Abstract: Real-time rendering of large-scale snow scenes with high realism has wide applications in the areas such as virtual reality,snowy disaster's prevention and rescue,military simulation and game design,etc.However,current methods cannot simultaneously simulate both the snow accumulation and falling effects for large-scale dynamic snow scenes.To solve this problem,we propose a new method of modeling and real-time rendering.To simulate the accumulation of snowflakes on different objects in the scene precisely,we present an adaptive occlusion map for the falling snowflakes,with respect to the object distribution within the current view frustum.The occlusion map will be updated in real-time allowing dynamic objects to be included in the scene.To simulate the huge amount of dynamically falling snow flakes,we adopt a view-dependent particles grouping technique to reduce the amount of the snowflakes,meantime move the view transform and evolution of snow particle system into GPU for acceleration.To enhance the realism of falling snowflake,we adopt a dynamic texture sequence of multi-rotary snowflake.Besides,a hybrid approach of geometry and texture is employed to render the distant view of large-scale snow scenes with less computation.Based on the above approaches,we successfully implemented the real-time walkthrough of large-scale snow scenes including both urban and suburb circumstances,and generate the beautiful winter view such as heavy snow accumulation on the tree branches and the covers of cars,etc.