## Calibrated probabilistic forecasting at the Stateline wind energy center: The regime-switching space-time (RST) method (2004)

### Cached

### Download Links

Venue: | Journal of the American Statistical Association |

Citations: | 21 - 10 self |

### BibTeX

@TECHREPORT{Gneiting04calibratedprobabilistic,

author = {Tilmann Gneiting and Kristin Larson and Kenneth Westrick and Marc G. Genton and Eric Aldrich},

title = {Calibrated probabilistic forecasting at the Stateline wind energy center: The regime-switching space-time (RST) method},

institution = {Journal of the American Statistical Association},

year = {2004}

}

### Years of Citing Articles

### OpenURL

### Abstract

With the global proliferation of wind power, accurate short-term forecasts of wind resources at wind energy sites are becoming paramount. Regime-switching space-time (RST) models merge meteorological and statistical expertise to obtain accurate and calibrated, fully probabilistic forecasts of wind speed and wind power. The model formulation is parsimonious, yet takes account of all the salient features of wind speed: alternating atmospheric regimes, temporal and spatial correlation, diurnal and seasonal non-stationarity, conditional heteroscedasticity, and non-Gaussianity. The RST method identifies forecast regimes at the wind energy site and fits a conditional predictive model for each regime. Geographically dispersed meteorological observations in the vicinity of the wind farm are used as off-site predictors. The RST technique was applied to 2-hour ahead forecasts of hourly average wind speed at the Stateline wind farm in the US Pacific Northwest. In July 2003, for instance, the RST forecasts had root-mean-square error (RMSE) 28.6 % less than the persistence forecasts. For each month in the test period, the RST forecasts had lower RMSE than forecasts using state-of-the-art vector time series techniques. The RST method provides probabilistic forecasts in the form of

### Citations

1998 | Time-Series Analysis: Forecasting and Control - Box, Jenkins - 1970 |

617 |
Generalized Autoregressive Conditional Heteroscedasticity
- Bollerslev
- 1986
(Show Context)
Citation Context ...l of order at most 4. The conditionally heteroscedastic AR-N-CH technique, furthermore, uses the S+FinMetrics function garch to fit a GARCH(1,1) model for the conditional variance of the innovations (=-=Bollerslev 1986-=-; Zivot and Wang 2003, Chapter 7). The AR-D and AR-D-CH methods estimate and extract a diurnal trend component of the form (13) and then proceeds as above. To obtain Gaussian predictive distribution f... |

334 | Regularized discriminant analysis
- Friedman
- 1987
(Show Context)
Citation Context ...ormance. 14sWe turn to the estimation of the conditional statistical models. The literature argues that maximum likelihood plug-in estimates may be suboptimal when the goal is prediction (Copas 1983; =-=Friedman 1989-=-). Gneiting, Westveld, Raftery and Goldman (2004) proposed the novel technique of minimum continuous ranked probability score (CRPS) estimation for estimating predictive distributions. In minimum CRPS... |

332 | Estimation of relationships for limited dependent variables. Econometrica - Tobin - 1958 |

289 | Evaluating density forecasts, with applications to financial risk management - Diebold, Gunther, et al. - 1998 |

279 |
Numerical recipes in fortran: The art of scientific computing
- PRESS, Teukolsky, et al.
- 1992
(Show Context)
Citation Context ...me, for instance, we find the minimum of the CRPS value as a function of the parameters in (10) and (11). This needs to be done numerically, and we use the Broyden-Fletcher-Goldfarb-Shanno algorithm (=-=Press et al. 1992-=-, Section 10.7) as implemented in the R language and environment (www.cran.r-project.org). The algorithm is iterative, and starting values based on past experience usually give good solutions. Converg... |

220 | Remarks on a multivariate transformation - Rosenblatt - 1952 |

187 | Statistical theory: The prequential approach - Dawid - 1984 |

174 | Strictly proper scoring rules, prediction, and estimation
- Gneiting, Raftery
- 2007
(Show Context)
Citation Context ...arameters, and minimize that function with respect to the parameter values. This technique is tailored to probabilistic forecasting and can be interpreted within the framework of robust M-estimation (=-=Gneiting and Raftery 2004-=-). Here the predictive distributions are cut-off normal and we minimize the CRPS value (2), where each term is computed from (5). For RST-D-CH forecasts in the easterly regime, for instance, we find t... |

170 |
Rational decisions
- Good
- 1952
(Show Context)
Citation Context ... score is proper in the sense that the forecaster maximizes the expected score for an observation drawn from F if she issues the probabilistic forecast F , rather than G �= F . The logarithmic score (=-=Good 1952-=-; Bernardo 1979; Roulston and Smith 2002) is a proper scoring rule, too, but it is highly sensitive to outliers and difficult to interpret for mixed discrete-continuous predictive distributions. There... |

