Effects of various digital filter initialization methods on results of weather research and forecasting (WRF) model

Numerical weather prediction (NWP) models without initialization techniques may beresult in unreal and inaccurate data. Many initialization methods, such as linear andnonlinear normal mode initialization, have been developed and applied in the field ofNWP by atmospheric researchers and modelers. In application, such techniques are very complex and expensive. One of the most efficient and simple techniques which can beused in operational forecasting is digital filter initialization. Digital filter initializationmethods are applied to eliminate non-physical and high frequency waves from NWPmodels. These unwanted waves can affect the results of the models and cause its results todepart from real world and observed conditions.In this paper, different filters (1- uniform filter, 2- Lanczos filter, 3- Hamming filter, 4-Blackman filter, 5- Kaiser filter, 6- Potter filter, 7- Dolph [Dolph-Chebyshev] window, 8-Dolph filter and 9- Recursive High-Order filter) are theoretically investigated.The theoretical study of these filters shows that the Dolf filter works better than theother filters. This superiority can be verified using a digital filter initialization techniqueassociated with the Dolf filter in the weather research and forecasting (WRF) model andinvestigating its results. Subsequently, the digital filter initialization methods provided inthe WRF model are tested for the region of Iran. Three different digital filter initializationtechniques, namely the digital filter launch, diabatic digital filter and twice digital filterinitialization, with nine aforementioned filters were prepared in the WRF model. TheWRF model was set with a 45-kilometer grid size for the region at 12-50 oN and 12-87 oS.The WRF model was run over this region with and without a digital filter initializationtechnique. In general, the initialization of the NWP models influences the first hours ofprediction of the meteorological parameters. In this study, two parameters, includingsurface pressure and rainfall, were considered as indicators of the effects of digital filterinitialization methods on the results of the WRF model. Therefore, the obtained resultsare investigated and compared for surface pressure fluctuation and rainfall.All results indicate that applying the digital filter initialization effectively liminatesnonphysical waves from surface pressure fields, especially in the first hours of prediction.This was determined by studying three parameters, including surface pressure fluctuationin some points, derivative of surface pressure fluctuation in some points, and integratedderivative of surface pressure. It was found that the twice digital filter initializationassociated with the Dolph filter works better than the other techniques and filters.For rainfall, three- and six-hours predictions of cumulative rainfall were investigated.The results of rainfall prediction with WRF model using digital filter initialization werecompared with the results of WRF model without digital filter initialization and observedstation data. This comparison showed that the twice digital filter initialization associatedwith the Dolph filter has its maximum effect during first three hours and in the secondthree hours has a minimum effect among other techniques. This means that unwantedfluctuations are eliminated properly during the first three hours. Also, a comparison ofrainfall prediction results with observed station data indicates that the diabatic digitalfilter initialization associated with the Dolph filter has the minimum root mean squareerror. Among digital filter initialization techniques studied, the digital filter launch hassudden effects on the amount of rainfall predicted during the first three hours ofprediction time, so this can induce significant errors in results of the model.