-N +

· -·· ---------

1 r:xsw iatrw

Figure 2 . 3 . ECMWF ensemble forecasts made at ( a ) 00 UTC and ( b ) 12 UTC on

August 26, 2005 . Katrina made landfall near New Orleans at 1 2 UTC on August
29, 200 5 .

2.1.7 References
The topic of weather modeling and prediction is vast , and we have summarized only
some aspects of the discipline to illustrate the role of uncertainty quantification for
predictive estimation. Additional discussion regarding basic weather phenomena
can be found in [28 , 1 73] , whereas detailed derivations of the atmospheric physics re­
lations ( 2 . 7 ) and parameterized constitutive relations are provided in [123, 193, 2 1 3] .
These references also contain an overview of numerical techniques for simulating
weather models including semi-Lagrangian integration techniques. The implemen­
tation of data assimilation algorithms, such as 4D-VAR, is addressed in [129, 1 82] ,
and [129, 1 50] provide further details about ensemble forecasting. Finally, the reader
is referred to [263] for discussion regarding statistical issues and techniques associ­
ated with weather modeling and prediction.

���:�7c�1n ter Location Z:'.i.J N B 1 .:5 W

llAax SUstalnedWlnd 7� rnph
CurrentMovementV/at S mph
@ CurrentCenterlocation
• Forecast Center Positions

-;z:::;:k:;:::ii\:"
: -s;�-r==�==���:-1 -
H Sustainedwind > 73 mph
� Potential Day1-3 TrackArea

----,:
Tropical Storm Warning
Tropical StormWatch

Figure 2 . 4 . NOAA image of the trajectory and cone of uncertainty for Katrina.
2.2. Climate Models 21

2.2 C l i m ate M o d e ls
The impression many people have of climate models is that they are simply weather
models extended to much longer timescales of years, decades, centuries, or millennia.
They subsequently conclude that since two-week weather predictions are typically
poor, the accuracy of climate models must therefore be very suspect . Whereas
climate and weather models both require the quantification of coupled atmospheric,
land, ocean, and solar radiation processes, the vastly differing timescales dictate that
different physical mechanisms be emphasized in the two modeling regimes.
As detailed in Section 2 . 1 , weather models are highly nonlinear, initial value
problems whose chaotic nature necessitates ensemble computations to predict sta­
tistical Qol whose accuracy is typically limited to 10 days to two weeks. Hence they
are based on atmospheric conditions such as wind patterns, radiative , convective,
and latent heat driven temperature changes, and aerosol and moisture levels on
fairly short timescales . They can thus neglect seasonal effects, long-term anthro­
pogenic and natural forcing terms such as increased C0 2 levels and volcanic ash,
and influences such as deforestation. The emphasis is to use data assimilation tech­
niques such as 4D-VAR to determine initial atmospheric conditions and parameter
values so that models match recent and current conditions in a statistically accu­
rate manner. Ensemble predictions are then used to compute future Qol such as
temperature, precipitation levels, and storm tracks.
Climate models differ in the sense that they are required to accurately main­
tain a balance between absorbed solar energy and lower frequency infrared radia­
tion emitted to space-typically termed the earth's global energy balance or energy
budget-for decades up to centuries. As illustrated in Figure 2.5 , this energy budget
is influenced by numerous natural and anthropogenic factors including greenhouse
gas levels, seasonal effects, volcanic eruptions, deforestation, ocean dynamics, and
polar ice coverage. The timescales dictate that the transient effects due to initial
atmospheric and terrestrial conditions are essentially negligible . Instead, compu-

I ncoming Outgoing
Solar Radiation Longwave
342 wm- 2 Radiation
235 w m - 2

G reenhouse
Gases

,
350
Back
Radiation

390
I
324

Absorbed
by Surface

Figure 2 . 5 . Earth 's energy budget modified from [137] .

22 Cha pter 2. Large-Scale Applications

tation of the energy budget requires quantification of energy fluxes at the earth's
surface and sources and sinks in the atmosphere. Hence climate models exhibit
the dynamics of forced boundary value problems. One ramification is that chaotic
dynamics associated with weather models are largely mitigated in climate models.
As with weather models, one goal is to compute statistical QoI such as long-term
change in C0 2 or temperature levels, with quantified uncertainties. The scope in
climate models is much broader, however, since it additionally involves questions
such as the following.
• Is the planet getting warmer and are manmade processes the cause?
• Are atmospheric and/or oceanic circulation patterns or currents changing and,
if so , what will be the effect?
• Are the weather and climate becoming more extreme or variable and, if so ,
what are the ramifications?
Answers to these questions, with quantified uncertainties, are required to ad­
dress societal questions .
• Will climate changes lead to improved agriculture and food supplies or reduced
supplies due to widespread drought?
• Will sea-level changes threaten large civilization centers?
• Will changes in ozone levels significantly increase the incidence of cancer?
In subsequent discussion, we highlight ways in which uncertainty quantifica­
tion is critical for obtaining climate predictions with uncertainties quantified in a
manner that informs both scientists and policy makers .

2.2.1 Climate Forces and Feedback

The equations (2. 7) quantify the basic processes associated with atmospheric physics.
The difference in how these relations are used to construct weather and climate mod­
els lies in the simplifying assumptions and parameterized phenomenological models
used to quantify the net radiative flux F, rate of internal heating q, and source
terms Sm; and Bx; . We summarize here physical phenomena and sources of un­
certainty associated with the radiative fluxes since these are central to the global
energy balance that drives climate models.
We illustrate in Figure 2.5 factors that influence the balance between solar
energy absorbed by the atmosphere and earth and infrared radiation emitted to
space. As detailed in [137] , approximately 342 W/m 2 of solar radiation enters the
earth's atmosphere, where it is absorbed or reflected by the atmosphere, ground,
ocean, or surface ice. In the atmosphere , the primary absorbers are water vapor in
the troposphere and ozone in the stratosphere. Clouds are the primary mechanism
that scatter and reflect radiation. On the earth's surface, oceans and phenomena
such as deforestation and changing polar icecaps directly affect absorption and re­
flection rates. Visible light constitutes the primary wavelength reaching the earth's
surface with lesser amounts of UV and near infrared (heat) radiation.
2.2. Climate Models 23

Radiation due to heat at the earth's surface and reflected solar radiation both
contribute to the longer wavelength infrared radiation that is emitted into space.
The amount of infrared radiation is governed by cloud and water vapor levels along
with greenhouse gas concentrations.
Latent heat associated with cloud formation and precipitation constitutes the
primary nonradiative heat source in the atmosphere . A significant emphasis in
climate modeling focuses on quantifying the mean behavior and uncertainties asso­
ciated with these radiative and nonradiative processes.
Climate forcing mechanisms are defined as changes imposed on the earth's
energy balance that produce changes in the climate. These can include external
changes due to variability in the earth's orbit , fluctuations in solar radiation, and
comet or meteor impacts, or internal factors such as volcanic eruptions, deforesta­
tion, or changes in aerosol and C0 2 levels . We note that the internal forces can be
both natural and human-induced. Feedback processes are those in which changes
in the climate state serve as forces that produce further climate changes. Exam­
ples include changes in cloud cover due to aerosols, changes in surface reflection
due to melting polar icefields, and changing greenhouse gas levels due to increased
temperature-induced evaporation. We point out that all of the climate forcing and
feedback mechanisms are quantified using phenomenological models, often having
a large number of nonphysical parameters. This introduces significant uncertainty
that must be quantified in final climate model predictions.
Various natural and human-induced forcing mechanisms, along with uncer­
tainties and a qualitative indication of the level of scientific understanding, are
illustrated in Figure 2.6. We summarize next aspects of these mechanisms and
indicate how associated uncertainties influence climate models and predictions.

