TY - THES
AU - Smith, Anthony David Miln
PY - 1979
TI - Adaptive management of renewable resources with uncertain dynamics
KW - Thesis/Dissertation
LA - eng
M3 - Text
AB - A range of adaptive policies is applied to the management of simulated fish stocks based on two simple models of stock dynamics, the Ricker model and the Schaefer model. Uncertainty about stock dynamics is represented as uncertainty about the parameters of these models. The policies tested include an active adaptive policy where management control options are chosen taking into account the uncertainty in the parameter estimates; a range of passive adaptive policies, where controls are chosen assuming parameter estimates are correct, but the estimates are updated from time to time; and a non-adaptive policy where the initial parameter estimates are assumed to be correct and are never updated.
Analysis of the regression problems for estimating the parameters of the two models shows that a major factor determining uncertainty about parameter estimates is the variability in observed values of the independent variables in the regression. Where there is more than one independent variable, contrasts between variables are also important.
Comparison of policy performances shows that the active adaptive policy always performs well relative to the optimal policy (where the true stock parameters are known). The passive adaptive policy with regular parameter estimate updating generally performs very well but occasionally very poorly. Poor performance occurs when the data points in the regression problem are clustered close to the apparent optimal levels of the independent variables. In most other cases poor initial parameter estimates cause sufficient perturbations in controls to correctly identify parameter values. The non-adaptive policy generally performs poorly, except in the case of the Ricker model where observations are available near the equilibrium stock size. This is due to the insensitivity of the optimal escapement to variations in the productivity of the stock. The good performance of the active adaptive policy is achieved at the expense of short term performance, which is sacrificed to improve parameter estimates. Infrequent updating of parameter estimates and low significance attached to new data are both shown to lead to marked deterioration in performance for the passive adaptive policy.
The major conclusion from the cases studied is that good estimation (equivalent to good understanding about stock dynamics) and hence good policy performance requires sufficient variability in, and contrasts between, the independent variables in the corresponding regression problems. It is suggested that this conclusion can be extended to more general problems of uncertainty about system dynamics in managing renewable resources provided that the problems can be simplified to an understanding of the key processes and uncertainties involved. The quantities corresponding to independent variables in an appropriate regression problem can then be identified and appropriate experimental management strategies devised to discriminate between alternative hypotheses.
N2 - A range of adaptive policies is applied to the management of simulated fish stocks based on two simple models of stock dynamics, the Ricker model and the Schaefer model. Uncertainty about stock dynamics is represented as uncertainty about the parameters of these models. The policies tested include an active adaptive policy where management control options are chosen taking into account the uncertainty in the parameter estimates; a range of passive adaptive policies, where controls are chosen assuming parameter estimates are correct, but the estimates are updated from time to time; and a non-adaptive policy where the initial parameter estimates are assumed to be correct and are never updated.
Analysis of the regression problems for estimating the parameters of the two models shows that a major factor determining uncertainty about parameter estimates is the variability in observed values of the independent variables in the regression. Where there is more than one independent variable, contrasts between variables are also important.
Comparison of policy performances shows that the active adaptive policy always performs well relative to the optimal policy (where the true stock parameters are known). The passive adaptive policy with regular parameter estimate updating generally performs very well but occasionally very poorly. Poor performance occurs when the data points in the regression problem are clustered close to the apparent optimal levels of the independent variables. In most other cases poor initial parameter estimates cause sufficient perturbations in controls to correctly identify parameter values. The non-adaptive policy generally performs poorly, except in the case of the Ricker model where observations are available near the equilibrium stock size. This is due to the insensitivity of the optimal escapement to variations in the productivity of the stock. The good performance of the active adaptive policy is achieved at the expense of short term performance, which is sacrificed to improve parameter estimates. Infrequent updating of parameter estimates and low significance attached to new data are both shown to lead to marked deterioration in performance for the passive adaptive policy.
The major conclusion from the cases studied is that good estimation (equivalent to good understanding about stock dynamics) and hence good policy performance requires sufficient variability in, and contrasts between, the independent variables in the corresponding regression problems. It is suggested that this conclusion can be extended to more general problems of uncertainty about system dynamics in managing renewable resources provided that the problems can be simplified to an understanding of the key processes and uncertainties involved. The quantities corresponding to independent variables in an appropriate regression problem can then be identified and appropriate experimental management strategies devised to discriminate between alternative hypotheses.
UR - https://open.library.ubc.ca/collections/831/items/1.0100273
ER - End of Reference