- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Faculty Research and Publications /
- Updating Short-Term Probabilistic Weather Forecasts...
Open Collections
UBC Faculty Research and Publications
Updating Short-Term Probabilistic Weather Forecasts of Continuous Variables Using Recent Observations. Nipen, Thomas 2011
pdf
Page Metadata
Item Metadata
Title | Updating Short-Term Probabilistic Weather Forecasts of Continuous Variables Using Recent Observations. |
Creator |
Nipen, Thomas West, Greg Stull, Roland B. |
Publisher | American Meteorological Society |
Date Created | 2011-11-22 |
Date Issued | 2011 |
Description | A statistical postprocessing method for improving probabilistic forecasts of continuous weather variables, given recent observations, is presented. The method updates an existing probabilistic forecast by incorporating observations reported in the intermediary time since model initialization. As such, this method provides updated short-range probabilistic forecasts at an extremely low computational cost. The method models the time sequence of cumulative distribution function (CDF) values corresponding to the observation as a first-order Markov process. Verifying CDF values are highly correlated in time, and their changes in time are modeled probabilistically by a transition function. The effect of the method is that the spread of the probabilistic forecasts for the first few hours after an observation has been made is considerably narrower than the original forecast. The updated probability distributions widen back toward the original forecast for forecast times far in the future as the effect of the recent observation diminishes. The method is tested on probabilistic forecasts produced by an operational ensemble forecasting system. The method improves the ignorance score and the continuous ranked probability score of the probabilistic forecasts significantly for the first few hours after an observation has been made. The mean absolute error of the median of the probability distribution is also shown to be improved. Copyright 2011 American Meteorological Society (AMS). Permission to use figures, tables, and brief excerpts from this work in scientific and educational works is hereby granted provided that the source is acknowledged. Any use of material in this work that is determined to be “fair use” under Section 107 of the U.S. Copyright Act or that satisfies the conditions specified in Section 108 of the U.S. Copyright Act (17 USC §108, as revised by P.L. 94-553) does not require the AMS’s permission. Republication, systematic reproduction, posting in electronic form, such as on a web site or in a searchable database, or other uses of this material, except as exempted by the above statement, requires written permission or a license from the AMS. Additional details are provided in the AMS Copyright Policy, available on the AMS Web site located at (http://www.ametsoc.org/) or from the AMS at 617-227-2425 or copyright@ametsoc.org. |
Genre |
Article |
Type |
Text |
Language | Eng |
Collection |
Atmospheric Science Program |
Date Available | 2011-11-22 |
DOI | 10.14288/1.0041971 |
Affiliation |
Science, Faculty of Earth and Ocean Sciences, Department of |
Citation | Nipen, Thomas N.; West, Greg; Stull, Roland B. 2011. Updating Short-Term Probabilistic Weather Forecasts of Continuous Variables Using Recent Observations. Weather and Forecasting, 26 (4) 564-571, http://dx.doi.org/10.1175/WAF-D-11-00022.1 |
Peer Review Status | Reviewed |
Scholarly Level | Faculty |
Copyright Holder | Stull, Roland B. |
URI | http://hdl.handle.net/2429/39213 |
Digital Resource Original Record | https://open.library.ubc.ca/collections/32536/items/1.0041971/source |
Download
- Media
- Stull_AMS_2011_WAF-D-11-00022.pdf [ 754.62kB ]
- Stull_AMS_2011_WAF-D-11-00022.pdf
- Stull_AMS_2011_WAF-D-11-00022.pdf
- Metadata
- JSON: 1.0041971.json
- JSON-LD: 1.0041971+ld.json
- RDF/XML (Pretty): 1.0041971.xml
- RDF/JSON: 1.0041971+rdf.json
- Turtle: 1.0041971+rdf-turtle.txt
- N-Triples: 1.0041971+rdf-ntriples.txt
- Citation
- 1.0041971.ris
Full Text