Nat ural External and Internal Forcing Mechanisms

Whereas we cannot control natural forces , their quantification is necessary to
determine their relative influence compared with anthropogenic forces that we can
control.
Solar Radiation
There are two primary mechanisms that affect the level of solar radiation
entering the earth's atmosphere: variability in the earth's orbit and fluctuations in
solar activity. The periodicity of sunspot activity with approximately 1 1-year cycles
has long been known, and the resulting variation in solar radiation is incorporated
in models. However , the effect of this variability on weather and climate is still
debated. Since the 1970's, radiation data has been measured by satellites and long­
term variability is inferred from carbon data in tree rings. Sources of uncertainty
include the parameterized models and measured data.
Volcanic Eruptions
The eruption of Krakatoa in 1 883 caused average global temperatures to drop
by approximately 1 ° C in the subsequent year and produced noticeable variability
in weather for approximately 5 years. The eruption of Mount Pinatubo in 1991
occurred after the advent of a global monitoring network, so it significantly advanced
24 Cha pter 2. Large-Scale Applications
Rad iative forci ng o f cl i m ate between 1 750 a n d 2005
Radiative iForc ng Terms

Long-lived
greenhouse gases
{
Ozone Tropospheric
VJ

·;;: Stratospheric
�
t5
al water vapou r

E S u rface albedo Black carbon

c:

on snow
al

::J
J:

Total
Aerosol
{ °'� """
Cloud albedo
effect

Linear contrails
VJ

iii
Q)

:s Solar i rrad iance

VJ
VJ

"1ii e
Q)
u

Total net
z c.

h u man activities

- 2 -1 0 2
Rad i ative Forcing (watts per squ are metre)

Figure 2 . 6 . Contribution and uncertainties associated with climate forcing mech­

anisms from the 2007 IPCC Report [228, Figure SPM.2] . Level of scientific under­
standing - Long-lived greenhouse gases: High; Ozone: Med; Surface albedo: Med­
Low; Total aerosol: Med-Low; Solar irradiance: Low.

our understanding of how large volcanic eruptions could force climate changes.
furthermore, it provided a unique opportunity to advance and test volcanic inputs
to climate models.
Volcanic forcing is due to the high levels of particulates and gases that are
introduced into the atmosphere. The manner in which these aerosols affect the
earth's energy balance is largely dependent upon the height to which they are
inj ected. Nonabsorbing aerosols reduce the amount of solar radiation that reaches
the earth's surface, whereas greenhouse effects are increased by aerosols that absorb
and emit in the infrared spectrum. The aerosols produced by Mount Pinatubo
reduced the solar energy reaching the earth's surface by 3-4 W/m 2 and cooled
global temperatures by approximately 0.5 ° C , as illustrated in Figure 2.7. The
climate changes due to volcanoes are relatively short-term unless they occur in
conjunction with other climate forces or feedbacks.
2.2. Climate Models 25
- - - Data
0.2
- Model

E
w
"'
c
rn
.c
0 -0.2
�
2
� -0.4
�
� -0.6

1991 1 99 2 1 993 1 994

Figure 2 . 7. Measured and predicted tropospheric temperatures surrounding the

June 15, 1991 Mount Pinatubo eruption; data from [106] .

Human-Induced Forcing Mechanisms

The quantification of anthropogenic climate forcing mechanisms is of funda­
mental importance since we can control them.
Greenhouse Gas Emissions
As illustrated in Figure 2.5, the greenhouse effect is the process in which
thermal radiation from the earth's surface is absorbed and reflected by greenhouse
gases such as water vapor, C0 2 , methane, ozone, and chlorofluorocarbons (CFCs)
such as the refrigerant Freon. Because much of the thermal energy is radiated
back to earth, increased levels of greenhouse gases produce an elevation in average
surface temperatures. The name is somewhat of a misnomer since warming is due
to changes in the absorption and reflection of radiated thermal energy rather than
restriction of convective heat loss, as is the case in a glass greenhouse .
Warming due to increased greenhouse gas levels constitutes one of the most
heavily studied areas of anthropogenic climate forcing, and it was stated in the 2007
IPCC Assessment Report that "It is very likely that greenhouse gas forcing has been
the dominant cause of the observed warming of globally averaged temperatures in
the last 50 years" [228] .
Because C0 2 is the most abundant greenhouse gas after water vapor and
because it is the byproduct of burning fossil fuels, its concentration levels have been
extensively studied and documented. C0 2 concentrations monitored at Mauna
Loa, Hawaii, since 1958 are plotted in Figure 2.8 (a) . The fluctuations have not
adequately been explained but are likely due to complicated feedback with short­
term atmospheric conditions.
Ice-core data has been used to establish C0 2 concentrations for the past
800 ,000 years at locations such as Antarctica and Greenland. Figure 2.8(b) illus­
trates the increase of C0 2 concentrations for the last 2000 years. Because measured
carbon isotopes can be used to establish the sources, the figure also illustrates that
the increase in C0 2 levels since the Industrial Revolution in the mid-nineteenth
century is largely influenced by C0 2 emission from fossil fuels.
Based on data published by [1 64] , the 200 1 IPCC published Figure 2.9, which
illustrated reconstructed temperatures from 1000 to 1980 , measured temperatures
from 1902 to 1999, 40-year smoothed averages, and two standard error limits [1 15] .
26 Cha pter 2. Large-Scale Applications

- Carbon Dioxide
E 360 - - - N itrous Oxide
.S 380 :a
a.

" .e, 340

>
" o"'
� 360 z_ 320
"E
Cl
·;;:

0
0
.s 300
c
340 "'

;;;
0
-" 8 280
0 320

1 9 60 1 970 1 980 1 990 2000 20 1 0 0 500 1 000 1 50 0 2000

Year Year

(a) (b)

Figure 2 . 8 . (a) Concentration of C02 measured at Mauna Loa; data from http ://
www.esrl.noaa.gov/ gmd/ ccgg/trends/ . (b) Concentrations of greenhouse gases over
the last 2000 years; data from 2007 IPCC report [228, Figure SPM. l] .

The reconstructed temperatures are based on tree ring data, coral, ice core mea­
surements, and historical records. C02 and temperature results of the type plotted
in Figures 2 . 8 and 2.9 have been cited by numerous scientists and policy makers
as evidence that increasing anthropogenic C02 levels are leading to unprecedented
temperature increases which in turn produce global warming.
While accepted by a large percentage of the scientific community, this topic is
still highly contentious and serious scientific arguments have been made questioning
the data, conclusions , future predictions and resulting policies, and even the fun­
damental viability of mathematical and statistical models as a means of providing

Med ieva l warm Period _____

L_
illl_
e_
lce
_A�g�
e____,..

0.5

0:.1.a ftom rt'M!irrnOM�trS: (tied) ar'td frorn lfe� rings.

corals, ice cotes and historical records (blue} .
-1 .0

1, 000 1 200 1 400 1 600 i aoo 2000

Year

Figure 2 . 9 . Proxy temperatures from 1000- 1980, measured temperatures from

1902- 1 999, 40-year smoothed averages, and two standard error limits; from 200 1
IPCC Report [1 1 5 , SPM, Figure l] .
2.2. Climate Models 27

meaningful climate predictions . While the sources of scientific disagreement are nu­
merous, nearly all can be distilled to differing assumptions employed in statistical
analysis or differences in how uncertainties are quantified (or neglected) in data,
models, and predicted Qol. We summarize only a few representative examples of
where uncertainty must be adequately quantified and incorporated to establish and
predict the degree of anthropogenic global warming.

• The ice core and tree ring analysis, used to establish past C0 2 and temper­
atures, exhibit varying degrees of uncertainty or inaccuracy. For example,
it is detailed in [49] that unaccommodated age-dependent variability in tree
ring widths, for broad-leaved species such as oaks, can flatten longer-term cli­
mate fluctuations and potentially produce misleading data. Termed the seg­
ment length curse, this can cause serious underestimation of climate variability
when the age of ancient timbers cannot be established through independent
means. The initial debate regarding the accuracy of the temperature "hockey
stick" plot centered on the statistical methods used to construct proxy past
temperatures based on statistically blended tree ring measurements.
• In many cases, error bars or uncertainties are omitted, thus misrepresenting
the validity of past values or future predictions. For example, the standard
error limits shown in Figure 2.9 are not plotted in many of the reports that
use this figure to argue that present temperatures are significantly higher
than during the past millennium. When viewed with uncertainty measures on
past proxy values, this conclusion is less dramatic. Moreover , when viewed
with error measurements, it is difficult to correlate Figure 2.9 with reported
past climate variations such as the Medieval Warm Period from roughly 950 -
1250, when Greenland was colonized by the Vikings , and the Little Ice Age
from 1400 - 1 70 0 . However, the global nature of these events has not been
established and is debated.
• Most greenhouse gas analysis focuses on the role of C0 2 and methane, since
they are fossil fuel combustion byproducts, with some analysis of ozone and
CFCs. However , water vapor is by far the most abundant greenhouse gas.
Significantly less effort has focused on accurate measurement of water va­
por over the last millennium. Moreover, it was noted in Section 3 . 1 that
because moisture phase transitions exhibit complex aerosol and temperature
dependencies and occur on subgrid scales , phenomenological models having
nonphysical parameters will introduce substantial uncertainties in models and
parameters .
• The specification of which data is utilized and which is neglected has intro­
duced uncertainty into both present interpretations and future predictions.
The conclusion drawn from Figures 2 . 8 and 2.9 is that present C0 2 levels
and temperatures are higher now than in the past millennium . However, ice
core measurements have demonstrated significantly higher C0 2 and temper­
ature changes over the last 800,000 years-e .g. , temperature variations of up
to 1 5 ° C [1 1 7 , 1 27] . In this context , variations of 5 ° C in the next century
fit well within past levels. It must be noted, however, that whereas the earth
28 Cha pter 2. Large-Scale Applications

has exhibited significantly more extreme temperatures in the past , it was also
not habitable by humans under those conditions.
• Climate model predictions must incorporate future C0 2 levels to predict
greenhouse effects. This requires models that predict the growth of nations
and their economies and technologies since these factors influence fossil fuel
usage. This is very difficult and prone to significant uncertainty. For example,
it is unlikely that models from 1990 would have accurately predicted current
C0 2 emissions in China. As detailed in Section 2.2.3 , this had led to predic­
tions based on various population, economic, and technological scenarios.

