Evaluation of probabilistic prediction of 10meter wind speed using two post-processing methods

Document Type : Original Article

Authors

1 M.Sc. Student, University of Hormozganm Bandar Abbas, Iran

2 Assistant Prof., University of Hormozgan-Bandar Abbas- Iran

3 Associate Prof, ASMERC-Tehran-Iran

Abstract

Abstract

Wind energy has been considered as one of the clean energy sources. Due to the variability of wind speed and its effect on wind power plant, wind forecasting methods are of special importance. Fossil fuel consumption has destructive effects on the environment. According to Renewable Energy Policy Network for 21st century(REN21st) in 2014, nearly 20% of the total electricity was generated by wind energy, and the European Wind Energy Association (EWEA) predicts that it reaches 24.4 % in 2030. Wind speed has a great impact on increasing or decreasing air pollution and thus human health. Also it is one of the most important and effective factors in evaporation. Accurate prediction of wind speed is crucial in many social applications such as weather warnings in risk assessment and appropriate decisions in aviation, ship navigation, recreational sailing and agriculture.

With the advancement of computers and the ability to perform fast calculations, it became possible to implement weather forecasting models. At first, looking at weather forecasting was only a deterministic forecast, but since numerical weather prediction models include differential equations that approximately describethe physical and dynamic laws of the atmosphere, the answer obtained from the implementation of numerical models is an approximation of the real answer and is always in error. Also, factors such as the existence of errors in the initial boundary values, the inability of the model to take into account all atmospheric processes, the lack of primary data in some areas, the inability of the model to successfully simulate subnet phenomena and the chaotic nature of the atmospheric dynamic system can be mentioned. Hence, forecasting in such a system will be accompanied by high uncertainty. Therefore, to determine how the weather will be in the future, taking into account the mentioned uncertainty, there is an approach called probability forecasting. In this method, by using probabilistic prediction the chance of possible occurrence of future states of the atmosphere is calculated and the uncertainty is quantified and can be obtained more and accurate information. In this way, instead of considering only one model with an initial value, with a physical schema and a dynamic core, it can be created a finite number of prediction models by changing each of these three cases. In this study with the change in the model physics schema, the uncertainty caused by the model physics is considered and each produced prediction is considered a member of the system which is different from others. By applying statistical methods to the members of the ensemble system, a probability distribution function will be obtained in which post-processing is also performed and describes the uncertainty of the future state of the atmosphere and includes sufficient information for the needs of different users.

Materials and Methods

In this study, probabilistic wind speed predictions are generated after applying two methods of BMA and NGR on the raw output of ensemble system for 24, 48 and 72 hour forecasts and compared with raw probability predictions of the system which are selected democratic voting. The applied ensemble system consists of eight different physical configurations, with changes in the boundary layer scheme of the WRF model with a resolution of 21 kilometers over Iran.GFS forecasts are used for the initial and boundary conditions, and the forecast start time is 12 UTC per day. Observation data of 31 synoptic meteorological stations located in the provincial capitals have been used and the corresponding values of the predictions on these stations have been interpolated by bilinear method. The model run from 1 March to 31 August 2017, and the results from 11 April to 31 August 2017 are considered as the test period. After calculating the forecast errors with different training periods, 30 days are considered as the length of training period for prediction in both BMA and NGR methods.

Conclusion

Verification was performed by Barrier Score(BS), Barrier Skill Score(BSS), reliability diagram, ROC diagram and Economic Value diagram for 10-meter wind speed threshold less than 3 and more than 5,10 and 13 m/s for BMA, NGR and Raw probability prediction of the system in all forecast ages. The results show that BMA and NGR have improved BS, BSS 33% to 46% from Raw probability prediction of the system and the reliability and the resolution have improved 78% to 97% and 12 to 30 times respectively. Reliability diagram and ROC diagrams of NGR and BMA have been also improved significantly. Economic Value diagram shows also that using probabilistic predictions is important to reduce the cost of meteorological hazard.

In both BS and Reliability diagram the quality of NGR was better than BMA. In BSS, Roc diagram and Economic Value diagram the both two post-processing methods had the same advantage. The economic value diagram had the best performance at a wind speed threshold of 10 m/s.

