Eawag: Swiss Federal Institute of Aquatic Science and Technology Use of time-dependent parameters for improvement and uncertainty estimation of dynamic.

Slides:



Advertisements
Similar presentations
Eawag: Swiss Federal Institute of Aquatic Science and Technology Analyzing possible causes of bias of hydrological models with stochastic, time-dependent.
Advertisements

Bayesian Belief Propagation
Use of Kalman filters in time and frequency analysis John Davis 1st May 2011.
FTP Biostatistics II Model parameter estimations: Confronting models with measurements.
This presentation can be downloaded at – This work is carried out within the SWITCH-ON.
Geog 409: Advanced Spatial Analysis & Modelling © J.M. Piwowar1Principles of Spatial Modelling.
Gizem ALAGÖZ. Simulation optimization has received considerable attention from both simulation researchers and practitioners. Both continuous and discrete.
CHAPTER 16 MARKOV CHAIN MONTE CARLO
Stochastic approximate inference Kay H. Brodersen Computational Neuroeconomics Group Department of Economics University of Zurich Machine Learning and.
Computing the Posterior Probability The posterior probability distribution contains the complete information concerning the parameters, but need often.
PROVIDING DISTRIBUTED FORECASTS OF PRECIPITATION USING A STATISTICAL NOWCAST SCHEME Neil I. Fox and Chris K. Wikle University of Missouri- Columbia.
Statistics for Managers Using Microsoft® Excel 5th Edition
Particle filters (continued…). Recall Particle filters –Track state sequence x i given the measurements ( y 0, y 1, …., y i ) –Non-linear dynamics –Non-linear.
A Concept of Environmental Forecasting and Variational Organization of Modeling Technology Vladimir Penenko Institute of Computational Mathematics and.
Simulation Models as a Research Method Professor Alexander Settles.
End of Chapter 8 Neil Weisenfeld March 28, 2005.
1 Simple Linear Regression Chapter Introduction In this chapter we examine the relationship among interval variables via a mathematical equation.
Modern methods The classical approach: MethodProsCons Time series regression Easy to implement Fairly easy to interpret Covariates may be added (normalization)
Arizona State University DMML Kernel Methods – Gaussian Processes Presented by Shankar Bhargav.
Course AE4-T40 Lecture 5: Control Apllication
Computer vision: models, learning and inference Chapter 10 Graphical Models.
Space-time Modelling Using Differential Equations Alan E. Gelfand, ISDS, Duke University (with J. Duan and G. Puggioni)
Lecture II-2: Probability Review
Introduction to Regression Analysis, Chapter 13,
Robin McDougall, Ed Waller and Scott Nokleby Faculties of Engineering & Applied Science and Energy Systems & Nuclear Science 1.
Hydrologic Statistics
©2003/04 Alessandro Bogliolo Background Information theory Probability theory Algorithms.
1 Bayesian methods for parameter estimation and data assimilation with crop models Part 2: Likelihood function and prior distribution David Makowski and.
Calibration Guidelines 1. Start simple, add complexity carefully 2. Use a broad range of information 3. Be well-posed & be comprehensive 4. Include diverse.
Computational Stochastic Optimization: Bridging communities October 25, 2012 Warren Powell CASTLE Laboratory Princeton University
Particle Filtering in Network Tomography
Machine Learning1 Machine Learning: Summary Greg Grudic CSCI-4830.
CRESCENDO Full virtuality in design and product development within the extended enterprise Naples, 28 Nov
Hydrologic Modeling: Verification, Validation, Calibration, and Sensitivity Analysis Fritz R. Fiedler, P.E., Ph.D.
Applications of Bayesian sensitivity and uncertainty analysis to the statistical analysis of computer simulators for carbon dynamics Marc Kennedy Clive.
Modeling & Simulation: An Introduction Some slides in this presentation have been copyrighted to Dr. Amr Elmougy.
Stochastic Algorithms Some of the fastest known algorithms for certain tasks rely on chance Stochastic/Randomized Algorithms Two common variations – Monte.
Eawag: Swiss Federal Institute of Aquatic Science and Technology Problems of Inference and Uncertainty Estimation in Hydrologic Modelling Peter Reichert.
Time Series Data Analysis - I Yaji Sripada. Dept. of Computing Science, University of Aberdeen2 In this lecture you learn What are Time Series? How to.
Geographic Information Science
Multiple Regression Petter Mostad Review: Simple linear regression We define a model where are independent (normally distributed) with equal.
Getting started with GEM-SA. GEM-SA course - session 32 This talk Starting GEM-SA program Creating input and output files Explanation of the menus, toolbars,
1 11 Simple Linear Regression and Correlation 11-1 Empirical Models 11-2 Simple Linear Regression 11-3 Properties of the Least Squares Estimators 11-4.
Randomized Algorithms for Bayesian Hierarchical Clustering
Stress constrained optimization using X-FEM and Level Set Description
Model Selection and Validation. Model-Building Process 1. Data collection and preparation 2. Reduction of explanatory or predictor variables (for exploratory.
ECE-7000: Nonlinear Dynamical Systems Overfitting and model costs Overfitting  The more free parameters a model has, the better it can be adapted.
CHAPTER 17 O PTIMAL D ESIGN FOR E XPERIMENTAL I NPUTS Organization of chapter in ISSO –Background Motivation Finite sample and asymptotic (continuous)
- 1 - Overall procedure of validation Calibration Validation Figure 12.4 Validation, calibration, and prediction (Oberkampf and Barone, 2004 ). Model accuracy.
Eawag: Swiss Federal Institute of Aquatic Science and Technology Mechanism-Based Emulation of Dynamic Simulation Models – Concept and Application in Hydrology.
- 1 - Calibration with discrepancy Major references –Calibration lecture is not in the book. –Kennedy, Marc C., and Anthony O'Hagan. "Bayesian calibration.
September 28, 2000 Improved Simultaneous Data Reconciliation, Bias Detection and Identification Using Mixed Integer Optimization Methods Presented by:
The Unscented Particle Filter 2000/09/29 이 시은. Introduction Filtering –estimate the states(parameters or hidden variable) as a set of observations becomes.
Monte Carlo Linear Algebra Techniques and Their Parallelization Ashok Srinivasan Computer Science Florida State University
Bayesian Modelling Harry R. Erwin, PhD School of Computing and Technology University of Sunderland.
Classification and Regression Trees
1 Tom Edgar’s Contribution to Model Reduction as an introduction to Global Sensitivity Analysis Procedure Accounting for Effect of Available Experimental.
Eawag: Swiss Federal Institute of Aquatic Science and Technology Analyzing input and structural uncertainty of a hydrological model with stochastic, time-dependent.
Kevin Stevenson AST 4762/5765. What is MCMC?  Random sampling algorithm  Estimates model parameters and their uncertainty  Only samples regions of.
Eawag: Swiss Federal Institute of Aquatic Science and Technology Analyzing input and structural uncertainty of deterministic models with stochastic, time-dependent.
Institute of Statistics and Decision Sciences In Defense of a Dissertation Submitted for the Degree of Doctor of Philosophy 26 July 2005 Regression Model.
CPH Dr. Charnigo Chap. 11 Notes Figure 11.2 provides a diagram which shows, at a glance, what a neural network does. Inputs X 1, X 2,.., X P are.
Ch 1. Introduction Pattern Recognition and Machine Learning, C. M. Bishop, Updated by J.-H. Eom (2 nd round revision) Summarized by K.-I.
Bayesian analysis of a conceptual transpiration model with a comparison of canopy conductance sub-models Sudeep Samanta Department of Forest Ecology and.
CSCI 5822 Probabilistic Models of Human and Machine Learning
L. Isella, A. Karvounaraki (JRC) D. Karlis (AUEB)
Akio Utsugi National Institute of Bioscience and Human-technology,
Where did we stop? The Bayes decision rule guarantees an optimal classification… … But it requires the knowledge of P(ci|x) (or p(x|ci) and P(ci)) We.
Ch 3. Linear Models for Regression (2/2) Pattern Recognition and Machine Learning, C. M. Bishop, Previously summarized by Yung-Kyun Noh Updated.
Bayesian Model Selection and Averaging
Presentation transcript:

