Real-time numerical simulation of tropical cyclone Nilam with WRF: experiments with different initial conditions, 3D-Var and Ocean Mixed Layer Model

Abstract

In this study, real-time predictions of the cyclone Nilam in the Bay of Bengal during October 27, 2012–November 01, 2012, using ARW mesoscale model are presented. The model is run with two nested domains (27 km × 9 km resolution) using the NCEP GFS 0.5° × 0.5° analysis and three hourly forecasts. A series of simulations with different initial conditions varying from 00UTC28Oct to 12UTC30Oct indicated that the prediction errors in respect of the vector track position and intensity are reduced with initial conditions corresponding to 12UTC29Oct. The error reduction is attributed to the better representation of cyclone vortex in the initial conditions corresponding to depression stage. Further experiments with the use of three-dimensional variational data assimilation (3D-Var) and Ocean Mixed Layer Model (OML) and comparison with a control run (CTRL_RUN) indicated that the 3D-Var and OML experiments improved the track and intensity predictions. While the 3D-Var experiment well predicted the overall track with lesser along-track errors, the OML experiment predicted more accurate landfall with an error of 55 km in landfall position and 5 h in landfall time. All the experiments simulated early landfall. It has been found that intensity of simulated cyclone is highest with CTRL_RUN followed by OML and 3D-Var experiments. While OML simulated a stronger cyclone for a brief period, the 3D-Var simulated a less intensive cyclone for a prolonged period in better agreement with IMD estimates. The improvements in track and intensity prediction using 3D-Var and OML are attributed to the improvements in initial conditions with observation assimilation in the former and to better representation of air–sea interaction (fluxes) in the latter. The overall better performance of the 3D-Var experiment is evident from the comparisons of flow-field, rainfall and maximum reflectivity parameters.