The objective of this proposal is to design, analyze, and implement scalable algorithms for analysis-driven construction of high-resolution 3D terrain models from BIG terrain data sets, and to build a software infrastructure for making analysis-prepared terrain models available to data consumers on multiple platforms. Analysis-driven modeling means that the construction of the model is influenced by, and adapted for, the specific analysis that the terrain model will be used for by data consumers. The algorithms for constructing terrain models will be capable of handling heterogeneous and dynamic data. To handle large volumes of terrain data efficiently, the computational techniques will optimize both the CPU running time and the data communication cost. Models and algorithms will be developed that can construct hierarchical models at different levels of detail. Analysis-driven denoising algorithms, using techniques from persistent homology and machine learning, will be developed to handle noise in the data, and probabilistic models will be developed to handle uncertainty in data and to attach confidence levels to various features computed by the algorithm. Finally, computational methods and software infrastructure will be developed to make terrain models prepared for analysis available to data consumers on multiple different platforms.
Keywords: Terrain modeling and analysis, GIS, LiDAR, big data, scalable algorithms, denoising