Eawag: Swiss Federal Institute of Aquatic Science and Technology Use of time-dependent parameters for improvement and uncertainty estimation of dynamic models Peter Reichert Eawag Dübendorf and ETH Zürich

SAMSI meeting Nov. 6, 2006 Contents Motivation References Approach Preliminary Results Problems/ Challenges  Motivation  References  Approach  Concept  Implementation  Preliminary Results for a Simple Hydrologic Model  Problems / Challenges

SAMSI meeting Nov. 6, 2006 Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Motivation Fundamental Objectives:  Improve understanding of mechanisms governing the behaviour of the system described by the model.  Estimate realistic uncertainty bounds / decrease the width of uncertainty bounds of model predictions. Technical Objectives:  Improve the formulation of the deterministic model component.  Make the stochastic component of the model more realistic. Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Motivation Achieve these objectives by: 1.Improving the input error model. 2.Allowing model parameters to vary (e.g. in time) to address model structure error. 3.Improving the output error model (by addressing bias explicitly). In particular:  Search for statistical model components that cannot be rejected by the data.  Try to „explain“ the bias by input and/or model structure error („trace“ the causes of the bias). Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 References Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 References The idea of using time-dependent parameters for model structure deficit evaluation is very old (e.g. review by Beck, 1987). Our work applies this idea to continuous time models and provides algorithms to apply it to nonlinear dynamic systems. Motivation References Approach Preliminary Results Problems/ Challenges This talk is based on:  Brun, PhD dissertation, 2002: First trials with filtering algorithm.  Buser, Masters thesis, 2003: Smoothing, MCMC algorithm.  Tomassini, Reichert, Künsch, Buser, Borsuk, 2007: Estimation of process parameters, cross-validation.

SAMSI meeting Nov. 6, 2006 Approach Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Notation (according to Bayarri et al. 2005) Input: Field data = reality plus measurement error: Motivation References Approach Preliminary Results Problems/ Challenges An ideal model describes reality: A realistic model approximates an ideal model: Output: Field data = reality plus measurement error: All terms together: input error meas. error effect of model structure error

SAMSI meeting Nov. 6, 2006 Problem Motivation References Approach Preliminary Results Problems/ Challenges input error meas. error effect of model structure error Problem: The bias term describes the effect, but not the cause of the model structure error. This leads to a satisfying statistical description of the past, but is hard to extra- polate into the future. For uncertainty reduction and extrapolation it would be better to reduce the bias by improving the mechanistic description of the system. In particular, trends must be described by the model, not by the bias term. How can statistical procedures support this?

SAMSI meeting Nov. 6, 2006 Concept Motivation References Approach Preliminary Results Problems/ Challenges input error meas. error effect of model structure error Concept: 1.Allow model parameters to vary. Add parameters where appropriate (input, output). 2.Try to reduce the bias by finding an adequate behaviour of these parameters. 3.Explore dependency of parameter variability on external or model variables. If successful (from a statistical and physical point of view), modify the model structure to reflect this dependency. 4.Redo the analysis with improved model structure and reduced bias.

SAMSI meeting Nov. 6, 2006 Use for dynamic models Motivation References Approach Preliminary Results Problems/ Challenges input error meas. error effect of model structure error  x t : Correction accounting for input error.   t : Model-internal correction of model structure error.  y t : Model-external correction for remaining effect of model structure error. In the ideal case, this error could be neglected as it would be accounted for by the internal correction. Formulation for time dependent models:

SAMSI meeting Nov. 6, 2006 Approach Motivation References Approach Preliminary Results Problems/ Challenges 1.Fit model with constant parameters, identify presence of bias. If bias exists: 2.Identify, separately or jointly, time-dependent  input variation (  x t )  parameter variation (   t )  output variation (  y t ) 3.Identify dependences of time-dependent parameters on external or model variables. 4.Improve the model structure by deterministic or sto- chastic elements (according to statistical and physi- cal considerations), try to avoid output error (  y t ). 5.Use the extended model for understanding and prediction.

SAMSI meeting Nov. 6, 2006 Implementation Motivation References Approach Preliminary Results Problems/ Challenges This has the advantage that we can use the analytical solution: The time dependent parameter is modelled by a mean-reverting Ornstein Uhlenbeck process: or, after reparameterization:

SAMSI meeting Nov. 6, 2006 Implementation Motivation References Approach Preliminary Results Problems/ Challenges We combine the estimation of  constant model parameters, , with  state estimation of the time-dependent parameter(s),  t, and with  the estimation of (constant) parameters of the Ornstein-Uhlenbeck process(es) of the time dependent parameter(s),  =( , ,,  to ).

SAMSI meeting Nov. 6, 2006 Conceptual Framework Motivation References Approach Preliminary Results Problems/ Challenges Model extended by input- and output- parameter and measurement error Original deterministic model

SAMSI meeting Nov. 6, 2006 Simplified Framework Motivation References Approach Preliminary Results Problems/ Challenges Simplifications:  Omit representation of given measured input, x F.  Add parameter to input to represent input uncertainty by parameter uncertainty.  Add parameter to output to represent output uncertainty by parameter uncertainty.

SAMSI meeting Nov. 6, 2006 Numerical Implementation (1) Motivation References Approach Preliminary Results Problems/ Challenges Gibbs sampling for the three different types of parameters. Conditional distributions: Ornstein-Uhlenbeck process (cheap) simulation model (expensive) Ornstein-Uhlenbeck process (cheap)

SAMSI meeting Nov. 6, 2006 Numerical Implementation (2) Motivation References Approach Preliminary Results Problems/ Challenges Metropolis-Hastings sampling for each type of parameter: Multivariate normal jump distributions for the parameters  and . This requires one simulation to be performed per suggested new value of . The discretized Ornstein-Uhlenbeck parameter,  t, is split into subintervals for which OU-process realizations conditional on initial and end points are sampled. This requires the number of subintervals simulations per complete new time series of  t.

SAMSI meeting Nov. 6, 2006 Estimation of Hyperparameters by Cross - Validation Motivation References Approach Preliminary Results Problems/ Challenges Due to identifiability problems we selected the hyperparameters, , in a previous application (Tomassini et al., 2006) alternatively by cross- validation:

SAMSI meeting Nov. 6, 2006 Preliminary Results Preliminary Results for a Simple Hydrologic Model Motivation References Approach Preliminary Results Problems/ Challenges  Model  Model Application  Preliminary Results (based on Markov chains of insufficient length)

SAMSI meeting Nov. 6, 2006 Model A Simple Hydrologic Watershed Model (1): Kuczera et al Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Model A Simple Hydrologic Watershed Model (2): Kuczera et al Motivation References Approach Preliminary Results Problems/ Challenges A B 7 model parameters 3 initial conditions 1 standard dev. of meas. err. 3 „modification parameters“ C

SAMSI meeting Nov. 6, 2006 Model A Simple Hydrologic Watershed Model (3): Kuczera et al Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Model Application Model application:  Data set of Abercrombie watershed, New South Wales, Australia (2770 km 2 ), kindly provided by George Kuczera (Kuczera et al. 2006).  Box-Cox transformation applied to model and data to decrease heteroscedasticity of residuals.  Step function input to account for input data in the form of daily sums of precipitation and potential evapotranspiration.  Daily averaged output to account for output data in the form of daily average discharge. Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Model Application Prior distribution: Estimation of constant parameters: Independent uniform distributions for the loga- rithms of all parameters (7+3+1=11), keeping correction factors (f rain, f pet ) equal to unity and bias (b Q ) equal to zero. Estimation of time-dependent parameters: Ornstein-Uhlenbeck process applied to log of the parameter (with the exception of b Q ). Hyper-parameters:  = 5d,  fixed, only estimation of initial value and mean (0 for f rain, f pet, b Q ). Constant parameters as above. Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Preliminary Results (MC of insufficient length) Posterior marginals: Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Preliminary Results (MC of insufficient length) Max. post. simulation with constant parameters: Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Preliminary Results (MC of insufficient length) Residuals of max. post. sim. with const. pars.: Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Preliminary Results (MC of insufficient length) Residual analysis, max. post., constant parameters Motivation References Approach Preliminary Results Problems/ Challenges Residual analysis, max. post., q_gw_max time-dependent

SAMSI meeting Nov. 6, 2006 Preliminary Results (MC of insufficient length) Residual analysis, max. post., s_F time-dependent Motivation References Approach Preliminary Results Problems/ Challenges Residual analysis, max. post., f_rain time-dependent

SAMSI meeting Nov. 6, 2006 Preliminary Results (MC of insufficient length) Time-dependent parameters Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Problems / Challenges (= Working Group Opportunities) Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Problems / Challenges 1.Other formulations of time-dependent parameters? 2.Dependence on other factors than time. 3.How to estimate hyperparameters? (Reduction in correlation time always improves the fit.) 4.How to avoid modelling physical processes with the bias term? 5.Learn from more applications. 6.Compare results with methodology by Bayarri et al. (2005). Combine/extend the two methodologies? 7.? Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Problems / Challenges (= Working Group Opportunities) Discussion slides from talk at Oct. 16. Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Problems / Challenges Research Questions / Options for Projects (1) 1.Compare results when making different model parameters stochastic and time-dependent. (Ongoing with a postdoc in Switzerland extending earlier work with continuous-time stochastic parameters.) 2.Develop a better statistical description of rainfall uncertainty. (Option for a collaboration with climate/weather working groups.) 3.Explore alternative options for making parameters time-dependent. (Suggestions so far: storm-dependent parameters, time- dependent parameter as an Ornstein-Uhlenbeck process.) Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Problems / Challenges Research Questions / Options for Projects (2) 4.Investigate how to learn from state estimation of stochastic hydrological models. (Can the pattern of state adaptations lead to insights of model structure deficits or input errors?) 5.Develop uncertainty estimates when using multi- objective optimization. (How to use information on Pareto set for uncertainty estimation of parameters and results?) 6.Analyse differences in results of suggested approaches when using different models. (Is there a generic behaviour of different techniques when they are applied to different models/data sets?) Motivation References Approach Preliminary Results Problems/ Challenges

SAMSI meeting Nov. 6, 2006 Problems / Challenges Research Questions / Options for Projects (3) 7.Improve the efficientcy of posterior maximisation and posterior sampling. (Efficiency becomes important when having complex watershed models in mind. Efficient global optimizers and sampling from multi-modal posterior distributions becomes then important.) 8.More questions will come up during discussions. Motivation References Approach Preliminary Results Problems/ Challenges

Eawag: Swiss Federal Institute of Aquatic Science and Technology Thank you for your attention