Updating Short-Term Probabilistic Weather Forecasts of Continuous Variables Using Recent Observations THOMAS N. NIPEN, GREG WEST, AND ROLAND B. STULL University of British Columbia, Vancouver, British Columbia, Canada (Manuscript received 16 February 2011, in final form 5 June 2011) ABSTRACT A statistical postprocessing method for improving probabilistic forecasts of continuous weather variables, given recent observations, is presented. The method updates an existing probabilistic forecast by in- corporating observations reported in the intermediary time since model initialization. As such, this method provides updated short-range probabilistic forecasts at an extremely low computational cost. The method models the time sequence of cumulative distribution function (CDF) values corresponding to the observation as a first-orderMarkov process. Verifying CDF values are highly correlated in time, and their changes in time are modeled probabilistically by a transition function. The effect of the method is that the spread of the probabilistic forecasts for the first few hours after an observation has beenmade is considerably narrower than the original forecast. The updated probability distributions widen back toward the original forecast for forecast times far in the future as the effect of the recent observation diminishes. The method is tested on probabilistic forecasts produced by an operational ensemble forecasting system. The method improves the ignorance score and the continuous ranked probability score of the probabilistic forecasts significantly for the first few hours after an observation has been made. The mean absolute error of the median of the probability distribution is also shown to be improved. 1. Introduction Correctly predicting forecast uncertainty can bring significant economic benefits to many decision makers (AMS 2008). Unlike a deterministic forecast, which supplies only the expected weather outcome, a proba- bilistic forecast gives the likelihood of occurrence of all outcomes. Decisions are based on combining the rela- tive risks of various weather outcomes with the costs and losses corresponding to those outcomes. Thus, proba- bilistic forecasts are naturally preferred for economic decision making. Let ft(x) be the forecasted probability density function (PDF) of a continuous meteorological variable X (such as temperature) valid for time t. One can generate ft(x) froman ensemble of numerical weather prediction (NWP) models by using methods such as Bayesian model av- eraging (Raftery et al. 2005), the binned probability ensemble technique (Anderson 1996), the method of moments (Jewson et al. 2005), or local quantile re- gression (Bremnes 2004). Let Ft(x) denote the forecasted cumulative distribu- tion function (CDF) given by Ft(x)5 ðx 2‘ ft(s) ds. (1) In addition, let xt denote the observed state of X at time t. Let pt denote the CDF value corresponding to the observed state: pt5Ft(xt). (2) Often, pt is called the probability integral transform value (PIT value) corresponding to the observation. We will assume an operational ensemble forecasting system initialized at time t5 0 that gives hourly forecasts out to time t 5 T. At times t, where 0 # t # T, hourly observations from observing stations are made avail- able, but the models do not incorporate these observa- tions until the next forecast cycle starts. Figure 1a shows a sample temperature CDF forecast for a single location produced from an ensemble. At Corresponding author address: Thomas Nipen, Dept. of Earth and Ocean Sciences, 6339 Stores Rd., Vancouver BC V6T 1Z4, Canada. E-mail: tnipen@eos.ubc.ca 564 WEATHER AND FORECAST ING VOLUME 26 DOI: 10.1175/WAF-D-11-00022.1 2011 American Meteorological Society the time the figure was produced, observations up to 1000 UTC were available. What is clear from the figure is that the CDF value that the observation verifies on (PIT value) is highly correlated in time (Fig. 1c). Given that the most recent PIT value (at 1000UTC) is 0.75, the next PIT value (at 1100 UTC) will likely be near 0.75. The probability distribution can therefore be refined to take into account this new information that was not available at the time the model was initialized. The ef- fects of the most recent observation will diminish for longer lead times. The updated probability distribution