Short range precipitation forecasts evaluation of WRF model over IRAN

Iran has a complex topography and it consists of rugged, mountainous rims surrounding high interior basins. Because of this condition, in some cases the NWP output has a significant error from mesoscale variations induced by the diverse topography.
 Iran, covering an area of about 1,648,000 km², is located in the southwest of Asia approximately between 25° and 40° N and 44° and 64° E. This is predominately a semi-arid to arid region surrounded by Caspian Sea to the north and Persian Gulf to the south and crossed by the impressive Zagros and Alborz Mountains. However, for a next-generation mesoscale forecast model (the Advanced Research Weather Research and Forecasting model, ARW) developed by NCAR (Skamarock et al. 2005), the performance of this model employed in the operational forecasts over Iran is not fully tested.
A few previous model studies (Evans and Smith 2001; Evans et al. 2004; Zaitchik et al. 2007; Marcella and Eltahir 2008; Xu et al. 2009) provided some interesting results for the basic weather simulation in SWA using a regional climate model [the second-generation National Center for Atmospheric Research (NCAR) Regional Climate Model (RegCM2)] or the fifth generation Pennsylvania State University–NCAR Mesoscale Model (MM5) model. They pointed out that the regional model has difficulty producing an accurate simulation of meteorological variables in certain sub regions this includes an accurate description of storm tracks, topographic interactions, and atmospheric stability.
On the other hand, verification is a critical component of the development and use of forecasting systems. Ideally, verification should play a role in monitoring the quality of forecasts, providing feedback to developers and forecasters to help improve forecasts, and provide meaningful information to forecast users to apply in their decision-making processes.
One of the purposes, in this study is to evaluate the performance of the WRF model in the complex terrain of Iran. This evaluation primarily concentrates on the precipitation forecasts. In this paper, Real-time gridded 24h and 48-h precipitation forecasts from NCAR models (the Advanced Research Weather Research and Forecasting model; WRF model) are verified over Iran from one to 28 February 2007. Network has 209 * 194 points, which have the center point 54E longitude and 32N latitude and the horizontal resolution of main domain is 21km. The observed precipitation data were taken from the FNL (Final Operational Global Analysis data) reanalysis data with horizontal resolution 1°×1°.
All forecasts are mapped to a 21km latitude–longitude grid and have been verified against an operational precipitation analysis (more than 100 rain gauges), mapped to the same grid. In this study, first we will describe the forecasting errors of the WRFARW model for precipitation. Then, we will introduce some techniques for evaluating the forecasts. These techniques are particularly designed to examine the difficult features of the precipitation fields.
Forecasting weather variables by numerical models has a discrete nature. Assessment of discrete variables such as precipitation is examined by discrete evaluation techniques and evaluation of quantitative precipitation forecasts (QPF) is a challenging matter because of the noisy, discontinuous and non-normal nature of precipitation.
Common method for evaluation of quantitative precipitation forecast is a categorical procedure that is essentially based on contingency tables. X observations and Y forecasts convert to dichotomous events based on both of them more than a constant precipitation rate threshold ‘u’. Then, the behavior of binary categorical verification scores from contingency tables in some rainfall thresholds has been evaluated. Common measures of the binary events evaluation are hit rate, false alarm ratio, false alarm rate, skill scores, correct ratio etc.
Some measures from Signal Detection Theory (SDT) such as area under the ROC curve and discrimination distance is used. SDT offers two broad advantages. Firstly, it provide a means of assessing the performance of a forecasting system that distinguishes between the intrinsic discrimination capacity and the decision threshold of the system. The main analysis tool that accomplishes this is the relative operating characteristic (ROC) that is a graph of hit rate against false alarm rate as u varies, with false alarm rate plotted as the X-axis and hit rate as the Y –axis.Secondly, SDT provides a framework within which other methods of assessing binary forecasting performance can be analyzed and evaluated.
Due to the complexity of the Iranian plateau and lack of knowledge in the estimation of the physical processes in this area, forecasters should have greater awareness of these limitations of the model when forecasting in this region.
Examining the calculated scores for QPF, show that the WRF model correctly estimates the general pattern of precipitation bands but there are problems in the actual rain value. Moreover, skill scores for different thresholds on total investigated area for one month period and in the intensity activity of synoptic systems days show good performance of the WRF model for estimating precipitation in most areas. In addition, using ROC curves gives a measure of performance in all thresholds. For 0.1mm rainfall threshold at selected synoptic stations, model estimates rainfall frequency properly and skill scores is desirable, although precipitation rate estimates still have problems. In addition, the verification scores of model in estimation of quantitative precipitation in 24 forecasts are better than 48h forecasts. The results suggest that improvements in initialization may be as important, or more so, than improvements in the physics for the land surface processes.