Monitoring environmental phenomenon with a network of ground-based sensors can alleviate the cost of repeated field surveys and provide a consistent temporal record. The geographic configuration of the network with respect to the scale of spatial process and autocorrelation of a measured variable can have considerable impact on the quality or uncertainty of resulting models and maps. It is therefore desirable to place the sensors of an environmental monitoring network in locations such that the entire area of interest can be modeled within acceptable error thresholds. Given fixed and variable costs of deploying and maintaining sensors there are two quantifiable and competing objectives, subject to which a geographic configuration of sensors can be designed: cost of sensor infrastructure, and interpolation error on a map representing the geographical distribution of a phenomenon. Given spatial constraints, such as locations of strategic and sensitive areas, the problem then becomes how to optimally locate sensors in light of the objectives and constraints. As many phenomena are dynamic the configuration of sensors should be treated as dynamic, i.e. there is a need to re-design sensor configurations as conditions change.
Local decision makers (DMs) in charge of resource allocation pertinent to the management of a particular phenomenon in question. Analysts designing a geographic network of sensors.
A library of optimization algorithms and solvers residing on CI computers accessible through Web clients and CI interfaces. GIS data layers including locational constraints, sensor candidate locations, their attributes, geostatistical modeling tools for the computation of surface models and estimates and error variances, graphical tools for exploratory evaluation of trade-offs among non-dominated solutions (designs) to multiple objective spatial optimization problems, multiple attribute evaluation tools for rank-ordering non-dominated sensor network configurations.
An initial configuration (design) of sensor network is available. A geostatistical model of a phenomenon, based on the initial configuration of sensors, has been created and error variances are known.
1. Analyst presents the initial configuration of existing or proposed sensor network configuration in the form of queriable (GIS layer) map.
2. Decision makers and analyst, aided by GIS database and exploratory visual analytics tools, embark upon a collaborative brainstorming process to define optimization objectives, spatial criteria, and constraints.
3. Analysts defines a multiple objectives optimization model and submits it for processing to GI resources.
4. Non-dominated solutions are explored by analysts and DMs using visualization tools and acceptable trade-offs are selected for further processing.
5. Additional spatial criteria, other then the optimized objectives, and spatial constraints are taken into account in the process of arriving at a ranking of the selected non-dominated solutions.
6. Top-ranked sensor network configurations are displayed on maps and explored by DMs and analyst.
A sensor network configuration is selected for deployment. Alternatively, there is a lack of agreement among DMs and analyst concerning the preferred configuration, which may lead to subsequent iterations of the workflow.
The multiple objective optimization routine is abandoned in favor of interactively changing the locations of sensors in light of spatial criteria and constraints, running a geostatistical model, and exploring the distribution of prediction error variance and its co-location with the distribution of crucial spatial criteria (e.g. population density).