will therefore be narrow near the time of the observation and widen back to the original distribution for times in the future (Fig. 1b). The goal of this paper is to present a method for producing an updated probabilistic forecast F̂t(x) by mapping the original CDF Ft(x) by a function F as follows: F̂t(x)5F[Ft(x)]. (3) The mapping will concentrate F̂ in a narrower range with the hope of improving short-term verification scores. End users in need of rapidly updating probabilistic short-term forecasts at very low computational costs can benefit from this update method. Postprocessing weather forecasts is commonly done to increase the correspondence between forecasts and observations. For deterministic forecasts, methods such as model output statistics (Glahn and Lowry 1972), Kalman filtering (Homleid 1995), and analog methods (Delle Monache et al. 2011) are commonly used to re- duce forecast error. On the other hand, methods such as ensemble calibration (Hamill and Colucci 1998) and Bayesian model averaging (Raftery et al. 2005) can be used to improve probabilistic forecasts from an ensemble of deterministic forecasts. The method presented here also aims to improve probabilistic forecasts, but differs in that it is only invoked once observations are available after the raw forecasts are created. It is therefore of most use for operational short-term forecasts. This paper is organized as follows: the method for updating probabilistic forecasts is presented in section 2, the dataset and verification metric used for testing the method is described in section 3, the performance of the method is evaluated in section 4, and conclusions are drawn in section 5. 2. Method Assume that for a given forecast day, T 1 1 hourly probabilistic forecasts Ft(x) (where 0 # t # T) are pro- duced. Let tobs denote the time at which the most recent observation was made. This observation is then used to update all hourly forecasts that are still in the future (i.e., where tobs , t # T ). The probabilistic forecast n hours after tobs, that is for time t 5 tobs 1 n, can be updated according to F̂t obs 1n(x)5Fn[Ft obs 1n(x)], (4) whereFn(p) will in general be different for each value of n and can be constructed based on forecast and obser- vation data prior to the time tobs. Here, Fn(p) is the probability function that the verifying PIT value of the original forecast will be less than p. Combining Eqs. (1) and (4) and using the chain rule gives the following for the updated PDF: FIG. 1. (a) A sample probabilistic temperature forecast initial- ized at 0000 UTC. Forecasted cumulative probability values are shown by lines. Observations are shown by solid dots. (b) The updated probabilistic forecast (solid lines) based on the most re- cent observation. The original forecast is shown by dashed lines. (c) The probability integral transform values of the original forecast corresponding to the observations. AUGUST 2011 N I PEN ET AL . 565 f̂ t obs 1n(x)5Cn[Ft obs 1n(x)]ft obs 1n(x), (5) where Cn(p) is the derivative of Fn(p) and acts as an amplification factor for the original PDF.We note that Cn(p) increases probability density in regions where the PIT value is more likely to occur given the recent observation. That is, Cn(p) is also the probability den- sity of p being the verifying PIT value of the original forecast. a. PIT values as a random walk in time We model the time sequence of verifying PIT values within one forecast cycle as a random walk in time. Mirror barriers at 0 and 1 are used to handle the fact that PIT values are bounded on the interval [0, 1]. That is, any random steps across the boundaries are reflected back into the domain (Fig. 2). Mirror barriers are com- monly used to describe stochastic processes in other areas of modeling [Karlin and Taylor (1981); see also Rose (1995) for applications in economics]. Let ptobs be the PIT value of the most recent obser- vation, and letCn(p) be the probability density function of the verifying PIT value being p at n hours after tobs. When n 5 0, the PIT value is fully known and can therefore be described by C0(p)5 d(p 2 pt obs ), (6) where d is the Dirac delta function defined by d(s)5 1‘ s5 0 0, s 6¼ 0 and (7) ð‘ 2‘ d(s) ds5 1. (8) Let