Further details regarding the role of uncertainty quantification in climate mod­

els will be provided in Section 2.2.3.
A erosol Emission
It was noted in Section 2.1 that aerosols critically affect cloud formation,
which in turn affects the global energy balance . It was further noted that aerosol
models are complicated by the fact that they contain complex forcing terms that
must be quantified using parameterized phenomenological models ; see, e.g. , (2.9) .
In climate models, predicted aerosol levels are further complicated by the fact that ,
like greenhouse gas levels, they rely on socioeconomic growth models. Associated
uncertainties are partially addressed by considering scenarios of the type discussed
in Section 2.2.3.
D eforestation and D esertification
The absorption of C0 2 through photosynthesis, and absorption and emission
of solar energy at the earth's surface , is strongly influenced by the nature of surface
vegetation. At present , approximately 29% of the earth's land surface is forested
and 1 1 % is arable. However, this is changing as forests are cut-especially in the
tropics such as the Amazon Basin in Brazil-croplands are urbanized, and sparse
vegetation is grazed in semiarid areas . As with greenhouse gases and aerosols,
models quantifying future deforestation and desertification are highly uncertain
since they depend on demographic and socioeconomic factors .

Climate Feedback Mechanisms

Forced climate changes are complemented by feedback mechanisms driven by
changes in climate conditions such as temperature , precipitation levels, or aerosol
concentrations.
Ice A lbedo Effects
The percentage of solar energy reflected by a substance or object , termed
the albedo, is an important factor in the global energy balance . On the earth's
surface, ice and open ocean water have very different albedos, and the former reflects
significantly more solar energy than the latter. The dependence of ice levels on
temperatures is a positive feedback mechanism since increasing temperatures reduce
the area covered by snow and ice , which subsequently produces further increase
in temperatures . Whereas ice and snow albedo effects constitute an important
2.2. Climate Models 29

component in the global energy balance , they are also a source of uncertainty in
climate models due to the complexity of associated physics and its highly nonlinear
coupling in global models.
Greenhouse Water Vapor Levels
We noted previously that water vapor constitutes the most abundant green­
house gas. Because increasing temperatures enhance evaporation that increases
greenhouse gas levels , the interdependency between greenhouse gases and temper­
atures constitutes a positive climate feedback mechanism in addition to being a
forced response .

2.2.2 Climate M o dels

The general framework used to construct climate models is similar to that described
in Section 2. 1.1 for weather models. Primitive equations derived from the equations
of atmospheric physics (2.7) are numerically solved on a 3-D grid of the form illus­
trated in Figure 2.1. As for weather models, the numerical grid introduces significant
uncertainty for processes such as aerosol-induced cloud formation and greenhouse
gas levels that must be quantified using parameterized phenomenological models on
subgrid scales.
It was noted in Section 2.2 that climate models differ fundamentally from
weather models in the sense that they are forced boundary value problems rather
than initial value problems. This introduces the problem of resolving complex
climate forces and feedbacks over very long time scales (e.g . , millennia) but signifi­
cantly reduces the chaotic behavior associated with uncertain initial conditions.
Figure 2.10 illustrates the evolution of climate models over the last 30 years.
Whereas ocean currents can be represented by primitive equations constructed us­
ing conservation of mass, momentum, and energy, the maj ority of processes such
as cloud microphysics , radiation, surface energy fluxes, turbulence, aerosol levels,
chemistry, sea ice formation, and dynamic vegetation levels are quantified using
parameterized phenomenological models. As with weather models, values and un-

Mid-1 970s Early 1 990s Late 1 990s Present

Atmosphere Atmosphere Atmosphere Atmosphere

Land Surface Land Surface Land Su rface

Ocean & Sea Ice Ocean & Sea Ice Ocean & Sea Ice

Su lfate Aerosols S u lfate Aerosols

Non-Su lfate Aerosols

Carbon Cycle

Dynam i c Vegetation

Atmospheric C h e m istry

Figure 2.10. D evelopment of climate models modified from the 2001 IPCC report
[115] .
30 Cha pter 2. Large-Scale Applications

certainties for these parameters must be determined using model calibration tech­
niques.
Climate models are validated by testing their ability to simulate past climatic
events (paleoclimates) such as the Cretaceous and Last Glacial Maximum (LGM)
based on proxy, measured, or estimated climate forces. Their validity is also tested
by simulating climate responses to current forces such as the eruption of Mount
Pinatubo, as illustrated in Figure 2.7.
Examples of present climate models include the NSF , DOE, and NASA spon­
sored Community Earth System Model (CESM) and the Hadley Centre model
HadCM3 developed in the United Kingdom. The package CESM l .O is publicly
available and includes atmosphere , land, sea ice, ocean, and land ice modules. The
package HadCM3 was highly cited in the 2001 IPCC report [1 15] and includes atmo­
spheric and ocean components, including sea ice . As noted in [178] , the atmospheric
package HadAM3 has on the order of 100 parameters of which approximately 29
are considered to control key atmospheric and surface processes.

2.2.3 Role of U n certainty Quantification in C limate M o deling

It was detailed in Section 2.2.1 that the quantification of uncertainties in temper­
ature and greenhouse gas data and models is critical for ascertaining the present
levels of anthropogenic greenhouse effects and predicting future ramifications of po­
tential warming. We summarize here other ways in which uncertainty quantification
will be critical for making viable climate predictions.
For weather models, uncertainties associated with inputs were comprised pri­
marily of those associated with initial conditions and nonphysical model parameters
that were determined using data assimilation techniques such as 4D-VAR. For cli­
mate models, initial conditions are replaced by forced boundary conditions that
introduce significant uncertainty for phenomena such as predicted greenhouse gas
and aerosol emissions and rates of deforestation. Models for these phenomena are
highly uncertain since they are based on socioeconomic and demographic factors
that are also highly uncertain. Furthermore , since these forcing mechanisms and
associated feedback loops occur in the future, data assimilation techniques are not
viable for constructing associated error bounds or pdf. The uncertainties associ­
ated with parameters and models is augmented by the fact that numerous forcing
and feedback mechanisms occur at subgrid scales and require parameterized phe­
nomenological models to quantify complex or poorly understood physics.
Due to the highly uncertain nature of future aerosol and greenhouse gas emis­
sion levels, the IPCC reports predicted emission levels and climate changes for four
scenarios representing a range of demographic, economic, and technological growth.
• A l : Rapid economic growth with increasingly efficient technology and a mid­
century population peak-A lFl represents fossil fuel intensive energy usage ,
whereas Al T and AlB respectively represent nonfossil and balanced energy
resources .

• A2 : Slow economic and technological growth and large population growth.

2.2. Climate Models 31

• Bl: Same global population dynamics as Al but more rapid change toward a
service and information economy.
• B2: Intermediate population and economic growth with local solutions to
environmental, social, and economic sustainability.

The report does not assign likelihoods to these scenarios .

The IPCC report [188] lists key uncertainties as well as robust findings. Rep­
resentative uncertainties from that report include the following.
• "The magnitude of C0 2 emissions from land-use change and CH 4 emissions
from individual sources remain as key uncertainties."
• "Aerosol impacts on the magnitude of the temperature response , on clouds
and on precipitation remain uncertain."
• "Models differ considerably in their estimates of the strength of different feed­
backs in the climate system, particularly cloud feedbacks, oceanic heat uptake
and carbon cycle feedbacks, although progress has been made in these areas."
• "Large-scale ocean circulation changes beyond the 2 1 st century cannot be
reliably assessed because of uncertainties in the meltwater supply from the
Greenland ice sheet and model response to the warming."
• "Proj ections of climate change and its impacts beyond 2050 are strongly
scenario- and model-dependent , and improved projections would require im­
proved understanding of sources of uncertainty and enhancements in system­
atic observation networks."
• "Understanding of low-probability /high-impact events and the cumulative
impacts of sequences of smaller events, which is required for risk-based ap­
proaches to decision-making, is generally limited."
Since the obj ective of climate models is to predict trends for the future , it is
natural to consider statistical Qol such as average temperatures , precipitation levels,
or amounts of sea level rise. The manner in which Qol and associated uncertainties
are reported depends on the highly varied nature of input uncertainties. When in­
put uncertainties can be reasonable quantified, Monte Carlo simulations from input
densities or confidence intervals are used to construct uncertainty bounds for Qol.
For highly uncertain inputs such as future aerosol or greenhouse gas emissions, pre­
dictions based on the scenarios A l , A2 , B l , and B2 are provided. For example, Fig­
ure 2 . 1 1 , from the 2007 IPCC report [1 88] , illustrates the predicted average change
in global surface temperatures for these scenarios . The scenario AlF l , representing
intensive reliance on fossil fuel energy sources, predicts a best estimate increase of
4 ° C by 2100 with a likely assessed uncertainty range of 2 .4-6 .4 ° C . To place this in
perspective, it is predicted that a 4 ° C temperature rise would cause approximately
35% reduction in African crop production, up to 50% less water available in the
Mediterranean and Southern Africa, coastal flooding that could displace up to 300
million people annually, and loss of up to half the arctic tundra [49] . It is noted
32 Cha pter 2. Large-Scale Applications

.. -�
2020 - 2029 2090 . 2099
6.0

� 5.0
g> 4.0 �
-�
�
.,
�
3.0
-
A1 B . . �
l\?. 1J·. ·�
:- .

