| الوصف: |
Precipitation is variable and random in nature and therefore, it has different behaviors in different places and times resulting in uncertainty increase. In this regard, due to the present uncertainties, there are many variations in the amount of precipitation, which makes it difficult to forecast this important quantity. In this study, to reduce the uncertainty and estimate the output of the Cordex database properly, we have used the multilayer perceptron neural network method and the Levenberg–Marquardt educational function. For this purpose, seven monthly parameters including surface temperature, surface temperature 850 hPa, surface pressure, altitude 500 hPa, surface moisture, specific humidity of 850 hPa and durations of sunshine extracted from the Cordex database networks were used as independent input parameters and monthly precipitation of six synoptic stations of Zahedan, Kerman, Bandar Abbas, Chabahar, Iranshahr and Saravan in five climate models CanESM2, CSIRO_Mk, GFDL_ESM2M, MIROC5 and NorESM1_M were separately evaluated as dependent parameters in the artificial neural network in two time series: 1) 1979 to 2005 in the historical model; and 2) 2006 to 2018 in RCP4.5 and RCP8.5 mode. The two models MIROC5 and CSIRO_MK in historical mode and GFDL_ESM2M and CSIRO_MK in RCP modes were selected as the most effective models. Considering the temperature, air pressure and humidity information and given the high correlation of the forecasted outputs, we can utilize the nonlinear artificial neural networks method in order to skew the Cordex database precipitation data for the southeast region of Iran. |