S(p, q) represent the probability density of ar- riving at a PIT value of p, given that the previous PIT value was q. Since our stochastic model for PIT values is a first-order Markov model, the probability of a certain PIT at time n can be found from all transitions to that PIT from time n 2 1. The probability density after a transition can therefore be determined by the following recursive equation: FIG. 2. (a) A hypothetical time series of verifying PIT values (solid line). Mirror barriers at 0 and 1 reflect any steps back into the domain. The dashed line shows the PIT time series without reflections. The transition from time 3 to 4 involves a reflection across 1 as shown by the arrows. (b) The PDF (thick solid line) of the PIT value for time 9, given that the PIT value at time 8 was 0.80. The dashed line shows the probability of the Gaussian distribution that has been reflected back into the domain. 566 WEATHER AND FORECAST ING VOLUME 26 Cn(p)5 ð1 0 S(p, q)Cn21(q) dq. (9) b. Determining the transition function We assume that the step length from one PIT to the next is Gaussian distributed with mean 0 and variance s2. That is, the transition function S can be constructed as follows: S(p,q)5f(p;q;s2) 1 f(2p;q;s2) 1 f(2 2 p;q;s2) 1 (10) 5 1‘ i52‘ [f(p 1 2i;q;s2) 1 f(2p 1 2i;q;s2)], (11) where f(x; m; s2) is a Gaussian PDF with mean m and variance s2. The first term in Eq. (10) comes from steps within the domain, the second comes from steps reflected across 0, and the third term comes from steps reflected across 1. Equation (11) includes all possible steps, inclu- ding steps that cross both boundaries one or more times. A transition function that combines n number of steps can also be constructed and is denoted by Sn. The variance of multiple steps (under the assumed model) increases linearly with time, and Sn can therefore be computed by Sn(p, q)5 1‘ i52‘ [f(p 1 2i; q; ns2)1 f(2p1 2i;q; ns2)]. (12) Since s is small in our study (around 0.15), and we use values of n no larger than 24, we restrict the summation to i 2 [210, 10]. A wider range for imay be required for large s and n values. Constructing Sn allows us to simplify Eq. (9) to the following: Cn(p)5 ð1 0 Sn(p, q)C0(q) dq (13) 5 Sn(p, pt obs ), (14) FIG. 3. An example sequence of PDFs of PIT values for different numbers of hours (n) after an observation has been made. In this case at n 5 0, the PIT value is fully known to be 0.7. AUGUST 2011 N I PEN ET AL . 567 where again p tobs is the verifying PIT value at time tobs. This simplification avoids the need to recursively compute Cn [as in Eq. (9)]. Note that for forecast var- iables that require a non-Gaussian transition function, it is possible that Eq. (12) cannot be constructed ana- lytically in which case the above simplification may not be possible. Figure 3 shows an example sequence of Cn(p) for various values of n. The PIT value distribution clearly widens as time goes on, indicative of the disappearing effects of the last observed PIT value. c. Parameter estimation To create the updated forecasts, an estimate of s2 is needed by Eq. (12). The variance of the step sizes of past PIT values (s20) can be used: s205 1 jT j i2T (pt112pt) 2, (15) where T represents a set of time points from the past forecast cycles that compose the training period and where jT j is the size of this training set. In general, s20 will underestimate s2 since some steps will appear to be short steps when in fact they are longer steps that have reflected across a boundary. For a given s, the expected value of s0 can be com- puted by the integral over all possible PIT transitions from p to q: s205 ‘ i52‘ ð1 0 ð1 0 [f(p 1 2i; q;s)(p2q)2 1 f(2p 1 2i; q;s)(p2q)2] dp dq. (16) Solving this equation for s [as required by Eq. (12)] was not possible analytically. We found that the following is a good approximation for s in terms of s0: s’ tan(3:5s0)/3:5, (17) where the input to the tangent function is in radians. This approximation has errors of less than 3.4% for s0 values up to 0.3 (Fig. 4). A summary of the process involved with updating a probabilistic forecast goes as follows: the variance of past PIT transition distances (s0) is computed by Eq. (15), which is used to approximate s in Eq. (17); s is then used in Eq. (12) to compute the transition function (Sn); and the transition function, combined with the latest available verifying PIT value, are used to calcu- late the PIT distribution (Cn) by Eq. (14), which is used to update the original probabilistic forecast through Eq. (5). 