81 - :.�
��-�- , . �
�
2.0
::I
"'
'" 1 .0
.0
S2
Cl 0

-1 �
. �����-�� co < � < � <
0 f- aJ u:: .

2000 2 1 00
Year (oC)
1 9 00
0 0:.5 1 1 .5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5

Figure 2 . 1 1 . Levels of average global warming predicted using the scenarios A l ,

A 2 , Bl , and B2 ; from the 2007 IPCC report [188, Figure 3 . 2] .

in the report that due to uncertainties in the ocean-atmosphere couplings , neither

best estimate predictions nor uncertainty bounds could be reasonably established
for sea level rise in 2 1 0 0 . Instead, ranges are reported.
Due to the impact that climate change can have on civilization, it is critical
that physical, biological, and socioeconomic modeling, data collection and analy­
sis, statistical and mathematical analysis, large-scale computing, and uncertainty
quantification continue to be investigated in concert to advance the state of cli­
mate models so they can inform scientists in a manner that includes quantified
uncertainties. This information must then be conveyed to policy makers and the
general public in a manner that encourages economic and technological growth that
minimizes the degree to which anthropogenic forces accelerate climate change.

2.2.4 Notes and References

As with weather modeling, we have focused only on aspects of climate modeling
to highlight issues and motivate the central role of uncertainty quantification for
understanding present climate trends and predicting future climate conditions. The
basic atmospheric phenomena are the same as those for weather , and the underlying
physical principles are detailed in [123, 147, 1 93 , 2 1 3 , 252] . The reader is referred
to [1 72] for a perspective of research issues at the intersection between weather
and climate and [163] for an overview of how the stochastic properties of turbulent
dynamical systems pertain to climate models. The texts [28 , 49, 1 69] provide very
readable descriptions of the global energy balance , the natural and human-induced
causes of climate change, issues associated with proxy measurements obtained using
ice-core and tree ring data, required parameterized components of climate models,
consequences of climate change, and debates pertaining to the subject . Details
regarding various forcing mechanisms can be found in [136] . The 1995, 200 1 , and
2007 Intergovernmental Panel on Climate Change (IPCC) Assessment Reports on
Climate Change [ 1 1 5 , 188, 228] provide a comprehensive discussion of the scientific
basis for and impact of climate change . These reports list key uncertainties and
 Another Random Document on
Scribd Without Any Related Topics
 Einsicht

Eine schneeweiße Pfauentaube saß mit dem Tauber auf dem

Dach. Sie glänzten in der Sonne und schnäbelten sich zärtlich.
»Das ist stark,« sagte das Truthuhn, das seinen Kopf ganz schief
halten mußte und dazu blinzeln um hinaufzusehen. Es wollte weiter
reden; aber da ging der Truthahn vorbei, kollerte und blähte sich,
und das Truthuhn warf sich platt auf die Erde, verliebt und demütig.
Es sah mit seinen blöden Augen zu dem stattlichen Tier empor, das
mit Rasseln und Trommeln dafür dankte und sich aufblies wie ein
Luftballon.
»Daß man einen Tauber anbeten kann!« kreischte das Truthuhn.
»Einen kleinen, unbedeutenden, farblosen Vogel, der keinem
Geschöpf Respekt einzuflößen imstande ist.« Es lag nun flach da,
wie ein breiter, bräunlicher Eierkuchen. Dem Truthahn schwoll der
rote Zierat an Kopf und Hals. Er wurde purpurrot.
»Daß er die Zärtlichkeit der Taube überhaupt für voll nimmt,«
kollerte er. »Daß er so wenig Einsicht hat und glaubt, was die Kleine
da oben girrt.« Er schüttelte sich. Das Truthuhn vor ihm wurde noch
flacher.
 »Er ist ein Tauber,« sagte es verächtlich. »Kein Herrscher, kein
König unter seinesgleichen, kein ...« Es konnte nicht weiter, und
schnappte nach Luft. Sein bläuliches Köpflein bewegte sich vorwärts
und rückwärts. Es schloß die Augen und wartete, ob der Truthahn
seine Ergebenheit belohnen werde. Aber er rauschte weiter. Wie
dunkles Gold glänzte sein Gefieder. Er wußte, daß er der Stolz des
Hühnerhofes war.
Der große, weiße Hahn hatte dem Zwiegespräch zugehört. Er
schwieg. Stolz drehte er den gebogenen Hals, und gravitätisch ging
er seinen Hühnern voran durch den großen Hof. Eine der Hennen
sagte, daß sie sich wundere, daß der Truthahn sich mit der dummen
Dinde abgeben möge, die Verehrung und Zärtlichkeit heuchle. »Und
er glaubt das alles,« sagte ein braungesprenkeltes Huhn, und
trippelte zum Hahn. Der hob sich, schüttelte sich und krähte. Alle
Hühner sahen sich an.
»So wie du, kräht keiner,« sagte eines.
»Wer hat dein stolzes Auge?« fragte ein anderes, und gab der
Nachbarin einen Hieb, denn sie hatte ihm eine Mücke vor dem
Schnabel weggeschnappt.
»Wessen Schwanzfedern wölben sich wie die deinen?«
»Wer ist so weiß wie du?«
»Wer könnte uns beschützen, wie du es tust?« Der Hahn schwieg.
Er war klug. Aber er stolzierte durch den Hof, schlug mit den Flügeln
und krähte, daß alle Hähne der Nachbarschaft antworteten.
Der Enterich, der am Zaun in der Sonne lag, hatte mit seinen
beerenschwarzen Augen dem allem zugesehen. Er war aber zu faul,
um zu sagen, was er dachte. Er wippte nur mit dem Schwänzlein
und schnatterte ganz leise. Seine beiden Enten konnten sich nicht
genug wundern, daß der Hahn solche grobe Schmeicheleien glaube.
Sie sahen hinüber zum Hahn und schnatterten empört und
verächtlich. Dann begannen sie gleichzeitig den Enterich zärtlich zu
lausen. Er ließ es sich gefallen.
Warum auch nicht?
 Eintagsfliegen

