نیوار

نیوار

پس پردازش برونداد مدل WRF برای پیش بینی بارش روزانه با استفاده از شبکه عصبی مصنوعی(مطالعه موردی: استان آذربایجان شرقی)

نوع مقاله : مقاله پژوهشی

نویسندگان
1 بخش حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی آذربایجان شرقی، تبریز، ایران
2 پژوهشگاه هواشناسی و علوم جو، تهران، ایران.
3 مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی آذربایجان شرقی، تبریز، ایران.
4 اداره کل هواشناسی آذربایجان شرقی، تبریز، ایران.
5 اداره کل هواشناسی آذربایجان شرقی، تبریز، ایران
10.30467/nivar.2026.553313.1357
چکیده
رخداد بارش به متغیرهای هواشناسی زیادی وابسته است و سبب پیچیدگی پیش بینی بارش می شود یکی از روش ها جهت افزایش دقت پیش بینی استفاده از مدل های شبکه عصبی مصنوعی(ANN) است که به طور مستقیم برای پیش بینی بارش و یا پس پردازش برونداد مدل های عددی پیش بینی وضع هوا استفاده می‌شود. در این مقاله متغیرهای هواشناسی از برونداد مدل عددی WRF استخراج و برای بهبود پیش بینی بارش ساعتی و روزانه از مدل شبکه عصبی مصنوعی برای پس پردازش برونداد مدل WRF استفاده گردید. متغیرهای هواشناسی که طی دوره آماری سه ساله (2023-2021) به صورت ساعتی از برونداد مدل WRF گرفته شده، شامل بارش، نقطه شبنم، میانگین فشار سطح دریا، رطوبت و دمای دومتری و سرعت باد است. این متغیرها در برد‌های پیش‌بینی 6، 12، 18، 24، 48 و 72 ساعته تقسیم‌بندی و هر بردی پیش‌بینی به طور جدا به عنوان ورودی برای هر ایستگاه به مدل شبکه عصبی مصنوعی داده شد. شبکه عصبی مصنوعی مورد استفاده یک پرسپترون چندلایه (MLP) با یک لایه‌ی پنهان و تابع فعال ساز سیگموئید است. بارش‌های مشاهداتی 6 ساعته برای هر ایستگاه به عنوان خروجی مدل شبکه عصبی مصنوعی استفاده شد. مدل روی 16 ایستگاه هواشناسی سینوپتیک استان آذربایجان شرقی اجرا گردید. در هر بار اجرای مدل شبکه عصبی مصنوعی، عملکرد مدل با شاخص‌های آماری میانگین توان دوم خطا (MSE) و ضریب همبستگی پیرسن(R) سنجیده شد. نتایج نشان داد که عملکرد مدل WRF پس از اعمال شبکه عصبی مصنوعی روی برونداد مدل برای شاخص MSE بین 36 تا 92 درصد برای ایستگاه‌های مورد نظر بهبود یافت. همچنین شاخص همبستگی R برای کل ایستگاهها به طور متوسط 85/0 درصد بوده است. همچنین نمودارهای خطای روزانه 95 درصد مدل‌ها نشان داد که در تمام ایستگاهها خطای روزانه مدل اصلاح شده نسبت به مدل WRF به صفر نزدیک‌تر و تغییرات یا واریانس خطا بسیار کمتر است.
کلیدواژه‌ها
موضوعات

عنوان مقاله English

Post-Processing of Daily Precipitation Forecast Output of the WRF Model Using Artificial Neural Nnetwork Models (Case study: East Azerbaijan Province)

