Document Type : Full length article
Associate Professor, Department of Irrigation and Reclamation Engineering, University of Tehran
Assistant Professor, Climate Change Department, Climatological Research Institute of Mashhad
Associate Professor, Atmospheric Science and Meteorological Research Center
PhD Candidate of Agricultural Climatology, Department of Irrigation and Reclamation Engineering, University of Tehran
The main perspective in seasonal prediction of precipitation is presentation of a qualitative prediction for upcoming seasons. The information gained from such predictions can be used for decision making in various disciplines such as agriculture, water management and hydropower production. Besides, it can help us reduce the adverse effects of climatic changes like drought and flood. But General Ciculation Models (GCMs) outputs have coarse resolution (>100 km). Dynamic downscaling is a method to obtain high-resolution climate data from relatively coarse resolution global climate models which do not capture the effects of local and regional forcing in areas of complex surface physiography. GCMs outputs at spatial resolution of 150-300 km are unable to resolve important sub-grid scale features such as clouds and topography. Many impact models require information at scales of 50 km or less. Several statistical and dynamical methods are developed to estimate the smaller-scale information. Dynamical downscaling uses a limited area, high resolution model (a regional climate model: RCM) driven by boundary conditions from a GCM to derive smaller scale information.
The aim of this study was application of RegCM4 dynamic model (Reginal climate model) in forecasting rainfall and improving the outputs using post processing techniques in northwest Iran during period 1982-2011. The recorded data of precipitation were collected from Urmia, Tabriz, Ardebil and Khuy Stations. The data required for running the regional climate model RegCM4 were obtained from center ICTP (International Centre for Theoretical Physics), in the format of NetCDF including three sets of weather data, NNRP1 with a 6-hour-scale and a horizontal resolution of 2.5×2.5 on the reanalysis databases of National Center of environmental prediction of United States, sea surface temperature, (SST) with a horizontal resolution of 1×1 from the type of SST belonged to America and National Oceanic and Atmospheric surface SURFACE, which were consisted of three topographic data of GTOPO, the vegetation or land use, GLCC, and the soil type data GLZB, with a horizontal resolution of 30×30 seconds from United States Geological Survey. These data were organized for the period 1982 to 2011. In order to execute the dynamic model, a test was conducted to determine the Convective Precipitation scheme and the amount of horizontal resolution for the year 2009 (as a normal year). Accordingly, Kuo scheme with minimum mean bias error (MBE), in comparison with the observed precipitation in 36 synoptic stations of the region, was implemented as the main scheme with horizontal resolution of 30 × 30 square kilometers. The number of grid points including 152 in longitude (iy) and 168 in latitude (ix) was conducted during the statistical period of 1982 to 2011. Geographical area center implemented in the intended period was located in 30.5 degrees north latitude and 50 degrees east longitude. The output of the model included atmospheric data (ATM), surface cover (SRF) and radiation cover with the format of NetCDF, each containing a large number of meteorological variables among which, except precipitation from the Model (tpr), 9 variables that were associated more with precipitation were extracted. These variables are including q2m and t2m, ps, v1000, v500, u1000, u500, omega1000, omega500. For post-processing the output of the model we used the Multi-Layer Perceptron (MLP) and Moving Average (MA) methods. For MLP Entering variables were those 10 aforementioned and the target variable was observatory precipitation in the stationary point. At any one time, 80% of the data at the beginning of the series were for train and final 20 percent of data was used for test.
Results and Discussion
The results of the study demonstrated that, in the study area, the mean bias error of raw annual precipitation outputs of the RegCM4 model was 124.3 mm in the validation period, which by conducting Post Processing it was reduced to 8.9 mm. In the seasonal and monthly time scales, also, mean bias error were 31.1 and 10.4 mm, respectively, which were reduced to -0.3 and zero mm, respectively, after post processing. The MA model was the preferred post processing method, in all time scales.
In conclusion, it can be stated that the RegCM4 regional climate model with the implementing conditions and in the study area, contained mainly overestimate in precipitation forecasting. However, the application of post-processing will optimally reduce bias. The appropriate method is also the simple moving average (MA) method.