3. Operational test case a. Model data and configuration Hourly surface temperature forecasts from the Me- soscale Compressible Community (MC2; Benoit et al. 1997) model, the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (Penn State–NCAR) Mesoscale Model (MM5; Grell et al. 1994), and the Weather Research and Forecasting Model (WRF; Skamarock et al. 2005) were used for the case study period: 0000 UTC 1 September 2005– 2300 UTC 1 February 2008. Two runs for the WRF model were used: one using Global Forecast System (GFS) initialization (WRFG) and the other using North American Mesoscale Modeling (NAM) model initiali- zation (WRFN), while MC2 and MM5 both used NAM initialization. The MC2 and MM5 runs had outer do- mains with 108-km grid spacing, and inner 36-, 12-, and 4-km nested domains. The WRF runs were similar, but did not contain the 4-km nested domain. These domains composed our 14-member ensemble. The models were initialized once per day at 0000UTC, and hourly forecast output to 60 h was available. Probabilistic forecasts were generated for the same time period. The model runs and probabilistic forecasts were generally completed by 0600 UTC, after which we used the latest observation to update the probabilistic fore- casts valid for the subsequent 24 h. The update process FIG. 4. Standard deviation of PIT step sizes used in the transition function as a function of the measured standard deviation of step sizes of past PIT values (solid line) and the approximation s 5 tan(3.5s0)/3.5 (dashed line). 568 WEATHER AND FORECAST ING VOLUME 26 was repeated each hour as a new observation became available. This was done until 0600 UTC the next day, when the probabilistic forecasts from the next forecast cycle were used. This means that for each forecast cycle twenty-four 24-h updated forecasts were produced, yielding 576 forecasts per day. We tested the method on temperature probabilistic forecasts and observations for the following five airport stations in British Columbia, Canada: Vancouver In- ternational Airport station (CYVR), Abbotsford In- ternational Airport (CYXX), Victoria International Airport (CYYJ), Kamloops Airport (CYKA), and Kelowna Airport (CYLW). This group of stations provided a geographically diverse sample from within our smallest model domain. b. Original probabilistic forecasts We used the method of moments to produce the original probabilistic forecast from the forecast ensem- ble. The PDF using this method is computed by ft(x)5f(x; jt1m; s 2), (18) where again f is a Gaussian PDF, x is a temperature value, jt is the ensemble mean at time t, m is a bias- correction term for the center of the distribution, and s2 is the variance of the distribution. The last two parameters are determined by the fore- cast errors during the training period T : m5 1 jT j i2T xi 2 ji and (19) s25 1 jT j i2T (xi2m2 ji) 2. (20) Note that the spread in this case is independent of the ensemble spread. The parameters m and swere computed separately for each station and separately for each of the 24 forecast hours. They were computed from a 40-day sliding win- dow that ended the day before the forecast was initial- ized. A training period of 40 days is a compromise between the need to use statistics that adapt quickly to seasonal changes and the requirement to have enough data to robustly estimate the parameters. Similar train- ing lengths have been used to produce probabilistic forecasts using Bayesian model averaging (Raftery et al. 2005; Sloughter et al. 2007). The spread parameter s0 (and consequently s) was also computed separately for each station using a 40-day sliding window; however, all 24 forecast offsets for a given station were pooled together to give a more robust estimate. 