Ein paar leichtbeschwingte Fliegen summten um den schön

gezopften Misthaufen im Hühnerhof herum.
Eine von ihnen, eine behäbige, wie blaues Metall glänzende
Roßfliege setzte sich auf den hölzernen Zaun, der den Hof umschloß,
denn es war unter ihrer Würde, sich tiefer unten niederzulassen. Sie
hatte der Welt Großes geschenkt. Eine Entdeckung von
Ewigkeitswert war ihr gelungen: Sie hatte die Grenze der Erde
erreicht. Triumphierend sah sie sich um.
»Die ganze Welt ist nun unser,« sagte sie, und ein Schauer der
Ehrfurcht machte die zarten Flügel der andern erzittern.
»Unser, im wahren Sinn des Wortes,« sagte bewundernd eine
kleine, muntere Fliege. »Nach allen Richtungen haben wir sie
erforscht. Sie birgt kein Geheimnis mehr für uns.« Die tausend
Augen der Zuhörer richteten sich wieder auf die Roßfliege, die aber
unwillig surrte, denn sie liebte es nicht, wenn andere in der Mehrzahl
von ihrer Entdeckung redeten.
Die muntere Fliege kratzte sich etwas verlegen mit dem dünnen
Beinchen den Kopf.
»Ich sage wir, weil ich dadurch andeuten möchte, daß das
Universum teilnimmt an dem Großen, das in diesen Tagen geschah.
Und auch, weil wir andern es uns nicht nehmen lassen wollen, uns
als einen Teil des Ganzen zu fühlen, als einen Stein am Bauwerk der
Wissenschaft, als eine Staffel an der Leiter des Ruhms, deren
höchste Stufe unsere glorreiche Roßfliege – es verneigten sich alle
zum Zeichen des Respekts und schwirrten mit den Flügeln – erreicht
hat.«
Die Gefeierte sah bescheiden mit den tausend Facetten ihrer
Augen zum Himmel auf, mit der anderen Hälfte spiegelte sie im Kreis
herum, ob man ihr auch allseitig die gebührende Hochachtung
erweise.
Plötzlich flogen alle Anwesenden erschrocken auf, denn es nahte
sich ein großer, unbekannter Fliegenschwarm. Sie setzten sich aber
sogleich wieder, da die Herannahenden kamen, um dem
blauglänzenden Forscher Ehre zu erweisen.
Ein gegenseitiges, höfliches Flügelrauschen, Summen und Surren
erhob sich. Ein bewunderndes Auf- und Abwogen, ein Gratulieren,
bescheidenes Abwehren, interessiertes Fragen, bestimmtes, sicheres
Antworten. Eine grünschillernde Fliege sprach für die andern. Sie
wandte sich an die Roßfliege.
»Du hast es erreicht,« begann sie. »Ohnegleichen ist dein Ruhm.
Himmel und Erde sind dir kein Geheimnis mehr. Die Grenze der Welt
hast du erforscht. Unter die Unsterblichen bist du aufgenommen
worden.« Sie funkelte mit ihren geschliffenen Augen die Roßfliege
an, die zusehends dicker, größer, blauer und haariger wurde. Alles an
ihr wuchs und gleißte.
Sie surrte auf das korrekteste ihren Dank, nahm den Orden der
erlösten Paradiesfliegen entgegen, und geleitete darauf die
Deputation über den Misthaufen, durch den Hof, weit in den Garten
hinaus. –
»Die Blaue platzt noch vor Hochmut,« sagte eine Biene, die an ihr
vorüberflog. Sie kroch in eine rosafarbene Primel, blieb dort eine
Weile, und kam heraus, die Füßchen voll Blütenstaub. »Was hat man
davon, wenn man weiß, daß am Ende der Welt ein Berg ist, den
keiner überfliegen kann?« fragte sie.
 »Nichts,« sagte verächtlich ein Schmetterling, der auf der Primel
saß. »Aber was hast du von deiner Arbeit?«
Verblüfft sah die Biene ihn an.
»Genug, meine ich. Die Welt bewundert uns und braucht uns.
Ohne uns schritte der dürrbeinige Hunger durch das Land. Ohne uns
stürbe, was Odem hat. Was ich davon habe? Dumme Frage: Wir sind
die Ernährer der Welt.« Zornig schnellte sie ihren Stachel gegen den
samtnen Schmetterling.
Er wiegte sich jetzt auf einer frühen Narzisse, die weiß wie er, ihr
goldenes Krönchen auf der Stirne, ihren zarten Duft verbreitete.
»Arbeitstiere ihr,« sagte er verächtlich. »Ihr braucht auf euren
Stand auch noch stolz zu sein. Grobes Volk, aller Schönheit bar. Wir
Schmetterlinge sind der Zweck der Schöpfung. Wir sind das Schöne.
Wir tragen den blauen Himmel, die bunten Blumen, die
durchsichtigen Steine und den Schimmer des Goldes auf unsern
Flügeln. Wir baden uns im flirrenden Sonnenstrahl und nähren uns
von glitzerndem Tau. Wir leben um zu genießen. Ohne uns wäre die
Welt öde, glanzlos, traurig.«
Er berührte den silberschimmernden Atlas der Narzisse mit den
zarten Flügeln. Die Biene flog mürrisch summend davon, dem Garten
zu, wo ihr Korb stand. Sie flog mit ihren beschwerten Füßen langsam
an der blauen Roßfliege vorüber, die eben heimkehrte in der Mitte
ihrer Anbeter.
»Faulenzer,« brummte die Biene.
In der Nacht kam ein Frost. Am Morgen lagen sie alle starr und
steif am Boden, die Fliegen, die Biene und der Schmetterling. Auf
dem Rücken lagen sie und streckten die Beine gen Himmel. – Über
ihnen lächelten die Sterne.
 Der Gesangverein
Mitten in einem Steinbruch lag ein Tümpel, der von großen Blättern
beschattet war, und wo der Gesangverein »Froschenia« seine
Übungen abhielt. Jeden Donnerstag Abend.
Es war ein feiner Gesangverein, und nur feine Leute sangen mit.
Waren andere Elemente eingetreten, so wurden sie rechtzeitig
hinausgeekelt. Sämtliche Mitglieder hatten Grün als ihre Farbe
erwählt, und so erschienen bei den Übungen die Damen in grünen
Roben, die Herren in ebensolchen Fracks, mit weißen oder gelben
Piquéwesten*.
* Die Verpflichtung, nur grün gekleidet den Übungen beizuwohnen, hielt
auch Unbemittelte fern. Sie waren nicht erwünscht.

Es war kurz vor acht Uhr. Man fand sich immer ein paar Minuten
vor der Zeit ein, teils um des notwendigen, sehr beliebten Flirts
willen, teils um die ebenso berechtigte Medisance zu Wort kommen
zu lassen.
Nur die ganz jungen Fräuleins kamen naiv um des Singens willen,
sie schwärmten aber für den Kapellmeister. Daß er verheiratet war,
tat nichts zur Sache, da die Backfische gänzlich wunschlos
schwärmten.
 Der Meister kam und bestieg sein Pult.
»O Gott!« sagten die ganz jungen Fräuleins und sahen sich
errötend an.
Die älteren Damen warfen sich bedeutungsvolle Blicke zu; denn
der Musiker hatte mit einer unter ihnen einen kurzen Gruß
getauscht. Man wußte, daß sie einander sehr genau kannten.
»Ekelhafter Patron,« näselte ein Student, »glaubt sich hier Hahn
im Korb.« Die Studenten waren wütend, weil sie von den
Backfischen nicht beachtet wurden; die hatten nur Augen für den
verehrten Meister.
Der Kapellmeister gab das Zeichen zum Beginn. Man übte aus
Carmen.
»Ich bitte die Bässe gefälligst beginnen zu wollen!« rief der
Dirigent. Alle erhoben sich, und: Mut in der Brust, siegesbewußt!
erscholl es unter den großen Blättern.
»Halt! Etwas mehr Feuer, muß ich bitten!« rief der Kapellmeister.
Also noch einmal: Mut in der Brust, siegesbewußt! Diesmal wurde
der Chor zu Ende gesungen.
»Meine Herren,« schrie der Musiker, »ich habe Sie singen lassen,
um einmal den vollen Eindruck eines derartigen Singsangs zu
bekommen! Meine Herren! Ist das gesungen? Das ist geleiert. Haben
Sie denn keine Ahnung, was das ist: Mut in der Brust? Hat denn
noch keiner von Ihnen Siegesbewußtsein empfunden! Mut ist – –«
Da patschte es im Wasser, ein roter Schnabel wurde sichtbar, und
ein paar lange Beine traten mitten zwischen die grünen Herren und
Damen.
Eine Sekunde lang war alles starr; dann sprangen, huschten,
schwammen und quiekten sie durcheinander und waren im Nu
verschwunden. Der Storch hatte das Nachsehen und schrie
höhnisch: »Jawohl! Mut in der Brust! Ihr seid mir die Rechten!«
Dann suchte er sich anderswo sein Nachtessen.
Es dauerte eine lange Weile, ehe sich einer der Herren Frösche
hervorwagte, dann ein zweiter und dritter. Endlich war die ganze
Gesellschaft wieder beisammen. Sie waren noch ganz aufgeregt von
dem Abenteuer. Der Dirigent begab sich an sein Pult, bedeutend
milder gestimmt, klopfte dreimal mit seinem Stab und rief: »Meine
Herren und Damen! Ich bitte um das Lied: Ein Veilchen auf der
Wiese stand!« Das Lied wurde mit Empfindung gesungen, und
mußte nur ein einziges Mal wiederholt werden.
Da aber der Meister beim Schluß des Liedes die Augen jedesmal
nach der vorerwähnten Dame verdrehte, begann ein Backfisch zu
kichern, andere fielen ein, es lachten die älteren Fräuleins, es
lachten die Damen, die Herren, zuletzt alle.
Wütend klopfte der Kapellmeister.
»Ich bitte die ungezogenen Backfische den Saal zu verlassen!«
schrie er.
»Was! Ungezogenen Backfische!« brüllten die Studenten, »das
lassen wir uns nicht gefallen!« Sie stürmten das Dirigentenpult,
packten den armen Musiker an den Beinen, und in hohem Bogen
flog er ins Wasser.
Die Übung war aus. Arm in Arm verließen die Studenten mit den
Backfischen das Lokal. Diese schwärmten nun für ihre Verteidiger,
die ihren Vorteil wahrnahmen. Hinter jedem Blatt saß ein Pärchen.
Die Damen waren vorsichtiger; sie trafen sich mit den
betreffenden Herren erst weit draußen und benutzten die Schatten
der Binsen.
Die älteren Herren aber zogen in den »Goldenen Frosch«, um den
heutigen Abend durchzusprechen. Die Stimmung war sehr animiert,
es wurde viel getrunken. Nach Mitternacht schwankte eine Reihe
grünbefrackter Gestalten nach Hause und sang: Mut in der Brust,
siegesbewußt! mit so viel Verve und Feuer, daß auch der
Kapellmeister zufrieden gewesen wäre!
 Das kluge Huhn

