Rivers are ubiquitous in nature, and thus an important component of the visual simulation of natural scenes. In nature, rivers are dynamic, therefore animation of rivers is necessary in these visual simulation applications. However, animation of rivers is a challenging problem. It incorporates multi-scale surface details and flow motion, and many phenomena involved have complex underlying physical causes. River animation is particularly difficult in emerging interactive applications like Google Earth or video games, which allow users to explore a very large scene and observe rivers at a very small scale or a very large scale at any moment. Controlling the design of water simulations is another hard problem.

The goal of this dissertation is to achieve real-time, scalable, and controllable river animation with a detailed and space-time continuous appearance. To achieve this goal, we break down the river animation problem into macro-, meso-, and micro-scale subproblems ranging from coarse to fine. We propose appropriate models for each scale that capture relevant surface details and fluid motion. In the macro-scale, we propose a procedural method that can compute the velocities of rivers with curved banks, branchings and islands on-the-fly. In the meso-scale, we propose an improved featured-based simulation method for generating the crests of the quasi-stationary waves that are made by obstacles. We also propose a method for constructing an adaptive and feature-aligned water surface according to the given wave crests. In the micro-scale, we propose the use of wave sprites, a sprite-based texture model, to represent advected details with stationary spectrum properties on flow surfaces. Armed with wave sprites and a dynamic adaptive sampling scheme, we can texture the surface of a very large or even unbounded river with scene-independent performance. In addition, we propose a Lagrangian texture advection method that has other applications beyond river animation.

We demonstrate that combining our models in three scales helps us incorporate visually convincing animated rivers into a very large terrain in real-time interactive applications.