Research in Real-time Analytics of Continuous Phenomena

Over the last several years, the focus of our research has been on supporting the data management and real-time analytics of massive amounts of sensor data streams (>100K sensor) that continuously and in real-time sample environmental continuous phenomena (e.g. a toxic plume, radiation over Japan, air pollution, etc.). These sensors stream their samples directly to a centralized server or the cloud (the in-network aspect is not relevant is this part of our research). However, the question of how to provide appropriate spatio-temporal data management strategies for such massive numbers of high-frequency sampling sensors remain. How can all sensor values be integrated and analyzed in near-realtime? How can users be shielded from having to manage 100K streams individually but can work with higher-level abstractions?

We believe that data stream engines (DSE), a high-throughput version of database systems, used for real-time financial analysis, credit card fraud detection and traffic management, are capable of providing the necessary throughput for this need. However, today’s research and commercial DSE provide little data model, query language and data stream operator framework support for spatially and temporally continuous phenomena.

Currently, our research focus is on algebraic data model and query language extensions. Further, we have published several papers that investigate the feasibility of DSE for achieving the spatio-temporal interpolation of fields via up to 250K sensor samples in 1-2 seconds, and a throughput of a field per second, making it possible to visualize such phenomena changes in real-time. We also investigated a scalable data stream operator framework for spatio-temporal interpolation since sensors likely update not synchronously. 

S. Nittel, Q. Liang and J.C. Whittier.:Real-time Spatial Interpolation of Continuous Environmental Phenomena using Mobile Sensor Data Streams, SIGSPATIAL 2012, Nov 5-7 2012, Redondo Beach, CA.

J..C. Whittier, S. Nittel, Q. Liang and M.A. Plummer, Towards Window Stream Queries over Continuous Phenomena, 4th International Workshop “Geostreaming” in conjunction with SIGSPATIAL 2013, Orlando, FL.


Research in Geosensor Networks

A main part of our research has been conducted in the area of wireless sensor networks deployed for geographical and environmental applications (so-called geosensor networks). An survey paper of the state of the art can be found here.

Mainly, our research focus is on monitoring and tracking continuous phenomena such as air pollution, pollen distribution in the air and features within those phenomena (like regions with measurements above a threshold that can be toxic to humans), and tracking those events over space and time.

Several aspects are relevant with regard to continuous phenomenon monitoring:

  • we want to measure the phenomenon
  • we want to identify features
  • we want to track those features over space and time.

Due to the energy constraints in geosensor networks, which are battery driven, most processing has to be performed within the geosensor network between neighboring nodes and spatially constrained around events, avoiding any unnecessary computation and communication between nodes.

Area 1: Estimating Spatial Window Queries over Continuous Phenomena

To retrieve metric information about continuous phenomena these phenomena are queried using spatial queries. For example “Select pollenDistribution from sensors where sensor.location INSIDE polygon P” is a typical spatial window queries. Since sensor samples are discrete the query result of the distribution of the phenomenon over a region has to be spatially interpolated. We have investigated reformulating several spatial interpolation methods to be executed as an in-network algorithm (kriging, Gaussian kernel estimation). The work can be found in the following papers:

Kriging: S. Nittel, G. Jin, Y. Shiraishi,  In-Network Spatial Query Estimation in Sensor Networks,  IEICE Transactions (A), Vol.J88-A, No.12, pp.1413-1421, December 2005.

Adaptive clustering: G. Jin and S. Nittel: UDC: A self-adaptive uneven clustering protocol for dynamic sensor network, International Conference on Mobile Ad-hoc and Sensor Networks (MSN), Wuhan, China, 13 – 15 December 2005.

G. Jin and S. Nittel: UDC: A self-adaptive uneven clustering protocol for dynamic sensor network, International Journal of Sensor Networks, Vol 2. No 1/2, 2007, pp 25-33.

Gaussian Kernel Estimation: G. Jin and S. Nittel: Towards Spatial Window Queries Over Continuous Phenomena in Sensor Networks,  IEEE Transactions on Parallel and Distributed Systems (TPDS), Vol 19(4), pp. 559-571, April 2008.

Area 2: Qualitative Monitoring of Continuous Phenomena in Geosensor Networks

Often, geosensor network applications are deployed not only for monitoring purposes but also to alert when events such as toxic spills or toxic plumes at a chemical or power plant happen. In this case, we are less interested in the actual measurements, but we can abstract to the events themselves and their behavior over space in time (i.e. the parts of the phenomenon that are dangerous to humans). For these types of applications we have investigated using a qualitative approach to detect and monitor continuous phenomena. We can significantly reduce message size by encoding quantitative measurements (e.g. 1000 degree) via a limited set of qualitative values (e.g. 0=cold, 1=warm, 2=hot, 3=extremely hot). Each sensor converts its measurements into qualitative information, and event detection and tracking can be based on this type of information, thus, saving a significant amount of energy during in-network communication.

M. Duckham, S. Nittel and M. Worboys: Monitoring dynamic spatial fields using responsive geosensor networks, ACM-GIS 2005, Bremen, Germany, November 2005.

G. Jin and S. Nittel: NED: Efficient Event Detection in Sensor Network Workshop “Mobile Location-Aware Sensor Networks”, in conjunction with MDM, Nara, Japan, May 13 2006.

Area 3: Tracking Continuous Phenomena over Space and Time in Geosensor networks

Once an event such as a toxic plume is identified within a geosensor network, algorithms are necessary to track the phenomenon’s dynamic changes. We explored two different approaches: 1) representing the boundary of the event as a polygon and incrementally tracking the localized changes to the geometry over space and time with a force-based model to adjust the optimal observation coverage, and 2) tracking the topological changes of the phenomenon (did it split, merge, develop holes?).

Geometric change tracking:

G. Jin and S. Nittel, Supporting Spatio-Temporal Queries in Wireless Sensor Networks by Tracking Deformable 2D Objects,  ACM-GIS, Los Angeles, CA, Nov 2008.

G. Jin and S. Nittel, Efficient tracking of 2D objects with spatio-temporal properties in wireless sensor networksJournal of Parallel and Distributed Databases, Vol 29(1-2), pp.3-30. February 2011

Topological change tracking:

C. Farah, C. Zhong, M. Worboys and S. Nittel, Detecting Topological Change using Wireless Sensor NetworksGIScience, Park City, Utah, September 2008.

J. Jiang, M. Worboys and S. Nittel, Qualitative Change Detection in Sensor Networks based on Connectivity InformationGeoinformatica, Vol 15, Issue 15(2), 2010, pp.305.

To be continued with our research in Mobile Geosensor Networks and our work on Sensor Data Stream Management…