Spatial planning under uncertainty Brendan Wintle and Mark Burgman
Natural variation (aleatory uncertainty) Lack of knowledge (epistemic uncertainty) Probability arithmetic, ‘classical’ decision theory, Monte Carlo The engineer’s taxonomy of uncertainty
Linguistic uncertainty Ambiguity – words have two or more meanings, and it is not clear which is meant (‘cover’). Vagueness – borderline cases (e.g., ‘river’) Underspecificity – unwanted generality. Context dependence – a failure to specify context. (Regan et al 2002)
Underspecificity Gigerenzer, Hertwig, van den Broek, Fasolo, & Katsikopoulos, Risk Analysis (in press) There’s a 70% chance of rain Possible interpretations rain during 70% of the day rain over 70% of the area 70% chance of rain at a particular point (the weather station)
Habitat maps Reserve planning exercise Landscape data Habitat maps in conservation planning Decisions
Old Growth Solar data Topography Temperature Presence/ Absence Data
models habitat quality ~ environmental attributes Habitat Model pr(occupancy) ~ α + β 1 X 1 + β 2 X 2 + … β k X k
Habitat maps
Introduced from Asia Contradictory laws Hunters: utility Conservation: ecological damage Samba Deer
Questions 1.How many are there? 2.Where are they likely to disperse? 3.Can we manipulate the landscape to slow dispersal?
Subjective uncertainties
Bounds
95% CIs What are they? The Sooty Owl in the Eden Region Mean prediction Lower 95% Upper 95% What is the probability the species is present? How reliable is the probability? Is the map reliable ‘enough’?
Wintle and Burgman - prioritizing under uncertainty Prioritizing under uncertainty: data, models, decision theory How important is the uncertainty in my particular application? How can i find out? What can i do about it? Decision Theory Because the uncertainty is only important to the extent that it impacts on the quality or robustness of decisions
Wintle and Burgman - prioritizing under uncertainty Prioritizing under uncertainty: data, models, decision theory Case study: Spatial prioritization that is robust to uncertainty about habitat values. Goal: Prioritize areas of high quality habitat for protection against development in the Hunter Valley, NSW, Australia Uncertainty: Imperfect spatial representation of habitat quality for focal species Case study: Spatial prioritization that is robust to uncertainty about habitat values. Goal: Prioritize areas of high quality habitat for protection against development in the Hunter Valley, NSW, Australia Uncertainty: Imperfect spatial representation of habitat quality for focal species
Wintle and Burgman - prioritizing under uncertainty Prioritizing under uncertainty: data, models, decision theory Decision: Choose the reserve design that satisfies a minimum representativeness requirement, and that is most robust to uncertainty in the estimates of habitat quality for focal species. Decision theory: Info-gap decision theory (Ben-Haim 2002) Decision: Choose the reserve design that satisfies a minimum representativeness requirement, and that is most robust to uncertainty in the estimates of habitat quality for focal species. Decision theory: Info-gap decision theory (Ben-Haim 2002)
Wintle and Burgman - prioritizing under uncertainty YBG The Data SQGL SOWL GRGL POWL ETC.. predicted distribution of yellow-bellied glider habitat in the hunter region (Wintle,Elith,Potts (2005) Austral Ecology)
Wintle and Burgman - prioritizing under uncertainty The uncertainty habitat quality ~ environmental attributes Uncertainty: Imperfect spatial representation of habitat quality for focal species
Wintle and Burgman - prioritizing under uncertainty The uncertainty pr(occupancy) ~ α + β 1 X 1 + β 2 X 2 + … β k X k Uncertainty: Imperfect spatial representation of habitat quality for focal species
Wintle and Burgman - prioritizing under uncertainty The uncertainty pr(occupancy) ~ α + β 1 X 1 + β 2 X 2 + … β k X k detectability-classification error data age positional accuracy modelling method: -glm/gam -gdm/gbm -boosted regression -mars/cart -garp/neural nets non-independence NON EQUILIBRIUM STATES poorly mapped variables: classification error, measurement error distal variables model structure uncertainty parameter uncertainty sampling bias
