By Manzhu Yu
Dust storms represent serious hazards to health, property, and the environment in arid and semi-arid areas. In order to mitigate the hazardous consequences of dust storms, it is crucial to detect an upcoming dust event and predict its impact. Early warning and decision making processes rely on accurate information. Various dust models have been developed in the past decades to predict dust concentrations with increasingly high accuracy. Scientists and decision makers can simply visualize the model output in either 2-dimensional or 3-dimensional space, and manually interpret the potential impact areas.
With the advancement of spatial computing techniques and capabilities, dust model generated data volume is increasing as well. Operational dust models routinely produce terabytes of data, and the next generation of higher resolution simulations will produce terabytes of data on a per-timestamp basis. Therefore, manual interpretation is no longer adequate. Automated dust storm feature discovery must be conducted through more sophisticated analytical and spatiotemporal data mining methods. The ability to characterize simulation outputs helps rapidly identify features, track changes, and classify specific patterns of dust storms. Moreover, efficiently performing the characterization process results in new insights about prediction models improvements.
Figure 1. Dust storm moving through Phoenix Thursday, July 3, 2014 (Source: CBS 5 News)
The research comprises the use of predicted dust information and other atmospheric status data, which create a valuable resource for monitoring and early-warning information. Decision makers benefit from improved dust storm identification and tracking features, employing storm surface, volume, evolving speed, direction, as well as its evolution phase. In addition, specific types of dust events are classified to correctly inform the public about the characteristics of the dust event.
The proposed innovation is derived from the four-dimensional spatiotemporal analytics work addressing environmental phenomena (longitude, latitude, elevation, and time). The primary methods include image processing algorithm development, tracking algorithm development, spatial quantitative analysis, and data mining techniques, and data sources for dust storm information are generated from an atmospheric forecasting model. Normally, dust storm information is produced from 2D dust concentration maps, which can only indicate the areas of high dust coverage and distribution. However, neither view represent real dust storm objects in three dimensional world, nor capture the movement patterns of dust events. Therefore, this project improves the multi-dimensional and spatiotemporal view of dust storms to increase understanding of dust processes, and enhance the capability of feature identification and movement tracking for natural phenomena.
Figure 2. Analysis procedure of the project
The broader impacts of this research lie in its potential to transform dust model simulation data into direct and intuitive information, which enables better storm hazard mitigation. The project benefits scientists in describing and understanding the evolution and transportation of dust storm over space and time; policy makers obtain early information and design mitigation plans; and the general public receives warning information leading to relevant responses. In addition, by identifying the features and tracking the movement patterns of dust storm objects in four dimensional world, scientists are able to enhance the characterization of the natural phenomena, differentiate multiple types and evolving phases of dust events, and extend, expand, and formalize dust prediction capabilities.
To leverage the abilities of Spatiotemporal Hybrid Cloud Platform, we are storing and parallel processing of dust storm detection from simulation data. Historical simulated dust storm data is stored in storage nodes, while near-real-time forecasted data is also injected. Since the simulated dust storm datasets are in four-dimension (lon, lat, elevation, and time), the processing and handling of these datasets is highly computing intensive. Therefore, we allocate multiple computing nodes to process in parallel to provide the detected information in a timely fashion.
Requesting access to spatiotemporal Hybrid Cloud Platform resources is fairly easy. Simply fill out the online application and submit one page proposal (use your project name as file name) to describe the project objectives. Apply now