Im Hühnerhof war große Gesellschaft. Von überall her waren die

Hühner und Enten eingeladen. Zu einem Gericht frischer Maikäfer
hieß es, in Wahrheit aber um die neue Nachbarin in Augenschein zu
nehmen. Es ging das Gerücht in der Gegend, daß sie eine
Andalusierin sei. Und das mußte wahr sein, denn tief schwarz war
das Gefieder, und blau die Bäcklein, wirklich blau. Andere Spanier,
Minorka zum Beispiel, hatte man ja auch schon gesehen, aber
Andalusier noch nie.
Die Fremde benahm sich wirklich nett. Sie begrüßte jede der
Hennen einzeln, und nur ganz kurz den Hahn.
Sie beantwortete sämtliche Fragen mit »Ja« oder »Nein«. Selber
fragte sie nichts.
 Nur bei den Hennen, die Junge hatten, forschte sie eifrig, ob alle
die Kleinen gesund seien, und fügte hinzu, daß sie selten so hübsche
Jungen gesehen habe.
Diese weise Frage hatte sie von ihrer Großmutter.
»Kücken,« hatte die gesagt; »du gehst nun in die weite Welt. Klug
bist du nicht. Also gibt es für dich nur zweierlei zu beobachten:
Begegnet dir ein Hahn, so sei schweigsam, und begegnet dir ein
Huhn, so lobe seine Jungen. Beide werden deine Klugheit preisen.«
Die einzige Klugheit des spanischen Kückens war aber die, daß es
seiner Großmutter gehorchte. Und auch diese Klugheit verdankte es
nur seiner Dummheit. Es fiel ihm leichter zu gehorchen, als selbst zu
denken.
Der Rat des alten spanischen Huhnes bewährte sich.
»Es ist wirklich eine gescheite Henne,« sagten die mütterlichen
Hühner.
»Das ist sie,« bestätigte der Hahn und fügte anzüglich hinzu:
»Wenigstens gackelt sie nicht den ganzen Tag wie gewisse andre.
Sie muß klug sein.«
Nun war die Parole ausgegeben. Die kluge Spanierin wurde sie
überall genannt.
»Sie kann reizend zuhören,« sagten die guten, schwatzhaften
Hühner und merkten nie, daß die Fremde bei ihren Erzählungen die
Augen geschlossen hatte und träumte.
»Und so bescheiden ist sie,« sagte die alte Ente. Sie konnte es
nicht leiden, wenn ihr jemand widersprach, ganz besonders wenn es
junge Leute waren. Die Jungen hatten »Ja« zu antworten, und damit
basta.
Und »Ja« antwortete die Spanierin immer, warum hätte sie »Nein«
sagen sollen? Es war ihr ja ganz gleichgültig, was die Alte da
behauptete.
Der Hahn aber liebte seine schwarze Andalusierin sehr. Sie
bewunderte ihn schweigend. Mit kindlichen, runden Augen sah sie zu
ihm auf. Sie schwieg, wenn die anderen gackelten. Sie lief immer
dicht hinter dem Hahn und ging nie eigene Wege. Auch hatte sie nie
eine eigene Meinung.
Später hatte die Spanierin Junge. Reizende, schwarze, kleine
Geschöpfe. Sie hütete und fütterte sie und lief nie von ihrer Seite.
Das tut ein Huhn aus Instinkt, dazu braucht es keinen Verstand.
Als sie aufwuchsen, gab es freilich Hindernisse.
»Was muß ich tun, um in der Welt fortzukommen?« fragte einer
der jungen Gockel.
»Du mußt »ja« und »nein« sagen und die Kücken der Hennen
loben,« sagte das Huhn. »Das hat mich meine Großmutter gelehrt,
und ich bin gut damit ausgekommen.«
Das Huhn sah nicht, daß hinter dem Zaun eine schöne, bunte
Katze saß, mit feurigen, gelben Augen, die das Gockelchen
unverwandt anstarrte. Es lief auf sie zu. Die Katze packte es und
trug es im Maul davon.
»Was muß ich tun, um in der Welt fortzukommen?« fragte auch
eines der jungen Hühnchen.
»Du mußt dem Gockel gefallen, das hat schon meine Großmutter
mich gelehrt,« sagte das spanische Huhn und warnte das Hühnchen
nicht vor dem Habicht, der mit gierigen Augen über dem Hühnerhof
kreiste. Er schoß herab und packte das Hühnchen mit seinen
scharfen Krallen.
Auch die andern Kücken kamen gelaufen.
»Was ist das Schönste in der Welt?« fragten sie.
 »Das Schlafen,« sagte die Andalusierin und schloß ihre Augen. Die
Kücken schlossen auch ihre Augen.
Sie sagten des ganze Leben »ja« und »nein«. Sie fraßen und
schliefen.
Als das spanische Huhn starb, hielt der Hahn die Grabrede. Er
nannte die Andalusierin die Klügste des Hühnerhofes.
»Des Hühnerhofes,« nickte klappernd der Storch und flog davon.
 Der alte Schafbock

In der Schafherde lebte ein alter Bock. Er war nicht liebenswürdig

gewesen, als er jung war, und er glaubte nun das Recht zu haben,
noch viel weniger liebenswürdig zu sein, da er alt geworden. Um
seines Alters willen mußte die ganze Herde sich ihm beugen. Einzig
gegen die jungen Schäfchen war er freundlich, und die hätten es
lieber gehabt, wenn er weniger artig gewesen wäre. Dies fanden
auch die jungen Böcke. »Das Recht des Alters« nannte es der
Schafbock, wenn er an den Lämmern herumschnüffelte.
Er hatte aber noch andre Eigenschaften.
Meistens erzählte er lange, langweilige Geschichten, und vergaß
im Laufe der Erzählung das Ende. Er fing dann von vorne an und
erzählte die Geschichte noch einmal. Aber dann passierte es ihm
leicht, daß er die Pointe einer andern Erzählung dieser anfügte. Das
merkten aber nur die andern, er selbst nie.
Man sah es ihm immer an, wenn er erzählen wollte, er hatte dann
einen matten, in sich gekehrten Blick. Wer ihn bemerkte, nahm
Reißaus. Nur die ganz Jungen nicht, die sahen nie, was ihnen
drohte; noch nicht einmal merkten sie es, wenn der Alte sie zu sich
rief.
»Laßt euch einmal erzählen, wie zu meiner Zeit die Alten
behandelt wurden,« sagte er dann, und die Helden seiner
Geschichten wurden jedesmal tugendhafter, und die Böcklein, die
zuhörten, kamen sich jedesmal gemeiner vor, wenn sie sich mit den
Altersgenossen des alten Schafbockes verglichen. Aber wieder nur
die ganz Jungen.
Die andern kannten die Form, nach der die schönen, moralischen
Lügen geprägt wurden.
Der alte Bock hatte aber nicht nur Belehrendes zu erzählen. Stand
er unter den Schafen, so ging es nach einer andern Melodie, und
hatte er sich gar an die Jungen herangeschlichen, so hörte er
überhaupt mit Erzählen auf, und die Lämmlein mußten sich von ihm
lecken lassen, so sehr ihnen vor seinen kahlen Stellen im Pelz und
seinen roten Augen ekelte.
»Denkt, ich sei euer Großvater, meine Lämmchen,« sagte er. Aber
das dachten sie nicht und sprangen bei der ersten Gelegenheit
davon.
Der alte Schafbock hörte nicht mehr gut, deshalb mußte jedes
Wort, das in seiner Gegenwart gesprochen wurde, wiederholt
werden, auch das gleichgültigste.
»Das ist das Recht des Alters,« behauptete er auch da. Zudem
nahm er alles übel, und die Jungen mußten um Verzeihung bitten,
wenn sie es schon nicht böse gemeint hatten.
»Das ist die Pflicht der Jugend,« sagte er. Er hatte auch ein
schlechtes Gedächtnis und wiederholte fortwährend dasselbe. Wenn
den andern die Geduld ausging, und sie über ihn weg zusammen
redeten, wurde er wütend.
»Nie wäre so etwas zu meiner Zeit möglich gewesen,« schrie er.
»Die heutige Jugend ist entartet, der Respekt vor dem Alter ist tot!«
»Warum soll man eigentlich gerade vor dem Alter Respekt
haben?« fragte ein kräftiges Böcklein.
»Warum? Warum?« Der Alte schnappte nach Luft. Er erstickte fast
vor Zorn. Er schnaufte und nieste und schäumte und bespritzte die
Umstehenden. Aber als er fertig war, fand er doch keine Antwort.
»Darum!« mähte er endlich heiser. »Ich verlange Respekt von
euch, das ist mein Recht! Ihr habt zu schweigen, wenn ich rede, ihr
habt zuzuhören, wenn ich erzähle, ihr habt stillzuhalten, wenn ich
euch liebkose. Ihr habt mir nicht zu widersprechen, wenn ich etwas
behaupte, und ihr habt mich zu ehren und zu lieben und zu achten.«
Erschöpft schwieg er.
»Warum?« fragten sie wieder. »Wir wollen wissen warum!«
»Weil ich alt bin!« Der alte Schafbock ging seinem Stalle zu, um zu
schlafen.
»Wenn er freundlich wäre,« sagten die Schafe, »wir wollten ihm
gerne helfen und ihm dienen!«
»Wenn er würdevoll wäre,« sagten die jungen Böcklein, »wir
wollten ihm gerne gehorchen.«
»Wenn er weise wäre,« sagten die alten Schafe, »wir hörten
gerne seine Lehren. Aber er ist nur alt. Hat er darum ein Recht auf
unser aller Wohlbehagen?«
»Nein,« schrien alle, »er hat keines! Wir wollen ausziehen und uns
belehren über die Rechte des Alters.« Die ganze Schafherde ging
über Land.
Sie fanden ein altes Pferd auf der Weide. Still und ruhig graste es.
Sprangen unerfahrene, junge Pferde zu nahe an den Fluß, so hielt es
sie auf. Den Füllen wehrte es die Fliegen. Wollten die Pferde in
wildem Jagen ihre Glieder üben, so stand es beiseite, und freute sich
der tollen Sprünge und gedachte dabei der eigenen Jugend. Und die
jungen Pferde suchten die saftigsten Kräuter und führten das alte
Pferd dorthin. Sie rieben sich schmeichelnd an ihm und scherzten
mit ihm. Sie liebten es, denn es freute sich ihrer Jugend.
»Hat das alte Pferd von seinen Rechten gesprochen?« fragte der
Leiter der Schafherde.
 »Kein Wort!« riefen alle. Darauf fanden sie einen rissigen, uralten
Baum. Hohl war sein Stamm, und dürre Äste ragten traurig zum
Himmel auf. Aber fröhlicher Efeu war am Stamme in die Höhe
geklettert und schmiegte sich schmeichelnd an die Eiche.
»Kann der Baum den Efeu zwingen, ihn zu schmücken, darum weil
er alt ist?« fragte der Leitbock die Herde.
»Niemals,« antworteten die Tiere.
Am Bache lag ein alter, grauer Stein. Er lag mitten im Flußbett und
störte den Lauf des Bächleins. Aber er hatte sich mit grünem Moos
bedeckt, er hatte seine scharfen Kanten und Ecken vom lustigen
Wässerlein abschleifen lassen und hörte freundlich auf sein Murmeln
und Plätschern, und freute sich des munteren Gefährten, der sein
Alter erheiterte.
»Warum kräuselt sich der Bach so gerne um den alten Stein?«
fragte der Bock die Herde.
»Weil der Alte ihn nicht hemmt!« rief die Herde.
»So brauchen wir nicht weiter zu ziehen,« sagte der Bock. »Wir
wissen nun, was wir wissen wollten.« Und sie zogen heimwärts bis
zu ihrer Weide, wo der alte Bock mürrisch an der Sonne lag und
schalt, daß man ihn so lange allein gelassen.
»Ich habe ein Recht, zu verlangen, daß man bei mir bleibe,« rief
er und stieß die Nahestehenden mit den Hörnern.
»Fort mit dir,« schrie nun die ganze Herde. »Du hast kein Recht
auf uns, nur weil du alt bist! Gehe zu Pferd, Baum und Stein und
lerne von ihnen, wie man sich Liebe erwirbt.« Und sie ließen den
Bock stehen und rannten leichtfüßig hinauf in die Berge, in die
Sonne, zu duftendem Tymian und Vergißmeinnicht.
 Vom bescheidenen Hähnchen