4. Analysis a. Ignorance score We use the logarithmic score of Good (1952), which has gained popularity over the last decade and has been referred to as the ‘‘ignorance’’ score owing to its ties with information theory (Roulston and Smith 2002). It is defined as follows: IGN( f )5 1 jT j t2T 2log2[ft(xt)]. (21) IGN rewards forecasts that place high confidence in the value where the observation falls. Low ignorance scores are desired. The total ignorance scores of the original probabilistic forecasts were computed by averaging ignorance scores over all forecast cycles, and forecast hours, but sepa- rately for each station and each value of n in order to see how far into the future a recent observation can improve the ignorance score. Figure 5a shows the improvement in the ignorance score provided by the updated probabilistic forecast as a function of distance from the most recent observation. The updated forecasts at 0 h after an observation has been made has an ignorance score of 2‘ since the true state is fully known. However, this update forecast is of no value since it is only available after the observation has been made. As the time since the most recent ob- servation increases, the improvement in the ignorance score reduces down toward 0. b. Continuous ranked probability score We also computed the continuous ranked probability score (CRPS) to further evaluate the quality of the probabilistic forecasts. It is defined as CRPS(F)5 1 jT j t2T ð1‘ 2‘ [Ft(x)2H(x2 xt)] 2 dx, (22) where H(s) is the Heaviside function defined by H(s)5 1 s$ 0 0 s, 0 . (23) Low values of CRPS are preferred. Figure 5b shows the percentage improvement due to the updated forecast relative to the original raw forecast. This is defined as %improvement5 CRPS(Fraw)2CRPS(Fupdated) CRPS(Fraw) 3100%. (24) AUGUST 2011 N I PEN ET AL . 569 Results for CRPS show a similar pattern as for the ignorance score, with the update method providing less improvement as the time since the most recent obser- vation increases. The average CRPS of the five stations was 1.508C and the update method brought the values down to 1.068 and 1.278C at 3 and 6 h, respectively. c. Reliability A probabilistic forecast is reliable (or calibrated) when the PIT values are uniformly distributed between 0 and 1 (Gneiting et al. 2007). This can be diagnosed by a simple histogram of verifying PIT values, as reliable forecasts will give a flat histogram. Figure 5c shows the histogram of PIT values from all forecast hours, forecast cycles, stations, and values of n. The update method does not appear to degrade or im- prove the reliability of the original forecasts in any sig- nificant way. d. Mean absolute error A probabilistic forecast can also provide a best de- terministic estimate, by using the median of the proba- bility distribution (as shown by the 50% lines in Figs. 1a and 1b). We used the mean absolute error (MAE) as a verification measure of this deterministic forecast: MAE( f )5 1 jT j t2T jxt 2 F 21 t (0:5)j, (25) where F21t is the inverse of Ft giving the temperature value corresponding to a nominal proportion of 0.5. The MAE of the deterministic forecast (Fig. 5d) showed a similar pattern to the ignorance score and CRPS, with the update method improving the MAEs from 2.078C down to 1.428C and 1.738C at 3 and 6 h, respectively. Improvements in MAE suggest that the update method improves the central tendency of the probabilistic forecasts. 5. Conclusions We have presented a method to update probabilistic forecasts of continuous variables based on recent ob- servations, which should prove useful for a variety of nowcasting purposes. An alternative to this is to use data assimilation after new observations are available in FIG. 5. Verification statistics for the probabilistic forecasts used in the study. (a) Reduction (improvement) of the ignorance score by the updated probabilistic forecast relative to the original probabilistic forecast. Each of the five lines represents the score for a different station. (b) Percentage improvement in the CRPS by the updated probabilistic forecast. (c) PIT his- togram of the updated forecasts (black bars) and the original forecasts (white bars), indicating the reliability of the forecasts. (d) Percentage improvement in mean absolute error of the median of the updated probability distributions relative to the median of the original distri- bution. 