Abstract. The current assessment of the ice sheet surface mass balance and more generally of the atmospheric branch of the high latitude water cycle mostly relies on climate model simulations. The ability of climate models to reproduce polar precipitation depends not only on the simulation of the atmospheric dynamics and on the advection of moisture towards the poles but also on the representation of the subgrid-scale cloud and precipitation processes that govern the formation and growth of snowflakes and raindrops. The LMDZ atmospheric general circulation model, atmospheric component of the IPSL-CM climate model, is intensively involved in polar-oriented studies and recent developments were carried out to improve the representation of mixed-phase and ice clouds. However, recent studies also highlighted substantial shortcomings and biases that persist in the simulation of polar precipitation, both in the Arctic and in Antarctic. This study presents the development of a new precipitation scheme for LMDZ model that includes both an advanced microphysical treatment of snowfall and subgrid considerations to properly account for the interactions between hydrometeors and clouds. The scheme has been designed with a moderatecomplexity approach, retaining strong approximations such as stationarity while avoiding additional prognostic variables that would require transport, thereby limiting its computational cost and making it suitable for global climate simulations and easy to tune. Particular attention has been paid to the numerical treatment of the different processes to ensure numerical convergence and stability at typical time steps used in global climate models. The scheme is then evaluated using regional simulations conducted over the Svalbard Archipelago and over Adélie Land, East Antarctica. The cloud water contents are assessed using airborne radar and lidar data collected during the RALI-THINICE campaign and the simulated vertical profiles of precipitation and microphysical tendencies are compared with observational data from a ground-based polarimetric radar deployed during the APRES3 Antarctic campaign. Perturbed parameter ensemble experiments are also conducted to assess the parameteric sensitivity of the model and to disentangle calibration issues from genuine structural biases. Results show that the model is now able to physically capture the vertical evolution of snowfall and to simulate more realistically the melting layer in the Arctic case study. Future applications to simulate more reliably the present and future Antarctic surface mass balance with LMDZ as well as avenues for improvement particularly regarding some microphysical approximations can now be envisaged.