»Frau Mutter, wir möchten uns ein wenig in der Welt umsehen,«
sagte das jüngste Hähnchen zu der Henne.
»Ja, das möchten wir,« sagte auch das älteste.
»Was heutzutage die Kinder nicht alles wollen!« Die Henne
schüttelte den Kopf. »So geht! Ihr werdet bald genug wieder da
sein. Und was ich sagen wollte: Seid ja recht bescheiden und drängt
euch nirgends vor. Das können die Erwachsenen nicht leiden.«
Die Hähnchen machten sich eilends davon und krähten heiser und
vergnügt in die Welt hinaus. Die Henne sah ihnen nach.
»Um den Ältesten ist mir nicht bange,« sagte sie zum Hahn, »aber
der Jüngste.«
»Jugend hat keine Tugend,« bedeutete sie der Hahn.
Die Hähnchen zogen über das Feld, und das jüngste wurde
hungrig.
»Hast du etwas zu essen?« fragte es seinen Bruder.
»Nein,« sagte der Älteste; »aber da kriecht eine fette Raupe.«
»Danke!« sagte das Jüngste, und fraß sie auf. Verblüfft sah der
andere zu.
 »Eigentlich hätte sie mir gehört. Ich habe sie zuerst gesehen.«
»Aber ich habe sie zuerst gefressen,« sagte ruhig das Hähnchen.
Sie liefen weiter und liefen manchen Tag, und die Welt hatte
immer noch kein Ende. Es wurde ihnen fast unheimlich zumute.
»Ich wollte, ich wäre wieder daheim bei der Frau Mutter!« sagte
das Älteste.
»Das glaube ich!« lachte der Fuchs, der plötzlich vor ihnen stand.
»Welches von euch beiden möchte nun zuerst gefressen werden?«
»Bitte, Herr Fuchs, ich warte gerne,« sagte das jüngste Hähnchen
bescheiden.
Da packte der Fuchs den Ältesten und zerriß ihn. Das Jüngste aber
lief über das Feld heimwärts, so schnell es konnte. Es rannte und
flog und krähte, bis es endlich bei seiner Mutter war.
»Frau Mutter,« schrie es schon von weitem, »oh, wie recht haben
Sie gehabt. Bescheidenheit ist eine schöne Sache.«
»So,« sagte die Henne und sah ihren Jüngsten mißtrauisch an,
»und wo hast du denn deinen Bruder?«
»Den hat der Fuchs gefressen, Frau Mutter. Und hätte ich nicht
auf Sie gehört und mich unbescheiden vorgedrängt, so hätte die
Sache schief ablaufen können.«
 Das neue Buch

