Interactive visualizations of unstructured oceanographic data

Abstract

The newly founded company Oceanbox is creating a novel oceanographic forecasting system to provide oceanography as a service. These services use mathematical models that generate large hydrodynamic data sets as unstructured triangular grids with high-resolution model areas. Oceanbox makes the model results accessible in a web application. New visualizations are needed to accommodate land-masking and large data volumes.

In this thesis, we propose using a k-d tree to spatially partition unstructured triangular grids to provide the look-up times needed for interactive visualizations. A k-d tree is implemented in F# called FsKDTree. This thesis also describes the implementation of dynamic tiling map layers to visualize current barbs, scalar fields, and particle streams. The current barb layer queries data from the data server with the help of the k-d tree and displays it in the browser. Scalar fields and particle streams are implemented using WebGL, which enables the rendering of triangular grids. Stream particle visualization effects are implemented as velocity advection computed on the GPU with textures.

The new visualizations are used in Oceanbox's production systems, and spatial indexing has been integrated into Oceanbox's archive retrieval system. FsKDTree improves tree creation times by up to 4x over the C# equivalent and improves search times by up to 13x compared to the .NET C# implementation. Finally, the largest model areas can be viewed with current barbs, scalar fields, and particle stream visualizations at 60 FPS, even for the largest model areas provided by the service.