118 | Space-time modelling with long-memory dependence : assessing Ireland’s wind power resource (with discussion - Haslett, Raftery - 1989 |

92 |
Time Series: Theory and Methods, second edition
- Brockwell, Davis
- 1991
(Show Context)
Citation Context ...t−1 + a3Kt + a4Kt−1 + a5Gt, (9) where Vt, Kt, and Gt denote the wind speed at Vansycle, Kennewick and Goodnoe Hills at time t, respectively. This is akin to a vector autoregressive time series model (=-=Brockwell and Davis 1991-=-, Chapter 11; de Luna and Genton 2003) or an autoregressive distributed lag scheme (Zivot and Wang 2003, Section 6.4). However, the predictive distribution is cut-off Gaussian, and the model applies i... |

90 | Nonlinear gated experts for time series: discovering regimes and avoiding overfitting - Weigend, Mangeas, et al. - 1995 |

78 | 2005: Using Bayesian Model Averaging to Calibrate Forecast Ensembles - Raftery, Gneiting, et al. |

76 |
Expected information as expected utility
- Bernardo
- 1979
(Show Context)
Citation Context ...roper in the sense that the forecaster maximizes the expected score for an observation drawn from F if she issues the probabilistic forecast F , rather than G �= F . The logarithmic score (Good 1952; =-=Bernardo 1979-=-; Roulston and Smith 2002) is a proper scoring rule, too, but it is highly sensitive to outliers and difficult to interpret for mixed discrete-continuous predictive distributions. Therefore, we assess... |

57 | A nonhomogeneous hidden Markov model for precipitation occurrence - Hughes, Guttorp, et al. - 1999 |

53 | Forecast uncertainties in macroeconometric modelling: an application to the UK economy - Garatt, Lee, et al. - 2003 |

48 | Nonseparable, stationary covariance functions for space-time data - Gneiting - 2002 |

37 | Evaluating probabilistic forecasts using information theory
- Roulston, Smith
- 2002
(Show Context)
Citation Context ...nse that the forecaster maximizes the expected score for an observation drawn from F if she issues the probabilistic forecast F , rather than G �= F . The logarithmic score (Good 1952; Bernardo 1979; =-=Roulston and Smith 2002-=-) is a proper scoring rule, too, but it is highly sensitive to outliers and difficult to interpret for mixed discrete-continuous predictive distributions. Therefore, we assess probabilistic forecasts ... |

36 | Calibrated probabilistic forecasting using ensemble model output statistics and minimum CRPS estimation - Gneiting, Raftery, et al. - 2005 |

35 | A class of stochastic models for relating synoptic atmospheric patterns to regional hydrologic phenomena - Hughes, Guttorp - 1994 |

34 | Space-time model for daily rainfall using atmospheric circulation patterns. Water Resources Res 28: 1247–1259 Capel Molina JJ (2000) El clima de la penı́nsula Iberica - Bardossy, EJ - 1992 |

33 |
Decomposition of the Continuous Ranked Probability Score for ensemble prediction systems
- Hersbach
(Show Context)
Citation Context ...ed probability score is defined as crps(F, x) = � ∞ −∞ , (F (y) − 1(y ≥ x)) 2 dy, (1) 4swhich equals the integral of the Brier scores for binary probabilistic forecasts at all real-valued thresholds (=-=Hersbach 2000-=-). Gneiting and Raftery (2004) used a result of Székely (2003) to show that crps(F, x) = EF |X − x| − 1 2 EF � � ′ X − X � � , where X and X ′ are independent copies of a random variable with distribu... |

30 |
The State-Of-The-Art in Short-Term Prediction of Wind power. A literature overview. Deliverable 1.1 of Anemos Project
- Giebel, Kariniotakis, et al.
- 2003
(Show Context)
Citation Context ... to the series of hourly average wind speed at Goodnoe Hills, fit and extracted a diurnal trend component, and modeled the residual component as an AR process. This approach has found widespread use (=-=Giebel 2003-=-). For the forecasts at Vansycle, we experimented with the square root transform of Brown et al. (1984) and the more general transform proposed by Allcroft and Glasbey (2003), but found forecasts base... |

28 |
Regression, prediction and shrinkage
- Copas
- 1983
(Show Context)
Citation Context ...dictive performance. 14sWe turn to the estimation of the conditional statistical models. The literature argues that maximum likelihood plug-in estimates may be suboptimal when the goal is prediction (=-=Copas 1983-=-; Friedman 1989). Gneiting, Westveld, Raftery and Goldman (2004) proposed the novel technique of minimum continuous ranked probability score (CRPS) estimation for estimating predictive distributions. ... |

