Note: This page is a couple years out of date, I hope to update it soon! Please refer to my publications page for up to date references.
This research with TerraSpark Geosciences is focused on computer aided stratigraphic interpretation of seismic volumes using visualization and image processing techniques. This research is being accomplished by myself and other researchers at TerraSpark and is advised/managed by Geoff Dorn.
We examine the performance and scalability characteristics of the Los Alamos Sea Ice Model (CICE). The CICE model is being considered for inclusion as the Sea ice model in the next version of the Coupled Climate System Model (CCSM). In particular we evaluate what type of changes may be necessary to enable efficient execution of the CICE model at 0.1 degree resolution on computer systems with large numbers of processors like the IBM Blue Gene/L. We will examine an alternative partitioning technique which uses a weighted space-filling curve to load balance the CICE model. We will also look at alternative data-structures to reduce computational cost within the thermodynamic component of the model. This work is a component of a systematic approach to improve the scalability of the CCSM model in preparation for upcoming Petascale systems.
I have worked on the Nekton (NEK5000) Spectral Element CFD code with Henry Tufo at the National Center for Atmospheric Research. My job was to enable the code for automatic isosurface extraction of simulation results when computed on a massively parallel system.
Daniel Goldstein is simulating decaying incompressible isotropic turbulence using a stochastic coherent adaptive large eddy simulation approach (SCALES). This method uses a wavelet filter to resolve the coherent energetic sturctures in the 3-D turbulent flow field. The wavelet coefficients essentially track the turbulent regions in the simulation: larger coefficient levels describe the more turbulent flow. The goal of this visualization is to track the turbulent regions of the flow by seeing where high wavelet coefficients exist throughout the simulation. These coefficients appear in each timestep of the simulation and are scaled and colored according to their magnitude and level using spheres. Regions of high turbulence in the flow can be modeled using this visualization.
The Changbaishan (Baitoushan) volcano, located at the norththeast frontier of the North China craton, is the highest mountain in northeast Asia. This rhyolite stratovolcano is far from the Pacific Plate subduction zone, but the subducting slab extends into the interior of the Asian continent. A NNW deep earthquake strike belt, with source depth between 500-600km, is situated 250-300km northeast to the volcano. There are other Cenozoic volcanoes in the area, such as Tudingzishan, Xiaobaishan, Baotaishan, Lufeng, Huangfeng, which form a NW volcano belt nearly parallel to the 500km deep earthquake belt. In 1999, an integrated monitoring system was established in Changbaishan Volcano Observatory to observe the seismicity, surface deformation and geochemical variations. We have analyzed this data at a short timescale and using cluster analysis. When looked at across a span of 5 years there can be seen a relationship with clusters of seismic events grouped at a small time scale (in the order of a few years) which signal a precursor to impending eruptive events.
WEB-IS, for Web-based Interrogative System, allows remote, interactive visualization of large-scale 3-D data over the Internet, along with data analysis and data mining capabilities. WEB-IS allows users to navigate through their rendered 3-D data and interactively analyze the data for statistics or apply data mining techniques, such as cluster analysis. We take advantage of a client-server paradigm in order to keep the processor intensive tasks, such as visualization and data mining, on the server-side while the client provides the front-end interface. WEB-IS uses a combination of CORBA, Java, C++ and Python to seamlessly integrate the server-side processing and user interaction utilities on the client. The server renders 3-D data off-screen and sends the resulting image buffer data over the Internet to a Java applet on the client to be displayed. In turn, the client responds to user interaction through requests sent to the server for additional visualization or data analysis tasks. In this way, the user can interact with the client from a web browser located thousands of miles from the server and the data, but view the results as if they were occurring locally. Using this tool, researchers from around the world can visualize and analyze large-scale 3-D datasets from any web browser with a current Java1.4 plug-in (http://java.sun.com)
We employ agglomerative clustering methods intended for feature extraction and studying the predictions of large magnitude earthquake events. Data-mining is accomplished using a mutual nearest meighbor (MNN) algorithm for extracting event clusters of different density and shapes based on a hierarchical proximity measure. Clustering schemes used in molecular dynamics [Da Silva et. al., 2002] are also considered for increasing computational efficiency using a linked cell algorithm for creating a Verlet neighbor list (VNL) and extracting different cluster structures by applying a canonical backtracking search on the VNL. Space and time correlations between the events are visualized dynamically in 3-D through a filter by showing clusters at different timescales according to defined units of time ranging from days to years. This WEB-IS functionality was tested both on synthetic [Eneva and Ben-Zion, 1997] and actual earthquake catalogs of Japanese earthquakes and can be applied to the soft-computing data mining methods used in hydrology and geoinformatics.
Poster: http://www.msi.umn.edu/~kadlec/papers/03pha.pps