Es war einmal ein alter Uhu, der nicht mehr auf die Jagd gehen
konnte, und sich von seinen Söhnen füttern lassen mußte. Da dachte
er, daß er ein Buch schreiben wolle, und zwar ein Buch, in dem man
sehen konnte, wie es in der Welt zugehe. Er wollte es drucken lassen
für die Schulkinder.
Er ließ seine drei Freunde kommen: Den Maulwurf, den Hahn und
die Schwalbe; die sollten ihm berichten, was sie von der Welt
wüßten.
Es waren Leute, die viel erfahren hatten, zudem wichen sie nie
von der Wahrheit ab, und dem Uhu lag besonders viel daran, daß in
dem Buch nur die reine Wahrheit gesagt würde.
Sie begaben sich zusammen an den Rand des nächsten Waldes,
um ungestört verhandeln zu können. Der Uhu saß im Stamm einer
alten, hohlen Eiche, der Hahn ging gravitätisch davor auf und ab und
der Maulwurf grub sich ein Loch, aus dem er nur den Kopf
herausstreckte. Die Schwalbe aber flog auf den untersten Zweig des
Baumes, unter dem sie beraten wollten.
Der Uhu nahm sein Notizbuch, spitzte seinen Bleistift, und bat den
Maulwurf anzufangen. Der setzte sich in Positur und begann:
»Die Welt ist dunkel.«
»Dunkel?« fragte die Schwalbe verwundert.
 »Ja, dunkel,« antwortete der Maulwurf bestimmt. »Dunkel und
eng. Lange, schmale Gänge durchziehen sie, in denen man bequem
gehen kann. Man macht die Gänge selbst, und hat viel Arbeit damit.
Nahrung gibt es in Menge. Die Tiere besitzen alle einen schwarzen
samtnen Pelz.«
»Einen schwarzen Pelz!« rief der Hahn. »Was für ein Unsinn!«
»Jawohl, einen schwarzen Pelz! Es gibt auch Maulwürfe, die einen
weißen Pelz haben. Aber zum Glück sind sie sehr selten. Man
verachtet sie, weil sie nicht sind wie alle andern.«
Der Uhu schrieb alles, was der Maulwurf gesagt, in sein Notizbuch.
Zu einigen Mitteilungen machte er Bemerkungen. Er sagte aber
nichts, sondern fragte höflich den Maulwurf, ob er noch etwas
mitzuteilen habe.
»O ja,« sagte der Maulwurf, »die Hauptsache! In der Welt ist es
sehr langweilig. Ein Tag ist wie der andere, und man hat nur zwei
Zerstreuungen. Die eine ist das Essen. Die andere ist, daß man alle
anderen Tiere über die Achsel ansieht, die nicht in der Welt wohnen
und nicht leben wie die Maulwürfe. Und das ist die feinste Freude für
einen Maulwurf.«
Der Uhu notierte alles. Darauf bat er den Hahn, nun auch seine
Erfahrungen mitzuteilen.
»Die Welt,« begann der Hahn, »ist meistens eine lustige Sache.
Genug zu essen, genug zu trinken und Hühner, soviel man will!«
»Soviel man will!« stöhnte entsetzt der Maulwurf.
»Jawohl! Soviel man will! Die Welt ist viereckig und hat einen
Zaun aus Draht rings herum. Die Welt hat ein Licht am Himmel,
dann ist es warm. Manchmal fallen aber weiße Fetzen vom Himmel
und dann ist es kalt.«
»Weiße Fetzen?« fragte erstaunt die Schwalbe.
 »Ja, und wenn die herabfallen, wird die ganze Welt weiß davon.
Kein Tier legt dann Eier. Es gibt in der Welt jemand, der einem alle
Tage Futter bringt. In der Welt haben die Tiere Federn und einen
roten Kamm.«
»Einen Kamm?« riefen Maulwurf und Schwalbe. »Das ist nicht
wahr.«
»So! Nicht wahr!« krähte heftig der Hahn. »Ich habe doch einen,
und unsere Kücken haben einen, wenn sie zur Welt kommen, meine
Hühner haben einen, und dann: nicht wahr! Jedes Wort ist wahr, das
ich sage! Ich habe alles selbst beobachtet, ich lebe mitten in der
Welt und betrachte sie von morgens bis abends.«
Der Uhu bat höflich den Hahn, sich nicht zu ärgern. Es zweifle
niemand an der Wahrheit seiner Behauptungen, nur nehme eben
nicht jedes denselben Standpunkt ein. Da gebe es dann leicht
Differenzen.
»Das Schönste in der Welt,« fuhr der Hahn besänftigt fort, »ist der
Misthaufen. Das ist eine wahre Fundgrube. Würmer, Käfer, Körner,
kurz alles, was man sich wünschen kann, ist vorhanden. Das ist eine
Lust, wenn alle da kratzen und scharren, picken und gackern, und
nie fühlt man sich so als Mann, als wenn man auf seinem Mist steht
inmitten seiner Hühner und stolz in die Welt hinauskräht.«
Ganz ergriffen hörte der Uhu zu. Zu der letzten Bemerkung des
Hahns machte er ein Kreuz, damit er sie besonders sorgfältig
ausarbeite.
Dann bat er die Schwalbe, nun auch ihre Beobachtungen und
Erlebnisse zum besten zu geben.
»Die Welt,« fing die Schwalbe an, »ist unendlich groß. Sie besteht
aus Meeren und Ländern, aus Bergen und Tälern. Das Schönste in
der Welt ist, wie ein Pfeil die Luft zu durchmessen, von einem Land
ins andere, Meere zu überfliegen und seine Brust dem Sturme
preiszugeben.«
 »Ein gräßliches Vergnügen!« wimmerte der Maulwurf, und der
Hahn und der Uhu schüttelten ihre Köpfe. Der Uhu fragte nicht
weiter. Es kam ihm gar zu phantastisch vor, was die Schwalbe
erzählte, gar zu unwahrscheinlich und übertrieben. Jedenfalls würde
er sich in seinem Buch mehr an die beiden andern halten.
Der Uhu dankte den Dreien sehr für die nützlichen Mitteilungen,
und versprach jedem ein Exemplar des Buches, wenn es erscheinen
würde. Er sagte, daß die Ansichten der drei Freunde weit
auseinander gingen, daß aber, da alle drei ehrenwerte Leute seien,
an ihrem Worte nicht zu zweifeln sei. Er werde alles sorgfältig prüfen
und aus allen Darstellungen dasjenige nehmen, was ihm für die
Kinder das Passendste scheine.
Nach einigen Monaten kam das Buch für die Schulkinder heraus.
Lehrer Storch las in der Schule daraus vor. Es hieß da:
Die Welt ist dunkel. Oft ist eine Sonne da, doch scheint sie nicht
immer. Wenn sie scheint, sehen sie nicht alle.
In der Welt haben die Tiere einen Kamm, manchmal aber einen
schwarzen Pelz. Die Welt ist unendlich groß, und alles ist mit einem
Zaun umgeben. Sie ist viereckig.
Das Schönste in der Welt ist der Misthaufen. Einige fliegen darüber
weg und geben ihre Brust dem Sturme preis, die meisten aber
krähen und suchen Würmer.
In der Welt sind enge, dunkle Gänge und darinnen verachtet man
die andern Tiere. In der Welt ist es sehr langweilig, manchmal auch
lustig, besonders wenn man Hühner hat, soviel man will und genug
zu essen.
Viele Tiere sehen Flocken vom Himmel fallen, andere sehen sie
nie.
In der Welt bringt jemand den Tieren Futter ... usw.
 Als der Storch fertig vorgelesen hatte, mußten die Kinder es
durchbuchstabieren, und dann mußten sie es auswendig lernen.
Der Uhu hatte es sich lange überlegt, welche der verschiedenen
Ansichten der Tiere er bringen wolle, denn sie stimmten ja durchaus
nicht überein. Er wollte keinen seiner Freunde ärgern, indem er
etwas wegließ, auch war ihm alles gleich wertvoll und schien ihm
unentbehrlich für sein Buch.
Zuletzt fand er einen Ausweg. Er machte Zettelchen, schrieb
sämtliche Beobachtungen von Maulwurf, Hahn und Schwalbe einzeln
darauf, warf sie dann in eine Schüssel, schüttelte sie tüchtig und fing
an zu ziehen. Den ersten Zettel, den er zog, gebrauchte er für das
Buch, den zweiten nicht, den dritten wieder für das Buch, den
vierten nicht, und so weiter, bis er den letzten gezogen hatte.
Das war gerecht und einfach und konnte ihm keinerlei
Unannehmlichkeiten zuziehen. Und so entstand das Buch.
Der Storch stattete dem Uhu einen Besuch ab und dankte ihm
begeistert im Namen der heranwachsenden Jugend für das
interessante Werk.
 Die lieben Nachbarn

»Habt ihr es schon gehört, der Nachbar von nebenan will eine
Stadtmaus heiraten!« sagte eine Feldmaus zu ihren Besucherinnen.
Sie glättete ihr braunes Pelzlein und ringelte zierlich den Schwanz.
»Eine Stadtmaus? Doch nicht die Weiße mit den roten Augen, die
neulich hier auf Besuch war?«
»Gerade die!«
»Jetzt hört aber doch alles auf!« jammerte eine der drei, eine
fette braune Feldmaus. »Also die Weiße! Nun, der Nachbar kann sich
gratulieren!«
»Warum? Was wissen Sie von der weißen Maus?« schrien
aufgeregt die andern.
»Ich weiß nichts, und ich sage nichts; aber denken tue ich mein
Teil.«
»Woher wissen Sie es, Frau Feldmausin?« fragten die drei und
rückten näher zusammen.
»Das darf ich nicht sagen. Aber die Person, die es mir mitteilte, ist
zuverlässig, durchaus zuverlässig. Wenn das unser Nachbar wüßte!
Der würde sich schwer hüten, so eine zu heiraten.«
»Man sollte ihn warnen,« riefen alle; »das ist beinahe unsere
Pflicht!«
 »Jawohl, es ist eigentlich unsere Pflicht!« Alle nickten mit den
Köpfen und sahen sich bedeutungsvoll an. Es glänzte
unternehmungslustig in den beerenschwarzen Äuglein. Und die vier
machten sich eilig auf, und gingen zum Nachbarn hinüber.
»Herr Nachbar, wir kommen in einer delikaten Angelegenheit.«
»Liebe Freundinnen, ihr kommt gewiß, um mir zu gratulieren. Es
ist ja kein Geheimnis mehr, gar nicht.« Die vier lächelten sauersüß
und wünschten Glück.
»Meine Braut ist reizend,« rief der Verliebte. Die vier nickten.
»Das ist sie, gewiß; dagegen ist nichts zu sagen.«
»Und tugendhaft,« betonte nochmals der Nachbar.
Die langen Schnurrbarthaare der Feldmäuse zitterten vor
Erwartung.
»Jetzt!« sagte leise die eine, und stieß ihre Nachbarin an, damit
sie reden solle.
»Herr Nachbar,« begann die Fette und räusperte sich, »es ist
leider unsere Pflicht, Ihnen mitzuteilen, daß Ihre Braut ...«
»Daß meine Braut?«
»Das Lob, tugendhaft zu sein, nicht ganz verdient.«
»So,« sagte der Nachbar, »was wissen Sie denn von ihr?« Die
fette Maus kam etwas aus der Fassung: Der Bräutigam blieb gar zu
gelassen.
»Sie ist ... sie hatte ... kurz, man hat sie mit einer braunen Maus
im Mondschein spazieren sehen!« Erleichtert setzte sich die
Feldmaus; es war eben keine Kleinigkeit, einem Bräutigam so etwas
zu sagen.
»So!« sagte der Nachbar.
 »So! So, sagen Sie, Herr Nachbar? Und mit diesen Grundsätzen
wollen Sie in die Ehe treten? Bei so etwas bleiben Sie gelassen? Die
beiden haben sich nämlich auch geküßt!« Triumphierend sah die
Feldmaus im Kreise herum.
Der Nachbar lachte. Da erhoben sich alle vier würdevoll.
»Wir haben unsere Pflicht getan,« sagten sie. »Das Weitere ist
Ihre Sache!« Steif wandten sie sich zum Gehen, ihre Schwänzchen
fuhren aufgeregt hin und her. Sie waren schwer enttäuscht. »Wir
bedauern gestört zu haben!«
»Gar nicht, aber gar nicht!« rief der Nachbar. »Die große,
dunkelbraune Maus bin ich nämlich selber gewesen. Übrigens lade
ich Sie alle zur Hochzeit ein.«
Und er öffnete die Türe und machte eine tiefe Verbeugung ...
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookgate.com