27 |
2002: The economic value of ensemble forecasts as a tool for risk assessment: from days to decades
- Palmer
(Show Context)
Citation Context ...c in nature, taking the form of probability distributions over future events. Indeed, over the past two decades probabilistic forecasts have become routine in applications such as weather prediction (=-=Palmer 2002-=-; Gel, Raftery and Gneiting 2004) and macroeconomic forecasting (Garrat, Lee, Pesaran and Shin 2003). A probabilistic forecast for a real-valued quantity, such as wind speed or wind power, takes the f... |

23 | A latent Gaussian Markov random-field model for spatiotemporal rainfall disaggregation
- Allcroft, Glasbey
- 2003
(Show Context)
Citation Context ...d the associated tobit model has been widely used in econometric applications since. Cut-off normal distributions can also be interpreted in terms of latent Gaussian variables (Sansó and Guenni 2000; =-=Allcroft and Glasbey 2003-=-). 3 Data We introduce and describe the data on which our case study is based. 3.1 The meteorological towers at Vansycle, Kennewick and Goodnoe Hills The meteorological data used hereinafter were coll... |

21 | Time series models to simulate and forecast wind speed and wind - Brown, Katz, et al. - 1984 |

20 | Calibrated probabilistic mesoscale weather field forecasting: The geostatistical output perturbation (GOP) method - Gel, Raftery, et al. - 2004 |

20 | Quality assurance procedures in the Oklahoma Mesonet - Shafer, Fiebrich, et al. - 2000 |

18 | Ozone Exposure and Population Density in Harris County - Carrol, Chen, et al. - 1997 |

16 |
A nonstationary multisite model for rainfall
- Sansó, Guenni
(Show Context)
Citation Context ...ctive distributions, and the associated tobit model has been widely used in econometric applications since. Cut-off normal distributions can also be interpreted in terms of latent Gaussian variables (=-=Sansó and Guenni 2000-=-; Allcroft and Glasbey 2003). 3 Data We introduce and describe the data on which our case study is based. 3.1 The meteorological towers at Vansycle, Kennewick and Goodnoe Hills The meteorological data... |

13 | Wind speed and power forecasting based on spatial correlation models - Alexiadis, Dokopoulos, et al. - 1999 |

11 | Probabilistic wind power forecasts using local quantile regression - Bremnes - 2004 |

10 | Predictive spatio-temporal models for spatially sparse environmental data - LUNA, GENTON - 2005 |

10 | 2001: The impact of unique meteorological phenomena detected by the Oklahoma Mesonet and ARS Micronet on automated quality control - Fiebrich, Crawford |

8 | Weather Forecasting for Weather Derivatives, working paper (2002 - Diebold, Campbell |

8 | ǫ-statistics: The energy of statistical samples - Székely - 2003 |

7 | On-line assessment of prediction risk for wind power production forecasts.Wind Energy 2004; 7:119–132 - Pinson, Kariniotakis |

6 | Verifying probabilistic forecasts: Calibration and sharpness - Gneiting, Raftery, et al. - 2003 |

6 | Modeling transport effects on ground-level ozone using a non-stationary space-time model - Huang, Hsu - 2004 |

5 |
2002: Columbia Gorge gap flow: Insights from observational analysis and ultra-High-Resolution Simulation
- Sharp, Mass
(Show Context)
Citation Context ...ts is known as gap flow. The Columbia Gorge gap flow plays a profound role in defining the weather and climate within and near the Gorge, which is one of the windiest places in the Pacific Northwest (=-=Sharp and Mass 2002-=-, 200x). When the surface pressure is higher to the west, the flow within the gorge is normally westerly; conversely, when there is higher pressure to the east, the wind is usually easterly. Westerly ... |

5 | Output power correlation between adjacent wind power plants - Wan, Milligan, et al. - 2003 |

4 | Modeling the variability of Sydney Harbor wind measurments - Cripps, Dunsmuir |

4 | 2004: Neural network classifiers for local wind prediction - Kretzschmar, Eckert, et al. |

4 | 2003: Using mediumrange weather forecasts to improve the value of wind energy production. Renewable Energy - Roulston, Kaplan, et al. |

4 | A dimension reduction approach to spacetime Kalman filtering. Biometrika - Wikle, Cressie - 1999 |

3 |
Initial results of a mesoscale short-range forecasting system over the Pacific Northwest Wea. Forecasting 17
- Grimit, Mass
- 2002
(Show Context)
Citation Context ...n the introduction, numerical weather prediction (NWP) forecasts are not competitive at the forecast lead time that we consider here. However, NWP forecasts and mesoscale ensemble prediction systems (=-=Grimit and Mass 2002-=-) may provide independent information that could be used in the form of additional predictor variables for the conditional linear models, in identifying the forecast regimes, or in assessing and model... |

3 | A new reference for wind power forecasting - Nielsen, Joensen, et al. - 1998 |

1 | Improving wind energy forecasts - Larson, Gneiting - 2004 |