Successful forecasting of the number and/or intensity of tropical cyclones at subseasonal lead times would have tremendous societal benefits. It is generally accepted that both the ocean and the atmosphere are important for the formation and intensification of tropical disturbances; land surface conditions have also been associated with the modulation of tropical storm activity e.g., soil moisture over tropical Africa. It follows that comprehensive coupled numerical models are a good option for operational numerical subseasonal forecasting of TC activity (week 3 to month 2). However, how realistically tropical systems are represented by coupled models and which resolutions are optimal for such long range forecasts is still unknown. What is the response of the ocean model to such a forcing? Is coupling with the ocean model producing spurious effects? Here we try to shed some light on these questions by investigating the evolution of named tropical storms initialized at different phases of their lifecycles. We further check the evolution of model generated storms as a function of resolution and initialization by using the storm tracker developed at NCEP. For this study we use a series of 60-day hindcast that we performed under NOAA's Climate Test Bed (CTB) in which we ran the NCEP Climate Forecasting System (CFS) at three horizontal resolutions. These resolutions are: the current operational resolution of T62 (200 km x 200 km), the next operational implementation of the CFS at T126 (100 km x 100 km) and an experimental version at T254 (50 km x 50 km). These hindcasts were initialized by Reanalysis-2 and by the NCEP operational analysis (GDAS) every 5 days apart from May 23 to August 11th from 2002 to 2006.