نویسندگان English

Mahdi Eslahi 1
Majid Azadi 2
Jamshid Yarahmadi 3
Farnaz Pourasghar 4
Seyyedeh Atefeh Mohammadi 2
Younes Akbarzadeh 5
1 Soil Conservation and Watershed Management Department, East Azerbaijan Agricultural and Natural Resources Research and Education Center, Tabriz, Iran.
2 Meteorology and Atmospheric Sciences Research Institute, Tehran, Iran.
3 East Azerbaijan Agricultural and Natural Resources Research and Education Center, Tabriz, Iran.
4 East Azerbaijan Meteorological Department, Tabriz, Iran.
5 East Azerbaijan Meteorological Department, Tabriz, Iran
چکیده English

Precipitation occurrence is influenced by numerous meteorological variables, making its prediction a complex task. Consequently, both numerical weather prediction (NWP) models and statistical approaches have been developed to forecast daily precipitation. One promising method for improving forecast accuracy involves the use of Artificial Neural Networks (ANNs), which can be employed either directly for precipitation prediction or for post-processing the outputs of NWP models. In this study, meteorological variables derived from the output of the WRF (Weather Research and Forecasting) model were utilized to enhance hourly and daily precipitation forecasts. Specifically, an ANN-based post-processing approach was applied to refine the WRF model outputs. The meteorological variables extracted hourly over a three-year period (2021–2023) included accumulated precipitation (APCP), dew point temperature (DPT), mean sea-level pressure (PRMSL), relative humidity at 2 meters (RH2), 2-meter air temperature (T2m), and wind speed (Wspeed). These variables were categorized into forecast lead times of 6, 12, 18, 24, 48, and 72 hours, with each lead time used independently as input to the ANN model for each station.The ANN architecture employed was a multilayer perceptron (MLP) consisting of two layers: a hidden layer with sigmoid activation functions and an output layer with linear activation. The model was trained to predict 6-hour observed precipitation at each station. The ANN was implemented across 16 synoptic meteorological stations in East Azerbaijan Province. Model performance was evaluated using statistical metrics including Mean Squared Error (MSE) and Pearson correlation coefficient (R), calculated for the training, testing, and validation datasets. The model was iteratively trained until the minimum MSE and maximum R values were achieved. Results indicated that the ANN-enhanced WRF model significantly improved precipitation forecasts, with MSE reductions ranging from 36% to 92% depending on the station. Also, the correlation index R for all stations was 0.85 on average. Furthermore, 95% daily error plots demonstrated that the modified model exhibited lower error magnitudes and reduced variability compared to the original WRF outputs.

