Modelling and monitoring environmental outcomes in adaptive management
Norton, J. P., & Reckhow, K. (2008). Modelling and monitoring environmental outcomes in adaptive management. In A. J. Jakeman, A. A. Voinov, A. E. Rizzoli, & S. H. Chen (Eds.), Environmental Modelling, Software and Decision Support: State of the Art and New Perspective (pp. 181-204). New York, NY: Elsevier. DOI: 10.1016/S1574-101X(08)00611-X
The main aim of this chapter is to air questions about the future of adaptive management (AM) of natural resources, and more specifically about what approaches may be feasible which have not yet been explored well. The method adopted is to compare the histories, ideas, strengths and limitations of AM, control engineering and Bayesian analysis, which have superficial similarities, significant differences and perhaps lessons for each other. Questions arising in these comparisons include:
– What factors limit or prevent application of the principles of feedback control and AM in natural-resource management (NRM)? Do the apparent similarities in problems allow approaches developed in control engineering to be applied in NRM, or are there fundamental differences? Do social-political-economic-biophysical realities prevent a systematic approach to NRM, employing techniques portable from problem to problem? Do short-term accountability, short-term funding and difficulties in measuring outcomes prevent managers from implementing policies embodying the principles of AM, with its focus on monitoring, adaptation to evolving situations and attention to long-term results which may not be clear in the short term?
– In what sense is AM adaptive? If the rules which derive management actions from observed behaviour of the system are changed in the light of experience, AM is adaptive according to the usage of the word in control engineering, but not if management actions, but not management rules, are modified as time goes on. What light does the chequered history of adaptive control throw on the prospects for genuinely adaptive AM?
– Do the ideas of robust control offer anything for NRM? For instance, does maximising the worst-case benefit, or optimising subject to bounds on some aspects of performance, make sense?
– Are there roles in NRM for receding-horizon control based on predictive models, determination of future actions by constrained numerical optimisation of model response and model revision according to observed behaviour, as in robust schemes such as Model Predictive Control? Do multiple and conflicting criteria prevent their use?
– How do the probabilistic (Bayesian) and bound-based alternatives for specifying uncertainty lend themselves to realistic use in NRM?
– What does the Bayes updating paradigm offer for NRM?
In this chapter, Sections Sections 11.1 and 11.6 review the most relevant aspects of AM, control engineering and Bayesian analysis. These sections are closely based on the position paper for Workshop 1 of the Summit on Environmental Modelling and Software in Burlington in 2006. The appendix summarises the ensuing workshop proceedings, which consisted of three short presentations to give practical substance to the topics, followed by a free-flowing discussion only loosely mediated by the convenors. Because of the informal and at times complicated nature of the discussion, no attempt has been made to attribute opinions to individual participants.