Sven Schade and Alexander C. Walkowski
Institute
for Geoinformatics, University of Münster
Robert-Koch-Straße 26-28, 48149 Münster, Germany
E-mail: schades@uni-muenster.de, walkowski@uni-muenster.de
The traditional paradigm of observing environmental phenomena is based on a small number of fixed sensors. The deployment is carried out using a highly-controlled deployment strategy, which for example defines the deployment location and the calibration parameters. Currently, the observation process is moving from a centralized manner – based on isolated sensors – to a distributed and dense manner of observing phenomena. Distributed sensor networks comprises of large number of sensors spread logically and connected through a communication network. This results in a new data collection scheme, with continuous feeds from dense distributed sensors. A geo-sensor network can be defined as a distributed sensor network that monitors phenomena in geographic space and the geospatial content of the information gathered, aggregated and analyzed is fundamental. We define a mobile geo-sensor network as a geo-sensor network whose sensor nodes are not fixed on a certain location. Instead the sensor nodes are either self-propelled or carried by agents. Such mobile geo-sensor networks provide an unprecedented way of monitoring environmental phenomena. Information derived from such observations will be the dominating source in the future. Relating to these, we expect five revolutions in understanding and handling of geographic information.
In one revolution, interaction with geographic data will drastically change by interacting with models of continues field which encapsulate heterogeneous raw data sources. At present, the gathered data is used in a data centric manner, i.e. being interested in information at a certain location, requests like “Give me the wind direction at 12:00 at point p” are posted. Since it is unlikely that a sensor is directly located at this position, a certain region in space and time is asked in a second step for time series for the sensors in the defined spatial region or – in the case of mobile sensor – for the trajectory of the sensor within the spatial region associated with the observations. It is up to the user to derive the desired information from the provided observation data. Posting the same kind of question to the future model will return the desired information directly. On a high level of abstraction this approach results in a three-tiered architecture. The lowest layer is represented by the sensors gathering observation data of the phenomenon. The topmost layer is defined by the user (application), which is interested in information about the phenomenon. Both layers are bridged by the intermediate model layer.
A wider use of spatio-temporal models will lead to a second revolution where research will tackle dynamics of such models (dynamic spatio-temporal models). The well known approach underlying weather forecast models will be carried over to other phenomena and their properties enabled by arising technology. Models of people being in a city, for example, can be dramatically improved for emergency cases by processing indirect information about peoples’ location. In this example, current static models of cities based on residential data will be extended by including dynamically changing information like traffic flows, connection signals of mobile phones, used cash machines, numbers of transactions made at cash desks in shopping centers and so on. The number of potential used sensor data is endless. Following this approach information models that are always up to the current situation (continuous fields which change overtime, see first revolution) can be provided for lots of phenomena properties.
Considering the arising mobile geo-sensor networks one further field of research will cause revolutionary results: work on delegation of sensors and rearrangements of sensor networks. The classical distinction between in-situ or remote and stationary or moving sensors will be further subdivided into uncontrolled-moving, moving on a fixed track or delegatable. Moving sensors underlie some motion constraints, like a satellite moving on a fixed orbit and being only able to observe certain positions at a given time, a sensor in a car is bound to the street network and the driver’s behavior. Rearranging mobile geo-sensor networks according to variety observation within the model will open new research areas. Algorithms to identify areas with consistent values over time and more varying once need to be developed, and such for optimal sensor spreading (i.e. moving more sensors to areas where high variations in the observation results are detected) to maximize the precision of the model.
A forth revolution will be triggered by including data (and service) quality and will result to a completely new view to geographic information. A phase of renewed interest in the field of error-propagation and new research in this area will be included. In the future geographic information will, for example, look like this: “with a presumption of 90% the wind direction at point p is between 85° and 90°, where the positional accuracy of p is 5 meters”. Only the awareness of such quality information (for the value for the phenomenon’s property, as well as, for the location value) will finally enable to serve a phenomena model covering the complete surface of the earth. Since field models will always include some simulation (as for example an interpolation by Kriging), observation results that pretend to be precise, like “the wind direction at 51°79´21´´N 7°51´03´´E is 89°”, are misleading. Even the results of directly measurements at the supporting points underlie uncertainties due to the sensors` capabilities and to method used for locating them.
In a fifth revolution technical requirements will be dominated by integrating the continuously increasing amount of (micro) sensors, which show heterogeneities in underlying conceptual structures (schemas), used data encoding and protocols. This revolution enables the practical use of the previously presented once. The current trend to enable semantic interoperability gets more attention. For the future, we assume protocol heterogeneities and syntactic issues within schemas solved by standards, whereas semantic heterogeneities between sensors become essential. Within the model layer, the required sources (sensor data or sensor networks) need to be discovered, the specific queries need to be delegated and the results have to be retrieved according to the models internal schema. In order to reach this goal, sensors (and sensor networks) require to be that “intelligent”, to describe themselves semantically. Metadata attached to sensors (and sensor networks) needs to enable discovery even if different languages and/or expressions are used for describing similar content. The required queries to the data sources (for data or for rearranging the network) need to be translated from the models internal query schema to those of the source, and the encoding used by the sources needs to be translated to the models internal schema for observation data. Accordingly, huge amounts of data and meta-data need to be stored and processed, which pushes GRID technology.