عنوان مقاله [English]
نویسندگان [English]چکیده [English]
The outputs of General Circulation Models (GCMs) which are available for the users based on predefined scenarios have a low spatial resolution. Hence, downscaling techniques should be used for regional studies. Since climate is an effective factor in natural phenomena, a time series of future weather data is required for meteorological, agricultural and hydrological prediction or pre-warning applications. It is also important to select and utilize better and more accurate techniques for theses purposes. Generally, downscaling methods are classified into two categories: dynamical and statistical. The statistical downscaling is commonly considered due to its simplicity and wide applicability. As an example, LARS-WG is a parametric or semi-parametric model which has been used widely, but it underestimates monthly variances. Undoubtedly, it is more useful to use an approach having a non-parametric structure so that it does not rely on a statistical structure by default. These models use a set of observations in the simulation process not a certain value namely the "parameter".
The aim of this study was to use a nonparametric approach for downscaling the GCM outputs. This approach is composed of one weather generator (WG) and a technique called strategic re-sampling which creates series that match the GCM output. The weather generator itself is based on a Kernel Density Estimator (KDE) and it is a multivariate weather generator. In the KDE method, all of the observations with a definite kernel function, commonly standardized and normalized, are used. Firstly, one of the normal kernels (with probability 1/n) is selected randomly and its mean is considered as the basic vector. Bandwidth (h) is the only parameter that should be estimated. The strategic re-sampling method includes a stochastic function based on the definition of the "shape parameter", which determines the tendency of the new series. At the first step, a strategic re-sampling is run and then strategic series are prepared as input to the WG and the simulating climate prospect.
The study area is Karkheh Basin in Khuzestan Province, Southwest of Iran. Downscaling was done for two periods, 30 years (2011 to 2040) and 60 years (2011-2060) for monthly rainfall and air temperature variables based on the A1B scenario of the CGCM3T63 model. The results can be divided into three groups: an estimate of the strategic re-sampling parameter, evaluation of the weather generator applicability and finally, the climate change simulation. Results showed that in case of temperature, by selecting a unit value for the shape parameter, the generated series coincide with the observed or historical series. Substituting the values less or more than one resulted in warmer and colder simulated series, respectively. Similarly, for rainfall series the optimum value was 0.9. Accoring to the results, the ability of the WG in simulation of moments of different orders (mean, variance and skewness coefficient) was acceptable and the coefficients were cross validated. The applied GCM showed warmer and drier series for both study periods. The findings of the study revealed that future climate would be simulated accurately and non-significantly different from GCM outputs. In general, the suggested non-parametric approach can be recommend due to the following features: non existence of a default pattern in its structure, the least number of parameters for running, coincidence of the high accuracy in downscaling with the GCM outputs and simplicity. More case studies are recommended for further scrutiny.
Keywords: GCMs, re-sampling, KDE, Karkheh basin, downscaling