Projects > Cloud Computing Initiative
Due to the recent development in geographic sciences, massive data are available hosted by data vendors, which calls for a transparent and scalabe visualization system.
With the emerging cloud computing and ubiquitous computing, new paradigm enabling fast interactive the spatial temporal visualization can be formed for different research communities. The new framework should handle the distributed data and computing sources, which also be able to adapt to various clients in a collaborative environment. At the same time, under the regulation of spatial principles, these techniques should be transparent and easy for scientists from Earth Sciences to develop their own visualization system and visual analytical tools.
Our center proposes a virtual globe based visualization framework to visualize large scale Earth Science data. The NASA's World Wind Virtual Globe is used as the multidimensional visualization platform to promote the understanding of scientific modeling. To enhance the ability to visualize massive data, we introduce octree based data organization as well parallel processing in distributed computing environment. Considering the limitation of bandwidth, progressive transmission strategy is developed as well.
The project starts from June 2009, and is funded by NASA GIO managed by Ms. Myra J. Bambacus, and the project is managed by Dr. Chaowei Yang.
To demonstrate the capabilities of this framework, we show three types of time enabled visualization based on the output by WRF (Weather Research and Forecasting) -NMM (Nonhydrostatic Mesoscale Model) model. 1)Display the transformation of climatic parameters, such as the dust density, from the ground surface to a certain height continuously, 2) show climate evolution clearly with time lapses, 3) simulate comprehensible climate change process with animations generated simultaneously by selecting interested routes and 4) volume rendering of dust storm. The development has been integrated into spatial web portals to support access by the public.
For more information please visit http://cischpc.gmu.edu.
2.1 System Architecture
Figure 1 shows a general architecture of distributed visualization with Virtial Globe. The server sides is responsible for the data generation, prepocessing and organization with support of cloud computing resources. The client side is responsible for the display or local visualization. The massive data are first processed at the server side and then send to the client in the form that the viewer of the client can display.
Currently, our work focuses on the 4D visualization of massive scientific data , where an integreated server is present. Beyond the computing enviroment, clients are able to view the visual analytical results in a timely fashion.
Figure1 System architecture of distributed visualization with Virtial Globe
2.2 Data organization and study area
Dust storm data produced by Nonhydrostatic Mesoscale Model (WRF) - Weather Research and Forecasting (WRF) are used to demonstrate the our work. The spatial coverage of data is from 25.560N to 41.480N along the latitude direction and from 123.000W to 96.510W along the longitude direction. The most southwestern states of United States are located in this region. In this area, due to the high possibility of dust storm, the observation dust storms are of highly environmental concerns.
The original data from the model are stored in NetCDF files. These array oriented datasets also have spatial information associated with them. Due to the characteristics of the data structure, a regular octree is introduced to organize the 4D data, The octree structure is typically used to compress data and construct multiresolution model. In this case, the multiresolution model can reduce the transmission and visualization intensity by a Level of Details(LOD) mechanism.
Figure 2 Study area of dust storm simualtion
2.3 Progressive transmission
With the multi-resolution model, the progressive transmission is implemented. The transmission process typically starts from lower resolution to higher resolution. A progressive rendering process is implemented while the visualization intensity is under controlled. The levels of the octree correspond to the progressive transmission level in general. The node and the leaves of the octree are interactively inquired during the transmission process.
2.4 Multithreading processing with multicore machine
The parallel strategy makes use of available computing resources through assigning processing tasks to different computing units. This is critical when large scale data and computing resources are present. Under such circumstance, octree along can not meet the requirement of organizing increasing volume data. Therefore, based on the effective data organization, the task is further decomposed into several subtasks and distributed to different sources. In this way, the massive data can be handled and visualized.
Our current solution is to distribute original data into several data blocks and process each data block respectively. In this process, two bottlenecks may be encountered, which are the data overloading and excessive visualization intensity. So before each bottleneck occurs, a decomposition process is done to distributed data to computing resources. Thus the double multithreading processing promises a fully use of shared facilities.
2.5 Client visualization
Three types of visualization are developed to help researchers to understand and analyze the geospatial data. There are surface view providing information at the same pressure or elevation layer, vertical profiles showing the data along selected routes and volume representation replicating the dust storm event on the Earth. The first two types of visualization have less rendering intensity than the volume rendering, which could be implemented at the server side. The volume rendering is locally implemented in this framework, which will be further distributed in the next phase.
3.1 Progressive transmission
According to the storage part, it is clear that from the coarser level to finer level, the data storage increases. The storage for Level 1 data is about 3KB, which is the coarsest level. By contrast, Level 6, as the finest level, costs more than 4MB memory space.
In terms of transmission speed, both wired and wireless based transmissions are done to test the performance of transmission. The test environment is internet with a speed of 23.3 mps also the wireless with a speed of 18.5 mps. As the data is hosted by the data sever in our lab in the university, the test is done by off-campus access. Table1 shows the transmission time for each level of data. The average transmission speed for wireless is about 28KB per second and for wired is about 79KB per second.
For the first four levels, there are no significant differences in terms of transmission time, which are less than 1 seconds. When it comes to the fifth and the sixth level, a discernable increase in transmission time is observed.
Table 1 Progressive transmission results of the NetCDF file
3.2.1 Surface view
Surface view exhibits the changes of dust density across the region of the same pressure layer. Figure illustrates the dust density at the lowest pressure level, which is the closest to the earth surface. The highest density of dust zones denoted by red color appear in while lowest density denoted by the blue color .Combing with the based map provided by the World Wind, these low density areas are part of New Mexico, Texas in the United States and northern Mexico.
Figure 3 Surface view of dust storm density data
Figure4 shows time series pattern of the dust storm through a vertical transaction. According to the figure 2, the density changes as elevation increases. Traversing the upper pressure layers, the density of dust storm decreased to zero. The vertical profile indicates the particular of dust storm is more likely to accumulate at lower layers and dissipate at upper layers. Similar to the surface view, the high density accumulates in northwestern area, which tends to be a center of the dust storm. And the value of density gradually decreased as the distance to the center increased. The six frames with an initial time of. According to the figure, the high density first appeared at the middle part of the routes and then moved to the end of the routes. This informs the transport process of dust particles to some degree. In addition, in figure (e), some dust particles were predicted in the upper pressure layers but few particles were found at their lower layers
Figure 4 Vertical view of dust density along selected routes
Besides profile of data, the dust storm simulation through the volume rendering is another way to model the dust storm. This allows a complete view of dust storm particles. The figure below shows the 4D visualization in the World Wind. It is clear that with time changes, the particles moves forward.
Jing Li, Huayi Wu, Chaowei Yang. and David.Wong, Octree-based LOD for dust storm simulation in Virtual Globes.(Submitted to Computers and Geosciences)
Jing Li, Zhenlong Li, Jibo Xie, Qunying Huang, Wenwen Li, and Chaowei Yang. (2008). "Geo-visualization for Geosciences data in World Wind". Eos Trans. AGU,
89(53), Fall Meet. Suppl., Abstract IN41A-1127(Poster)
Jing Li, Huayi Wu, Chaowei Yang, Jibo Xie, Zhenlong Li and Qunying Huang. (2009). "Progressive transmission of 3D/4D geospatial data over the internet to facilitate Geo-visualization in World Wind", Geoinformatics 2009, Farifax, USA(Oral)