W1: Workshop on Environmental Monitoring & Adaptive Management
organised by
John Norton and Kenneth Reckhow
Natural-resource management is usually viewed as making a one-off decisions or set of
decisions, aided by models to predict effects of alternative decisions
Weaknesses of this process:
- it either ignores uncertainties in model predictions or takes them
into account by weighting predictions with probabilities (or some
similar method), which can't take any notice of unforeseen consequences
of decisions, unquantifiable effects of future changes in the system or
its inputs, unrecognised omissions from the model
- it assumes that the management criteria used in the decision-making will stay suitable
- it often doesn't explicitly consider how the success of the
management will be checked, i.e. doesn't design the monitoring together
with the management action
- at best it only partly investigates controllability and
observability, i.e. what can be made to happen by the management
actions and what can be detected in the response
- it takes no advantage of the ability of feedback (even without
adaptation) to reduce the sensitivity of outputs to disturbances,
changes in the system and modelling errors
- selection of spatio-temporal intervals (for model and monitoring)
is largely heuristic and may not match the system's dynamics and the
scales relevant for the objectives.
Questions arising from this situation:
- Is it unavoidable; what factors limit or prevent application of the
principles of feedback control to NRM? Do social-political-
economic-biophysical realities force us to treat the great majority of
NRM problems as "one-off"? Does short-term accountability prevent
managers from implementing continuing policies with good long-term
results which are not obvious in the short term?
- How can the case be made to sponsors, stakeholders and the public
that many NRM problems require active management over decades, and
hence funding arrangements and continuity to match?
- Is there anything for NRM in the ideas of robust control? For
instance, does maximising the worst-case benefit make sense? Does
optimisation subject to worst-case bounds on some aspects of
performance make sense? What sort of descriptions of ranges of system
behaviour would be realistic, obtainable and useable?
- Where is there a role for constrained optimisation in NRM (as in
robust schemes such as Model Predictive Control (MPC: see refs.
below))? Do multiple and conflicting criteria prevent it? (After all,
such criteria arise all the time in engineering).
- Is there a place for receding-horizon control in NRM?
- What is meant by adaptive management: that the management actions,
but not the management rules, are modified as time goes on (not
adaptive control according to control engineering nomenclature), or
that the rules deriving management actions from observed behaviour of
the system are changed as the situation develops (which is adaptive
control)? (It's useful to distinguish adaptive and non-adaptive
control/management, as they pose different problems of designing a
policy - not just an action).
- How do we design post-implementation monitoring programs, given
learning and assessment objectives? In particular, how do we account
for lags in system response?
References
- A. Bemporad and M. Morari (1999) Robust model predictive control: a survey, in Robustness in Identification and Control, Eds. A. Garulli,
A. Tesi and A. Vicino, LNCIS 245, Springer-Verlag, London, 1999,
207-226.
- C. Garcia, D. M. Prett and M. Morari (1989) Model Predictive Control-theory and practice-a survey, Automatica, 25, 335-348.
- J. M. Maciejowski (2002) Predictive Control with Constraints, Prentice Hall, Harlow, England, ISBN 0 201 39823 0.
- M. Morari and E. Zafiriou (1989) Robust Process Control, Prentice-Hall Inc., Englewood Cliffs, NJ, USA. ISBN 0 13 782153 0.
- R. Soeterboek (1992) Predictive Control: A Unified Approach, Prentice Hall International (UK), Hemel Hempstead, UK. ISBN 0 13 678350 3.