کلیدواژه‌ها English

Precipitation
WRF
FFN model
MSE index
R correlation index
1.    A. Chand and R. Nand, 2019, Rainfall prediction using Artificial Neural Network in the South Pacific region, IEEE Asia-Pacific Conference on Computer Science and Data Engineering (CSDE), Melbourne, VIC, Australia, 2019, pp. 1-7, doi: 10.1109/CSDE48274.2019.9162367.
2.    A. Kala and S. G. Vaidyanathan, 2018, Prediction of Rainfall Using Artificial Neural Network, International Conference on Inventive Research in Computing Applications (ICIRCA), Coimbatore, India, 2018, pp. 339-342, doi: 10.1109/ICIRCA.2018.8597421.
3.    Azadi, Majid, Jafari, Somayeh, Mirzaei, Ebrahim, Arabli, Parvin. (2008), Postprocessing of MM5 mesoscale model output for maximum and minimum temperature using Kalman filter, Journal of Earth and Space Physics, 2008, Volume 34, Issue 1; Page 45 to Page 61.(in Persian)
4.    Azadi, Majid, Shirgholami, Mohammad Reza, Hijam, Sohrab, and Sahraiian, Fatemeh. (2011). Postprocessing of WRF model output for daily rainfall in Iran. Iranian Water Resources Research, 7(4 (Issue 22)), 71-81. SID. https://sid.ir/paper/100248/fa. (in Persian)
5.    Azadi, Majid, Vashani, Saeed and Hijam, Sohrab (2011). Probabilistic Precipitation Forecasting Using Post-Processing of the Output of a Hamadi System, Quarterly Journal of Earth and Space Physics, 38(3), 203. magiran.com/p1069698. (in Persian)
6.    Biondi, Riccardo and Chkeir, Sandy and Anesiadou, Aikaterini and Mascitelli, Alessandra, 2022, Nowcasting Extreme Weather with Machine Learning Techniques Applied to Different Input Datasets. Available at SSRN: https://ssrn.com/abstract=4144317.
7.    Demuth, H., Beale, M., 2000. Neural Network Toolbox User’s Guide. Copyright 1992-2002. BT The Math Works, Inc, Version 4, 840P.
8.    Faraji, Mozafar, Rezaei Banafsheh, Majid, Sari Sarraf, Behrouz, Mohammad Khorshiddoost, Ali. 2023, Numerical Simulation of Iranian Climate Change Using Artificial Neural Network Algorithm, Climate Change Research, 4(14), pp. 43-64. doi: 10.30488/ccr.2023.397215.1135. (in Persian)
9.    Espeholt, L., Agrawal, S., Sønderby, C. et al. 2022, Deep learning for twelve hour precipitation forecasts. Nat Commun 13, 5145 (2022). https://doi.org/10.1038/s41467-022-32483-x.
10. F. Najib, Yusriadi, I. W. Mustika and S. Sulistyo, 2023, Rainfall Prediction using Artificial Neural Network with Forward Selection Method, 2023 IEEE International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT), BALI, Indonesia, 2023, pp. 102-107, doi:10.1109/IAICT59002.2023.10205930.
11. Fulop, I.A., Jozsa, J., Karamer, T., 1998. A neural network application in estimating windinduced shallow lake motion. Hydro informatics, 98, 2, pp 753-757.
12. Galanis, G., and Anadranistakis, M., 2002, A one-dimensional Kalman filter for the correction of near surface temperature forecasts: Meteorological Applications, 9, 437-441.
13. Galanis, G., Louka, P., Katsafados, P., Kallos, G., and Pytharoulis, I., 2006, Applications of Kalman filter based on non-linear functions to numerical weather predictions: Annals of Geophysics, 24, 2451-2460.
14. Gelman, A., Carlin, J.B., Stern, H.S., Dunson, D.B., Vehtari, A. and Rubin, D.B.,2013, Bayesian data analysis. Chapman and Hall/CRC.
15. Glossary of Meteorology, 2012. American Meteorological Society. Retrieved 2014-11-13
16. Haykin, Simon, 2001, Neural Networks: A comprehensive Foundation, An inprint of Pearson Education.
17. Homleid, M., 1995, Diurnal corrections of short-term surface temperature forecasts using the Kalman filter: Weather Forecast, 10, 689-707.
18. Ildarmi, Alireza, Zare Abyaneh, Hamid, Bayat Varkshi, Maryam. 2013. Precipitation estimation using artificial neural networks with non-rainfall meteorological data in three regions of Shiraz, Mashhad and Kerman, Geography and Planning, 17(43), pp. 21-40. (in Persian)
