Presenter/Author Information

Patrick M. Reed
Ruchit Shah
Joshua B. Kollat

Keywords

observation networks, many-objective analysis, bayesian networks

Start Date

1-7-2010 12:00 AM

Description

This paper explores three related propositions for designing environmentalobservation systems: (1) Nonstationarity in environmental data series has the consequentimpact of making observation network design itself a nonstationary, stochastic planningproblem where the value of alternative observation strategies should be evaluated based onplanners’ evolving conception of Pareto efficiency given new knowledge, technologies,and policies over long time-scales. (2) Real-world budgetary constraints within observationnetwork design problems yield resource allocation conflicts across space, time, andcompeting foci that are equivalent in form to the multiobjective d-dimensional knapsackproblem (MO-dKP). Consequently, the Pareto efficiency of observation networks can onlybe determined approximately for non-trivial problem instances. (3) Multiobjectivehierarchical Bayesian optimization provides a very promising tool for identifyingobservation alternatives that are approximately Pareto efficient while simultaneouslyproviding insights into the emergent dependencies of our decisions (both science andmanagement oriented) on critical observations.

Share

COinS
 
Jul 1st, 12:00 AM

Assessing the Value of Environmental Observations in a Changing World: Nonstationarity, Complexity, and Hierarchical Dependencies

This paper explores three related propositions for designing environmentalobservation systems: (1) Nonstationarity in environmental data series has the consequentimpact of making observation network design itself a nonstationary, stochastic planningproblem where the value of alternative observation strategies should be evaluated based onplanners’ evolving conception of Pareto efficiency given new knowledge, technologies,and policies over long time-scales. (2) Real-world budgetary constraints within observationnetwork design problems yield resource allocation conflicts across space, time, andcompeting foci that are equivalent in form to the multiobjective d-dimensional knapsackproblem (MO-dKP). Consequently, the Pareto efficiency of observation networks can onlybe determined approximately for non-trivial problem instances. (3) Multiobjectivehierarchical Bayesian optimization provides a very promising tool for identifyingobservation alternatives that are approximately Pareto efficient while simultaneouslyproviding insights into the emergent dependencies of our decisions (both science andmanagement oriented) on critical observations.