Comparing The Verification Results Of Wrf And Alaro Numerical Weather Prediction Models

Tez (Yüksek Lisans) -- İstanbul Teknik Üniversitesi, Fen Bilimleri Enstitüsü, 2015Thesis (M.Sc.) -- İstanbul Technical University, Institute of Science and Technology, 2015Son yıllarda gelişen teknoloji ile birlikte Sayısal Hava Tahmin (SHT) ürünlerinin çeşitliliği de artmıştır. Çoklu girdi verisi ile yapılan tahminler gibi SHT için istatistiki birçok yöntem uygulanmaktadır. Kısa ve uzun dönemli hava tahminleri, tarımdan ulaşıma, turizmden şehir planlamasına ve enerji ihtiyaçlarını karşılamaya yönelik birçok alanda önemli rol oynamaktadır. SHT modellerinin parametirizasyon setlerinin birbirine uyumluluğu, model girdi verisinin kalitesi, çalışma yapılan alanın coğrafi özellikleri vb. gibi durumlar atmosferik prognostik değişkenlerinde sistematik veya sistematik olmayan hatalar doğurmaktadır ve yapılan tahminin kalitesini etkilemektedir. Verifikasyon teknikleri, bu tahminlerdeki sistematik hataları ortaya çıkarmak ve tahmini gerçeğe yakınsatmak için kullanılmaktadır. Yapılan çalışmada, Weather Reseach and Forecasting (WRF) and transition step between ALadin and AROme (ALARO) modelleri Sınırlı Alan Modelleri (SAM) olarak kullanılmıştır. Bu modeller Meteoroloji Genel Müdürlüğü (MGM) tarafından ağırlıklı olarak kullanılan modellerdir. Integrated Forecasting System (IFS) modeli ise, global model olarak seçilmiştir. IFS modelinin ürettiği tahminler WRF ve ALARO modellerine girdi verisi olarak verilmiştir. WRF modeli Amerikan menşeli, açık kaynak kod yapısıyla bütün dünyada birçok kullanıcı tarafından kullanılan/geliştirilen, orta ölçekli SHT modelidir. Bu çalışmada, WRF modeli iç içe iki alan olarak koşturulmuştur. Dıştaki alanın özellikleri yatayda 13500 metre çözünürlükle ve 480x300 grid sayısı ve düşeyde 46 sigma seviye, içteki alanın özellikleri yatayda 4500 metre çözünürlükle 649x430 grid sayısı ve düşeyde de 46 sigma seviyedir. ALARO modeli, Fransa menşeli, Aire Limitée Adaptation dynamique Développment INternational (ALADIN) konsorsiyumunun geliştirdiği ve üye ülkeler tarafından kullanılan/geliştirilen, orta ölçekli SHT modelidir. ALARO modeli yatayda 4,5 kilometre çözünürlükle 709x439 grid sayısıyla ve düşeyde 60 sigma seviye ile hidrostatik olmayan eşitliği kabul ederek koşturulmuştur. IFS modeli, Avrupa Orta Ölçekli Hava Tahmin Merkezi (ECMWF) tarafından 1 Ağustos 1979 tarihinden beri operasyonel olarak koşturulan ve üye ülkelerce geliştirilen global modeldir. IFS global modelinin çıktıları, bu çalışmada koşturulan SAM 'ın girdi verisi olarak kullanılmıştır. Çalışmada kullanılan verifikasyon teknikleri, hata kareleri ortalamalarının karekökü (RMSE), ortalama hata (bias) ve isabet skorudur. RMSE, hatanın şiddeti hakkında bilgi verirken bias, hatanın yönü hakkında da bilgi vermektedir. İsabet skoru, modellerin parametreleri belirlenen aralıklardaki yakalama başarısını göstermektedir. Bu çalışma MGM'de operasyonel olarak çalıştırılan WRF ve ALARO modellerinin 5 yıllık Ocak ve Temmuz ayları için sıcaklık, aktüel basınç, rüzgar ve yağış parametrelerinin verifikasyonu niteliğindedir. ECMWF'den alınan tahmin verileri girdi verileri olarak, Türkiye'de 2010 yılında mevcut bulunan 361 adet otomatik gözlem istasyonunun 5 yıllık periyot içindeki ölçümleri de gözlem verileri olarak kullanılmıştır.Nowadays, the weather forecast plays an important role in our modern life. It has been used in extended area from agriculture to transport; from tourism to urban planning. Therefore the numerical products used in the prediction should be more realistic and effective. In recent years, number of the numerical products in the weather prediction area has increased with the developments in the simulation technologies. There are many statistical techniques in the literature such as multi input data predictions method also known as Ensemble Predictions System (EPS). The quality of the predictions can be affected by many error sources. Coherency of parametrization sets, quality of input data, geographic properties of study area and such circumstances lead to systematic or non-systematic errors. The systematic errors can be recognized by the verification techniques that are used to improve the predictions. The scope of this study is the evaluation of the verification techniques in Limited Area Models (LAM). The main goal of The Numerical Weather Prediction (NWP) is to produce prediction about the future or the past by means of meteorological observations that are processed with mathematical models in the simulation environment. The NWP models are classified in terms of temporal resolution and geographical scale. In the temporal resolution point of view, NWP models can be studied as the short-term predictions (per day or daily); the middle-term predictions (monthly or seasonal); and long-term predictions (decades to century). Besides that, when geographical range is considered, NWP models are splitted as Limited Area Models (LAM) and Global Models (GM). In this study, the Weather Research and Forecasting (WRF) and transition step between ALadin and AROme (ALARO) models are selected as a LAM. These models are typically used in the Numerical Weather Prediction Department at Turkish State Meteorological Service (TSMS). In addition, Integrated Forecasting System (IFS) is chosen as a global model. The prediction output from IFS model is used as input data for the WRF and ALARO models. WRF is an American originated, open source, non-hydrostatic mesoscale weather prediction model; that is used and developed by researchers around world. In this study, two nested domains are used for WRF model simulations. The outer domain has 13500 meters horizontal resolution; 480x300 grid points in the horizontal direction and 46 sigma levels in the vertical direction. On the other hand, the inner domain has 4500 meter horizontal resolution; 649x430 grid points in the horizontal direction and 46 sigma levels in the vertical direction. The current version of the model can be downloaded from the University Corporation for Atmospheric Research (UCAR) website. The other LAM used in this study is ALARO model. It is a French mesoscale weather prediction model that is generated/developed by Aire Limitée Adaptation dynamique Développment INternational (ALADIN) consortium and used by the member states. Its horizontal resolution is 4500 meters and it includes 709x439 grid points in the horizontal plane and the atmosphere is divided into 60 sigma levels in the vertical direction. Also, the model has hydrostatic and non-hydrostatic versions. In order to compare with WRF results, ALARO model's non-hydrostatic version is used in the simulations. IFS model, the input of the LAMs, has been run operationally at European Center for Mesoscale Weather Forecast (ECMWF) since 1 August 1979. The model is generated and developed by the member states of ECMWF. In this study, output of this model is used as input data for WRF and ALARO, without any processing operation. Objective weather forecast verification can be performed at least three different point of view: accuracy, skill and utility. From the accuracy point of view, the difference between prediction and measurements is evaluated. From the skill perspective, the predictions are compared with the reference data i.e compared with climate data or checked against an alternative forecast model. Last one, utility, the results of predictions are evaluated in terms of its political consequences or economic effects. Second and third viewpoints are both subjective interpretation, on the contrary, first one is an objective evaluation. In the thesis, forecasts are assessed from accuracy viewpoint. In the scope of above mentioned information, three verification methods are used in the present study. These are Root Mean Square Error (RMSE), Mean Error (ME) and Threat Score. RMSE, provides information about strength of the error. ME, gives information about strength and direction of the error. Finally, Threat Score demonstrates prediction accuracy of the model for predicting parameters in a predefined range. For the implementation of the mentioned verification techniques, matching sets of stations and model's grid points are determined by selecting the nearest grid point and weather observation stations by means of their latitude and longitude information. However, topography data sets of the models are different from actual situation. In other words, station's actual altitudes are differ from model's altitudes due to the nature of the models. It is assumed that the 361 stations from 2010 are common stations for a five-year calculation period (2009-2013). In the verification study, January and July months are selected for representing the winter and summer seasons respectively in the five-year calculation period. The WRF and ALARO models are run along 72-hour prediction for each day. The outputs of the models deal with the four prognostic parameters (temperature, actual pressure, wind speed and precipitation). For the temperature, actual pressure and wind speed the outputs are produced every three hours, on the other hand the precipitation output is taken for cumulative values in a day. Threat score is used for temperature, precipitation and wind speed values. However, the RMSE and bias values are generated for the temperature, actual pressure and wind speed. In addition, for the temperature parameter; the altitude correction implemented using calculated relative humidity and altitude values by means of linear method in order to decrease the error that comes from altitude differences between actual and model topography. This study involves a verification study of 5 years of January and July temperature, actual pressure, wind and precipitation parameters outputs of WRF and ALARO models that operationally run by TSMS. Prediction data from ECMWF is used as input data whereas automated weather observation stations data is used as observation data.Yüksek LisansM.Sc