19. Kalman, R. E., 1960, A new approach to linear filtering and prediction problems: Journal of Basic Engineering, 82(1), 35-45.
20. Kalman, R. E., and Bucy, R. S., 1961. New results in linear filtering and prediction problems, Trans. ASME, Ser. D., 83, 95-108.
21. Khalili, Najmeh, Khodashenas, Saeed Reza, Davari, Kamran, and Mousavi Bayegi, Mohammad. 2010. Daily Precipitation Forecasting Using Artificial Neural Networks (Case Study: Mashhad Synoptic Station). Watershed Management Research (Research and Construction), 23(4 (89)), 7-15. SID. https://sid.ir/paper/200564/fa. (in Persian)
22. Lam, R., , Sanchez-Gonzalez, A., Willson, M., Wirnsberger, P., Fortunato, M., Alet, F., Ravuri, S. Ewalds, T., Eaton-Rosen, Z. and Battaglia, P. 2023. Learning skillful medium-range global weather forecasting, Science, 14 Nov 2023, Vol 382, Issue 6677, pp. 1416-1421, DOI: 10.1126/science.adi2336.
23. Malboosi, Sharara, Seyed Mahdavi, Seyed Javad, Pakdaman, Morteza. 2014. 'Precipitation forecasting using deep neural networks (sparse stack autoencoder with hard neuron-based denoiser)', Climatological Research, 2014(50), pp. 201-215. (in Persian)
24. Marzban, C., 2003, Neural network for post processing model output, ARPS, Mon. Wea. Rev., 131, 1103-1111.
25. Makhamisa Senekane, Mhlambululi Mafu, Molibeli Benedict Taele, 2020. Book chapter: Data Mining - Methods, Applications and Systems DOI: http://dx.doi.org/10.5772/intechopen.84552
26. Mccollor, D. and Stull, R., 2008, Hydro meteorological accuracy enhancement via post processing of numerical weather forecasts in complex terrain, Wea. Forecasting, 23, 131-144.
27. Mobasheri, Mohammad Reza, Pourbagerkerdi, Seyed Mehdi, Farajzadeh Asl, Manouchehr and Sadeghi Nayini, Ali, 2010. Estimation of Total Precipitable Water Using MODIS Satellite Images and Radiosound Data: Tehran Region, Journal of the Humanities Lecturer, 14(1), 107-126. magiran.com/p785823. (in Persian)
28. Mohammadiha, Amir, Memarian, Mohammad Hossein, Azadi, Majid, Reyhani Parori, Mohammad. 2014. Investigation of WRF model predictions for precipitation and its relationship with precipitation estimates using Tehran radar data, Iranian Journal of Geophysics, 8(3), pp. (in Persian)
29. Monache, L. D., Nipen, T., Liu, Y., Roux, G., and Stull, R., 2011, Kalman filter and analog schemes to post-process numerical weaher predictions: Monthly Weather Review, 139, 3554-3570.
30. Omidvar, Kamal, Nabavizadeh, Masoumeh. 2014. 'Predicting Daily Precipitation in Kerman Province with Artificial Neural Networks (Case Study: Kerman, Baft and Jiroft Midpoint)', Regional Geography and Development, 12(2), pp. -. doi: 10.22067/geography.v12i23.26991. (in Persian)
31. Omidvar, Kamal, Nabavizadeh, Masoumeh, Samareh Ghasem, Meysam. 2015. Assessment of the Accuracy of the Recurrent Artificial Neural Network Narx in Forecasting Daily Precipitation in Kerman Province, Natural Geography, , 8(27), pp. 73-90. (in Persian)
32. Pirone  D., Cimorelli  L., Del Giudice  G. and Pianese D., 2022. Rainfall Nowcasting Exploiting Machine-Learning Techniques: A Case Study in Southern Italy. Environ. Sci. Proc. 2022, 21(1), 49; https://doi.org/10.3390/environsciproc2022021049.
33. Taghavi, Farahnaz, Jalali, Mojtaba. 2014. 'Post-processing of the WRF Model Output for the Precipitation Water Parameter Using Satellite Data', Iranian Journal of Geophysics, 16(3), pp. 37-55. doi: 10.30499/ijg.2022.321906.1387. (in Persian)
34.Tao Chen and M. Takagi, 1993. Rainfall prediction of geostationary meteorological satellite images using artificial neural network, Proceedings of IGARSS '93 - IEEE International Geoscience and Remote Sensing Symposium, Tokyo, Japan, 1993, pp. 1247-1249 vol.3, doi: 10.1109/IGARSS.1993.322107.
دوره 50، 132-133 - شماره پیاپی 132
فروردین 1405
صفحه 222-243

  • تاریخ دریافت 11 آبان 1404
  • تاریخ بازنگری 10 تیر 1405
  • تاریخ پذیرش 20 تیر 1405
  • تاریخ انتشار 01 فروردین 1405