This paper presents a real-time, continuous level of detail (LOD) rendering algorithm for height fields. The algorithm uses a compact and efficient regular grid representation and employs a variable screen-space threshold to bound the maximum error of the projected image. It dynamically computes and generates the appropriate level of detail in real-time, allowing for smooth changes of resolution across areas of the surface. The algorithm has been implemented for approximating and rendering digital terrain models and other height fields, and consistently performs at interactive frame rates with high image quality. Typically, the number of rendered polygons per frame can be reduced by two orders of magnitude while maintaining image quality such that less than 5% of the resulting pixels differ from a full resolution image. The algorithm is fast, meets user-specified image quality metrics, and provides for smooth, continuous changes between different surface levels of detail. It is a hybrid of algorithms that combines the large complexity reduction obtained in TIN constructions and the flexibility provided by regular grids, while avoiding many of the drawbacks inherent in these types of algorithms. The algorithm uses a regular grid representation, allowing for localized polygon densities and screen-space error-driven LOD selection. It ensures that the complexity changes smoothly between consecutive frames and that the simplified geometry doesn't lead to gaps or popping in the mesh. The algorithm satisfies all the criteria for a real-time LOD algorithm for height fields, including efficient queryability of mesh geometry, dynamic changes to the geometry not significantly impacting performance, high frequency data remaining local, small changes to view parameters leading to small changes in complexity, and bounding the loss in image quality incurred by the approximated geometry. The algorithm is implemented using a regular grid representation, allowing for recursive "fusion" of polygons where appropriate. It uses a variable screen-space threshold to bound the maximum error of the projected image and provides for continuous level of detail, which is discussed in detail. The algorithm uses a regular grid representation, allowing for localized polygon densities and screen-space error-driven LOD selection. It ensures that the complexity changes smoothly between consecutive frames and that the simplified geometry doesn't lead to gaps or popping in the mesh. The algorithm is implemented using a regular grid representation, allowing for recursive "fusion" of polygons where appropriate. It uses a variable screen-space threshold to bound the maximum error of the projected image and provides for continuous level of detail, which is discussed in detail.This paper presents a real-time, continuous level of detail (LOD) rendering algorithm for height fields. The algorithm uses a compact and efficient regular grid representation and employs a variable screen-space threshold to bound the maximum error of the projected image. It dynamically computes and generates the appropriate level of detail in real-time, allowing for smooth changes of resolution across areas of the surface. The algorithm has been implemented for approximating and rendering digital terrain models and other height fields, and consistently performs at interactive frame rates with high image quality. Typically, the number of rendered polygons per frame can be reduced by two orders of magnitude while maintaining image quality such that less than 5% of the resulting pixels differ from a full resolution image. The algorithm is fast, meets user-specified image quality metrics, and provides for smooth, continuous changes between different surface levels of detail. It is a hybrid of algorithms that combines the large complexity reduction obtained in TIN constructions and the flexibility provided by regular grids, while avoiding many of the drawbacks inherent in these types of algorithms. The algorithm uses a regular grid representation, allowing for localized polygon densities and screen-space error-driven LOD selection. It ensures that the complexity changes smoothly between consecutive frames and that the simplified geometry doesn't lead to gaps or popping in the mesh. The algorithm satisfies all the criteria for a real-time LOD algorithm for height fields, including efficient queryability of mesh geometry, dynamic changes to the geometry not significantly impacting performance, high frequency data remaining local, small changes to view parameters leading to small changes in complexity, and bounding the loss in image quality incurred by the approximated geometry. The algorithm is implemented using a regular grid representation, allowing for recursive "fusion" of polygons where appropriate. It uses a variable screen-space threshold to bound the maximum error of the projected image and provides for continuous level of detail, which is discussed in detail. The algorithm uses a regular grid representation, allowing for localized polygon densities and screen-space error-driven LOD selection. It ensures that the complexity changes smoothly between consecutive frames and that the simplified geometry doesn't lead to gaps or popping in the mesh. The algorithm is implemented using a regular grid representation, allowing for recursive "fusion" of polygons where appropriate. It uses a variable screen-space threshold to bound the maximum error of the projected image and provides for continuous level of detail, which is discussed in detail.