<p>This study demonstrates an effective approach to assimilate satellite observations of chlorophyll into coupled physical–biogeochemical model using a non-linear technique to improve model simulations of chlorophyll. For this purpose, ensemble-based particle filter technique, a non-linear technique, is developed to assimilate satellite observations of chlorophyll into the model for the year 2019 for the Bay of Bengal. Non-linear assimilation schemes pose a stringent computational limitation while employing in to a state-of-the-art coupled model. A unique approach is adopted in the study which is the use of the boot strap technique with particle filter to generate particles representing the probability density function of model chlorophyll. This approach has substantially reduced the computational time required for the generation of particles, which otherwise needs a distinct model run for each particle generation in the conventional particle filter technique. Higher weightage is assigned to the particles closer to observations and these strong particles are used to initialize the model for the next time step, while the weaker particles get discarded. Comparison with satellite data shows the efficacy of the data assimilation in correcting model bias significantly. The findings from the study indicate the influence of underlying physical processes and seasons dominating in the region on the efficacy of assimilation to improve on model simulations. Comparison with Bio-Argo floats highlights the significant improvement observed in model surface and sub-surface chlorophyll in both RMSE and bias correction. An 18% improvement in RMSE and ~ 23% improvement in bias correction is observed for surface chlorophyll, while sub-surface chlorophyll within the upper 40&#xa0;m water column showed an improvement of 15% in RMSE and ~ 45% in bias correction when compared with Bio-Argo buoy data.</p>