570 WEATHER AND FORECAST ING VOLUME 26 order to create new initializations for the ensemble, fol- lowed by a complete rerun of the ensemble. This is con- siderably more expensive from a computational point of view, andmay be infeasible formany operational systems. The method improves the ignorance score and CRPS of the probabilistic forecasts, and improves the MAE of the median of the distribution significantly for forecasts up to 6 h after a recent observation, while not affecting reliability negatively. Future work includes investigating the benefits of using a higher-order Markov model for modeling PIT transitions. In addition to accounting for the hour-by-hour correlation of PIT values, a higher-order Markov model can also incorporate any diurnal corre- lation of PIT values that may exist, thereby allowing for the potential to improve forecasts for 24 h after a recent observation. Acknowledgments. This research was made possible by funding from the Canadian Natural Science and Engineering Research Council, the Canadian Founda- tion for Climate and Atmospheric Science, and the BC Hydro and Power Authority. We also thank Kristian Soltesz and two reviewers for providing helpful insight. REFERENCES AMS, 2008: Enhancing weather information with probability forecasts. Bull. Amer. Meteor. Soc., 89, 1049–1053. Anderson, J. L., 1996: A method for producing and evaluating probabilistic precipitation forecasts from ensemble model in- tegrations. J. Climate, 9, 1518–1530. Benoit, R., M. Desgagne, P. Pellerin, S. Pellerin, Y. Chartier, and S. Desjardins, 1997: The CanadianMC2: A semi-Lagrangian, semi-implicit wideband atmospheric model suited for fi- nescale process studies and simulation.Mon. Wea. Rev., 125, 2382–2415. Bremnes, J. B., 2004: Probabilistic forecasts of precipitation in terms of quantiles using NWP model output.Mon. Wea. Rev., 132, 338–347. Delle Monache, L., T. Nipen, Y. Liu, G. Roux, and R. Stull, 2011: Kalman filter and analog schemes to postprocess numerical weather predictions.Mon. Wea. Rev., in press. Glahn, H., and D. Lowry, 1972: The use of model output statistics (MOS) in objective weather forecasting. J. Appl. Meteor., 11, 1203–1211. Gneiting, T., F. Balabdaoui, and A. E. Raftery, 2007: Probabilistic forecasts, calibration and sharpness. J. Roy. Stat. Soc., 69B, 243–268. Good, I. J., 1952: Rational decisions. J. Roy. Stat. Soc., 14B, 107– 114. Grell, G. J., J. Dudhia, and D. R. Stauffer, 1994: A description of the fifth generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Rep. TN-3981STR, 122 pp. Hamill, T. M., and S. J. Colucci, 1998: Evaluation of Eta–RSM ensemble probabilistic precipitation forecasts. Mon. Wea. Rev., 126, 711–724. Homleid, M., 1995: Diurnal correction of short-term surface tem- perature forecasts using the Kalman filter. Wea. Forecasting, 10, 689–707. Jewson, S., A. Brix, and C. Ziehmann, 2005: Weather Derivative Valuation. Cambridge University Press, 373 pp. Karlin, S., and H. Taylor, 1981: A Second Course in Stochastic Processes. Academic Press, 582 pp. Raftery, A. E., T. Gneiting, F. Balabdaoui, and M. Polakowski, 2005: Using Bayesian model averaging to calibrate forecast ensembles.Mon. Wea. Rev., 133, 1155–1174. Rose, C., 1995: A statistical identity linking folded and censored distributions. J. Econ. Dyn. Control, 19, 1391–1403. Roulston, M. S., and L. A. Smith, 2002: Evaluating probabilistic forecasts using information theory. Mon. Wea. Rev., 130, 1653–1660. Skamarock,W. C., J. B. Klemp, J. Dudhia, D. O.Gill, D.M. Barker, W. Wang, and J. G. Powers, 2005: A description of the Ad- vanced Research WRF version 2. NCAR Tech. Rep. TN- 4681STR, 88 pp. Sloughter, J. M., A. E. Raftery, and T. Gneiting, 2007: Probabilistic quantitative precipitation forecasting using Bayesian model averaging.Mon. Wea. Rev., 135, 3209–3220. AUGUST 2011 N I PEN ET AL . 571
Cite
Citation Scheme:
Usage Statistics
Country | Views | Downloads |
---|---|---|
China | 19 | 0 |
Japan | 5 | 0 |
United States | 3 | 0 |
City | Views | Downloads |
---|---|---|
Beijing | 19 | 0 |
Tokyo | 5 | 0 |
Redmond | 1 | 0 |
Sunnyvale | 1 | 0 |
Ashburn | 1 | 0 |
{[{ mDataHeader[type] }]} | {[{ month[type] }]} | {[{ tData[type] }]} |
Share
Share to: