Dynamical extended range prediction is subjected to various sources of errors, of which the errors arising from the uncertainties in the initial conditions and model are important. To eradicate such errors, ensemble prediction approach is one of the best options and has been attempted by various operational centers [Buizza and Palmer 1995, Toth and Kalnay 1993, Houtekamer et al. 1996, 2005]. Though there are several approaches to generate ensembles of different initial conditions, we use an approach [Abhilash et. al. 2013a] which is similar to the 'complex-and-same-model environment group' as classified in Buizza et al. [2008].

For the details on perturbation technique used in this study refer Abhilash et al., 2013a, b. Since the skill and spread of an EPS essentially depends on the ensemble size [Richardson 2001, Reynolds et al. 2011], an ensemble of 10 perturbed atmospheric initial conditions has been developed in addition to one actual initial condition keeping in mind the huge computational power required to run large ensemble members in real-time basis. It has already been observed that there is not much difference between the perturbed and the actual analysis field over most of the parts except in the extra tropics where wind field is higher [Figure 1, Abhilash et al. 2012].