This study examines the impact of lateral boundary conditions (LBCs) and initial conditions (ICs) in the prediction of land-falling tropical cyclones (TCs) using Weather Research and Forecasting (WRF) model and its three-dimensional variational (3D-VAR) data assimilation system for the Bay of Bengal (BoB) region. The LBCs are derived from the National Centre for Environmental Prediction (NCEP) Final (FNL) analysis and GFS forecasted datasets at different horizontal and temporal resolutions. The assimilated datasets in the model comprises of NCEP PREBUFR (PILOT, SOUND, SYNOP, METAR, SONDE, BUOY, SHIP and QuikSCAT) and satellite radiances from Advanced Microwave Sounding Unit (AMSU-A, AMSU-B), High Resolution Infra-Red Sounder (HIRS), Microwave Humidity Sounder (MHS) at different assimilation cycles. The impact of the LBCs with varying spatial and temporal resolutions are investigated in detail and the results indicate that the forecasted track during the first 48 h are not affected, however the subsequent forecasted track is significantly influenced by the 3 hourly GFS LBCs. The initial intensity, positional errors, and storm structure show improvements with the assimilation of NCEP PREBUFR and radiance observations at different assimilation cycles. Simulations carried out with improved initial conditions indicate that the forecast (in terms of track, intensity, and trends in the intensification and dissipation) of the cyclonic storm Sidr improved significantly. In addition, five severe land-falling BoB cyclones (Phailin, Lehar, Helen, Madi, and Hudhud) during 2013–2014 were studied using 3 hourly GFS LBCs and with assimilation of NCEP PREBUFR and satellite radiance observations. The results clearly signify the ability of the WRF modeling system in forecasting the track, intensity, and landfall of the storms. The forecasted mean track errors during 2013–2014 cyclones are 65 km, 58 km, 99 km, and 103 km from day-1 to day-4 respectively, while the initial positional errors was about 41 km. The mean landfall time and positional errors are about 3 h and 15 km respectively. The forecasted intensity of the storm in terms of maximum surface wind (MSW) is also reasonably well predicted, and these errors are 4 ms-1, 10 ms-1, 6ms-1, and 8 ms-1 from day-1 to day-4 respectively. © 2018 Elsevier Ltd