Keywords

References

  1. منابع

    1. Azadi, M., Mohamadi, S.A., 2019, Probabilistic forecasting of diurnal maximum and minimum temperature using a 2-member ensemble system over Iran. Nivar 43(1),57-66
    2. Fathi, M., Azadi, M., Kamali, G., Meshkatee, A., 2018, Calibration of probabilistic precipitation forecasts based on ensemble output using bayesian model averaging over Iran. Journal of Meteorology and Atmospheric Science 1(2),114-129
    3. Mohamadi, S.A., 2016, Deterministic and prognostic numerical prediction of near-surface temperature over Iran in autumn and winter. PhD Thesis.  Academic Center for Education, Culture and Research
    4. Baran, S., Hor, A., 2013, Comparison of BMA and EMOS statistical calibration methods for temperature and wind speed ensemble weather prediction, submitted on 13 Dec 2013, Idojaras, 118(2014), 217-241.
    5. Baran, Á., Lerch, S., El Ayari, M., & Baran, S. (2021). Machine learning for total cloud cover prediction. Neural Computing and Applications, 33(7), 2605-2620.‏
    6. Eckel, F., Allen, M., and Sittel, M., 2012: Estimation of ambiguity in ensemble forecasts, Weather and Forecasting, 27(1), 50-69.
    7. François, B., Hugo, M., 2020, Probabilistic thunderstorm forecasting by blending multiple ensembles, Tellus A: Dynamic Meteorology and Oceanography, 72(1), 1-19
    8. Gilbert, C., Messner, J. W., Pinson, P., Trombe, P. J., Verzijlbergh, R., van Dorp, P., & Jonker, H. (2020). Statistical post‐processing of turbulence‐resolving weather forecasts for offshore wind power forecasting. Wind Energy, 23(4), 884-897.‏
    9. Glahn, B., Peroutka, M., Wiedenfeld, J., Wagner, J., Zylstra, G., Schuknecht, B., and B., Jackson, 2009: MOS uncertainty estimates in an ensemble framework, Monthly Weather Review, 137(1), 246-268.
    10. Gneiting, T., Balabdaoui, F., and A., Raftery, 2007: Probabilistic forecasts, calibration and sharpness, Royal Statistical Society, 69(2), 243-268.
    11. Gneiting, T., Raftery, A., Westveld, A., and T., Goldman, 2005: Calibrated probabilistic forecasting using ensemble model output statistics and minimum CRPS estimation, Monthly Weather Review, 133(5), 1098-1118.
    12. Han, K., Choi, JT., 2018, Comparison of Statistical Post-Processing Methods for Probabilistic Wind Speed Forecasting, Asia-Pacific Journal of Atmospheric Sciences, 54(1): 91-101
    13. Konstantinou, T., Savvopoulos, N., Hatziargyriou, N., 2020, Post-processing Numerical Weather Prediction for Probabilistic Wind Forecasting, published in: 2020 International Conference on Probabilistic Methods Applied to Power Systems (PMAPS), Date of Conference: 18-21 Aug. 2020
    14. Lorenz, E., 1996: Predictability: A problem partly solved, In Proc. Seminar on Predictibality, Conference LocationShinfield Park-Reading, 10829,1-18
    15. Marcos, R., Gonzales, N., 2018, Characterization of the near surface wind speed distribution at global scale: ERA-Interim reanalysis and ECMWF seasonal forecasting system 4, Climate Dynamics 52, 3307–3319.

     

    1. Mohammadi, S., A., 2016, Meta-heuristic CRPS minimization for the calibration of short-range probabilistic forecasts, Meteorology and Atmospheric Physics, 128(4),429-440.
    2. Raftery, A., Gneiting, T., Balabdaoui, F., 2005, Using Bayesian model averaging to calibrate forecast ensemble, Monthly Weather Review, 133(5), 1155-1174.
    3. Sloughter, J. M., Gneiting, T., Raftery, A. E., 2010: Probabilistic wind forecasting forecasting using ensembles and Bayesian model averaging., Journal of the American Statistical Association, 105(489), 25–35.
    4. Unger, D. A., 1985, A method to estimate the continuous ranked probability score, in Preprints, Ninth Conf. on Probability and Statistics in Atmospheric Sciences, Virginia Beach.
    5. Unger, D., Van Den Dool, H., O’Lenic E., 2009, Ensemble regression, Monthly Weather Review, 137(7), 2365-2379.
    6. Wilks, D., 2006: Comparison of ensemble-MOS methods in the Lorenz ’96 setting, Meteorol. Appl., 13, 243-256.
    7. Wilks, D., 2011, Statistical Methods in the Atmospheric Sciences, New York: Academic Press.
    8. Xu, W., Ning, L., & Luo, Y. (2020). Wind speed forecast based on post-processing of numerical weather predictions using a gradient boosting decision tree algorithm. Atmosphere, 11(7), 738.‏
Journal of Climate Research
Volume 1400, Issue 48
Vol. 12. No. 48, Winter 2022
April 2022
Pages 69-84

Files

Share

How to cite

Statistics