Statistical Postprocessing of Ensemble Forecasts

• Main Authors: Vannitsem, Stéphane, Wilks, Daniel S, Messner, Jakob
• Format: eBook
• Language: English
• Published: Chantilly Elsevier 2018
• Edition: 1
• Subjects: Forecasting; Weather forecasting-Popular works
• ISBN: 9780128123720; 0128123729
• DOI: 10.1016/B978-0-12-812372-0.09993-3

This volume brings together chapters contributed by international subject-matter experts describing the current state of the art in the statistical postprocessing of ensemble forecasts. It illustrates the use of these methods in several important applications including weather, hydrological and clim...

Table of Contents:

• Front Cover -- Statistical Postprocessing of Ensemble Forecasts -- Copyright -- Contents -- Contributors -- Preface -- Chapter 1: Uncertain Forecasts From Deterministic Dynamics -- 1.1. Sensitivity to Initial Conditions, or ``Chaos´´ -- 1.2. Uncertainty and Probability in ``Deterministic´´ Predictions -- 1.3. Ensemble Forecasting -- 1.4. Postprocessing Individual Dynamical Forecasts -- 1.5. Postprocessing Ensemble Forecasts: Overview of This Book -- References -- Chapter 2: Ensemble Forecasting and the Need for Calibration -- 2.1. The Dynamical Weather Prediction Problem -- 2.1.1. Historical Background -- 2.1.2. Observations -- 2.1.3. The Equations of Motion for the Atmosphere -- 2.1.4. Computation of the Initial Conditions (Analysis) -- 2.2. The Chaotic Nature of the Atmosphere -- 2.3. From Single to Ensemble Forecasts -- 2.3.1. Forecast Reliability and Accuracy -- 2.3.2. Are Ensemble Forecasts More Valuable than a Single Forecast? -- 2.4. Sources of Forecast Errors -- 2.5. Characteristics of the Operational Global Ensemble Systems -- 2.6. The Value of a Reforecast Suite -- 2.7. A Look Into the Future -- 2.8. Summary: The Key Messages of This Chapter -- References -- Chapter 3: Univariate Ensemble Postprocessing -- 3.1. Introduction -- 3.2. Nonhomogeneous Regressions, and Other Regression Methods -- 3.2.1. Nonhomogeneous Gaussian Regression (NGR) -- 3.2.2. Nonhomogeneous Regressions With More Flexible Predictive Distributions -- 3.2.3. Truncated Nonhomogeneous Regressions -- 3.2.4. Censored Nonhomogeneous Regressions -- 3.2.5. Logistic Regression -- 3.3. Bayesian Model Averaging, and Other ``Ensemble Dressing´´ Methods -- 3.3.1. Bayesian Model Averaging (BMA) -- 3.3.2. Other Ensemble Dressing Methods -- 3.4. Fully Bayesian Ensemble Postprocessing Approaches -- 3.5. Nonparametric Ensemble Postprocessing Methods
• Chapter 6: Verification: Assessment of Calibration and Accuracy -- 6.1. Introduction -- 6.2. Calibration -- 6.2.1. Univariate Calibration -- 6.2.2. Multivariate Calibration -- 6.2.3. Example: Comparing Multivariate Ranking Methods -- 6.3. Accuracy -- 6.3.1. Univariate Assessment -- 6.3.2. Simulation Study: Comparing Univariate Scoring Rules -- 6.3.3. Assessing Extreme Events -- 6.3.4. Example: Proper and Nonproper Verification of Extremes -- 6.3.5. Multivariate Assessment -- 6.3.6. Divergence Functions -- 6.3.7. Testing Equal Predictive Performance -- 6.4. Understanding Model Performance -- 6.5. Summary -- References -- Chapter 7: Practical Aspects of Statistical Postprocessing -- 7.1. Introduction -- 7.2. The Bias-Variance Tradeoff -- 7.3. Training-Data Issues for Statistical Postprocessing -- 7.3.1. Challenges in Developing Ideal Predictor Training Data -- 7.3.2. Challenges in Gathering/Developing Ideal Predictand Training Data -- 7.4. Proposed Remedies for Practical Issues in Statistical Postprocessing -- 7.4.1. Improving the Approaches for Generating Reforecasts -- 7.4.2. Circumventing Common Challenges Posed by Shorter Training Data Sets -- 7.4.3. Substandard Analysis Data -- 7.5. Case Study: Postprocessing to Generate High Resolution Probability-of-Precipitation From Global Multimodel Ensembles -- 7.6. Collaborating on Software and Test Data to Accelerate Postprocessing Improvement -- 7.7. Recommendations and Conclusions -- References -- Further Reading -- Chapter 8: Applications of Postprocessing for Hydrological Forecasts -- 8.1. Introduction -- 8.2. Univariate Hydrological Postprocessing -- 8.2.1. Skewness and the Assumption of Gaussianity -- 8.2.2. Univariate Hydrological Ensemble Postprocessing -- 8.2.3. Postprocessing of Hydrological Forecasts Versus Postprocessing of Meteorological Input -- 8.3. Multivariate Hydrological Postprocessing
• 3.5.1. Rank Histogram Recalibration -- 3.5.2. Quantile Regression -- 3.5.3. Ensemble Dressing -- 3.5.4. Individual Ensemble-Member Adjustments -- 3.5.5. ``Statistical Learning´´ Methods for Ensemble Postprocessing -- 3.6. Comparisons Among Methods -- References -- Chapter 4: Ensemble Postprocessing Methods Incorporating Dependence Structures -- 4.1. Introduction -- 4.2. Dependence Modeling Via Copulas -- 4.2.1. Copulas and Sklar's Theorem -- 4.2.2. Parametric, in Particular Gaussian, Copulas -- 4.2.3. Empirical Copulas -- 4.3. Parametric Multivariate Approaches -- 4.3.1. Intervariable Dependencies -- 4.3.2. Spatial Dependencies -- 4.3.3. Temporal Dependencies -- 4.4. Nonparametric Multivariate Approaches -- 4.4.1. Empirical Copula-Based Ensemble Postprocessing -- 4.4.2. Ensemble Copula Coupling (ECC) -- 4.4.3. Schaake Shuffle-Based Approaches -- 4.5. Univariate Approaches Accounting for Dependencies -- 4.5.1. Spatial Dependencies -- 4.5.2. Temporal Dependencies -- 4.6. Discussion -- References -- Chapter 5: Postprocessing for Extreme Events -- 5.1. Introduction -- 5.2. Extreme-Value Theory -- 5.2.1. Generalized Extreme-Value Distribution -- 5.2.2. Peak-Over-Threshold Approach -- 5.2.3. Nonstationary Extremes -- 5.3. Postprocessing of Univariate Extremes: Precipitation -- 5.3.1. Data and Ensemble Forecasts -- 5.3.2. Approaches and Verification -- 5.3.3. Variable Selection -- 5.3.4. Comparison of Postprocessing Approaches -- 5.4. Extreme-Value Theory for Multivariate and Spatial Extremes -- 5.4.1. Extremal Dependence and Multivariate Extreme-Value Distributions -- 5.4.2. Spatial Max-Stable Processes -- 5.5. Postprocessing for Spatial Extremes: Wind Gusts -- 5.5.1. Postprocessing for Marginal Distribution -- 5.5.2. The Spatial Dependence Structure -- 5.6. Conclusions -- 5.7. Appendix -- References
• 8.3.1. Temporal Dependencies -- 8.3.2. Spatial Dependencies -- 8.3.3. Spatio-Temporal Dependencies -- 8.4. Outlook -- References -- Chapter 9: Application of Postprocessing for Renewable Energy -- 9.1. Introduction -- 9.2. Preliminaries: Relevant Forecasting Products and Notation -- 9.3. Conversion of Meteorological Variables to Power -- 9.3.1. Data and Empirical Features -- 9.3.2. Local Polynomial Regression as a Basis -- 9.3.3. Time-Varying Estimation to Accommodate Nonstationarity -- 9.3.4. From Least Squares Estimation to Fitting of Principal Curves -- 9.4. Calibrated Predictive Densities of Power Generation -- 9.4.1. Calibration Prior to Conversion -- Kernel dressing of wind speed -- Inverse power curve transformation -- 9.4.2. Calibration After Conversion -- Nonhomogeneous regression of wind power -- Adaptive kernel dressing -- 9.4.3. Direct Calibration of Wind Power -- 9.5. Conclusions and Perspectives -- 9.6. Appendix: Simulated Data for the Conversion of Wind to Power -- References -- Chapter 10: Postprocessing of Long-Range Forecasts -- 10.1. Introduction -- 10.2. Challenges of Long-Range Forecasts -- 10.3. A Statistical Framework for Postprocessing -- 10.3.1. Statistical Hypotheses -- 10.3.2. Reliability of Long-Range Forecasts -- 10.4. Multimodel Combination or Consolidation -- 10.5. The Use of Multimodels for Probabilistic Forecasts -- 10.6. Drift and Trend Correction Techniques -- 10.7. Ensemble Postprocessing Techniques -- 10.8. Application of Postprocessing in an Idealized Model Setting -- 10.8.1. Experimental Setup -- 10.8.2. Postprocessing Single-Model Ensembles -- 10.8.3. Multimodel Ensemble Forecasts -- 10.9. Application Using an Operational Long-Range Forecasting System -- 10.10. Conclusions -- 10.11. Appendix: The Idealized Model -- Acknowledgments -- References -- Chapter 11: Ensemble Postprocessing With R -- 11.1. Introduction
• 11.2. Deterministic Postprocessing -- 11.2.1. Data -- 11.2.2. Model Fitting -- 11.2.3. Prediction -- 11.2.4. Verification -- 11.3. Univariate Postprocessing of Temperature -- 11.3.1. Data -- 11.3.2. Model Fitting -- Nonhomogeneous Gaussian regression -- BMA and other ensemble dressing approaches -- 11.3.3. Prediction -- 11.3.4. Verification -- 11.4. Postprocessing of Precipitation -- 11.4.1. Data -- 11.4.2. Model Fitting -- Nonhomogeneous regression -- Bayesian model averaging -- Logistic regression -- 11.4.3. Prediction -- 11.4.4. Verification -- 11.5. Multivariate Postprocessing of Temperature and Precipitation -- 11.5.1. Data -- 11.5.2. Model Fitting -- 11.5.3. Prediction -- 11.5.4. Verification -- 11.6. Summary and Discussion -- 11.7. Appendices -- 11.7.1. Appendix A: Code for Some Functions Used in This Chapter -- 11.7.2. Appendix B: Available R Packages for Ensemble Postprocessing -- Available data sets and data processing -- Ensemble postprocessing models -- Verification -- References -- Author Index -- Subject Index -- Back Cover