Wintle and Burgman - prioritizing under uncertainty The uncertainty pr(occupancy) ~ α + β 1 X 1 + β 2 X 2 + … β k X k detectability-classification error data age positional accuracy modelling method: -glm/gam -gdm/gbm -boosted regression -mars/cart -garp/neural nets non-independence NON EQUILIBRIUM STATES poorly mapped variables: classification error, measurement error distal variables model structure uncertainty parameter uncertainty sampling bias
Wintle and Burgman - prioritizing under uncertainty The uncertainty pr(occupancy) ~ α + β 1 X 1 + β 2 X 2 + … β k X k detectability-classification error data age positional accuracy modelling method: -glm/gam -gdm/gbm -boosted regression -mars/cart -garp/neural nets non-independence NON EQUILIBRIUM STATES poorly mapped variables: classification error, measurement error distal variables model structure uncertainty parameter uncertainty sampling bias
Wintle and Burgman - prioritizing under uncertainty The uncertainty pr(occupancy) ~ α + β 1 X 1 + β 2 X 2 + … β k X k detectability-classification error data age positional accuracy modelling method: -glm/gam -gdm/gbm -boosted regression -mars/cart -garp/neural nets non-independence NON EQUILIBRIUM STATES poorly mapped variables: classification error, measurement error distal variables model structure uncertainty parameter uncertainty sampling bias
Wintle and Burgman - prioritizing under uncertainty The uncertainty Uncertainty: Imperfect spatial representation of habitat quality for focal species mean uncertainty
Wintle and Burgman - prioritizing under uncertainty Case study – Hunter Valley 1.objective – identify the conservation strategy that maximizes our immunity to uncertainty (in habitat predictions) while achieving a satisfactory proportion of preserved habitat for each species (minimum area). robust satisfycing 2.maximize robustness to uncertainty while achieving a satisfactory outcome – infogap decision theory
Wintle and Burgman - prioritizing under uncertainty Case study – Hunter Valley 1.objective – identify the conservation strategy that maximizes our immunity to uncertainty (in habitat predictions) while achieving a satisfactory proportion of preserved habitat for each species. robust satisfycing 2.uncertainty characterized by bounds on p 3.solution - info-gap decision theory (Ben-Haim 2001):
Wintle and Burgman - prioritizing under uncertainty Case study – Hunter Valley design 1 design2 horizon of uncertainty (α) habitat included in reserve (ha) two questions: is this amount of uncertainty plausible? what is this minimumally satisfactory performance?
Wintle and Burgman - prioritizing under uncertainty Case study – Hunter Valley Solution (find a geek): implemented in Zonation (Moilanen et al. 2005) - Implementation hardwired in Zonation for all to use - Load in uncertainty files (prediction lower bounds)
Wintle and Burgman - prioritizing under uncertainty Case study – Hunter Valley pr(occupancy) ~ α + β 1 X 1 + β 2 X 2 + … β k X k α = 0 α = 2α = 3 Increasing robustness to uncertainty in habitat quality estimates
Wintle and Burgman - prioritizing under uncertainty Adaptive management Your decision will be wrong, so have a plan to learn and adapt (adaptable spatial priorities?) Linkov et al Integ. Env. Ass. Manage.
Wintle and Burgman - prioritizing under uncertainty Conclusions 1.it is possible (though not trivial) to explicitly identify management strategies that are most robust to uncertainty 2.optimal policies are often not robust to uncertainty 3.including all uncertainties is hard, but including as many as possible is worth it 4.your decision will definitely be wrong, so have a plan for learning and adapting
Wintle and Burgman - prioritizing under uncertainty Conclusions 5.life without uncertainty is boring the future 6.make this easier 7.extension - case studies – variable costs 8.rules of thumb
Wintle and Burgman - prioritizing under uncertainty References
Wintle and Burgman - prioritizing under uncertainty References
Wintle and Burgman - prioritizing under uncertainty This one’s the easiest to follow!
Wintle and Burgman - prioritizing under uncertainty Prioritizing under uncertainty: data, models, decision theory Mark Burgman, Brendan Wintle