Model Calibration and Economic Optimization in Conservation Planning with Bayesian Methods

Abstract

Abstract

A proper interpretation and treatment of uncertainty enables decision makers to make confident decisions by better use of knowledge and successful communication of what we know and what we do not know. Thus uncertainty is important in decision analysis of evidence based conservation planning. A Bayesian approach to statistics shows more advantages over the classical frequentist in treating uncertainties. It allows for consideration of knowledge based uncertainties, which are likely to exist in our understanding of ecological and environmental systems. It allows transparent integration of multiple sources of evidence and modelling of causal relationships. Bayesian methods are more and more adopted in scientific research. The fact that Bayesian approaches incorporate uncertainty in scientific evidence with decision analysis in a transparent way, makes them useful in simulation based approaches in evidence based management. An important step in simulation based approaches is the model calibration which means to assign values to parameters in the assessment model.
In this thesis, I put forward three model calibration approaches using Bayesian inference: traditional Bayesian modelling, Bayesian emulation and Bayesian Evidence Synthesis. I demonstrate how these methods handle calibration using a simple Population Viability Analysis assessing the risk of a population going extinct. The PVA is chosen to exemplify producing evidence for conservation planning. Then Bayesian Evidence Synthesis is applied to an economic optimization problem on strawberry production relying on pollination services. The link with conservation is that an efficient management of ecosystem services nowadays is part of conservation planning. A causal model was formulated that link characteristics in the surrounding landscape to visitation rates of pollinators (wild bees and honey bees) and from visitation rates to the yield and quality of strawberries. The model has been calibrated in OpenBUGS based on field measurement data. Simulated annealing was applied to find the optimal combination of management actions in different strawberry fields (sampled from a real landscape) that maximizes the strawberry farmers’ profits.
Our case shows that Bayesian Evidence Synthesis is a suitable framework for conservation planning. In the end, I discuss the advantages and problems with traditional Bayesian modelling, Bayesian emulation and Bayesian Evidence Synthesis and compare these methods.

Abstract

Popular science summary:

Having problems with statistics? Try Bayesian!

While many of us are afraid of the abstruse concepts and annoyed by a lot of limitations in classical statistics, Bayesian methods provides user-friendly way to do statistics. They are intuitive, explicit and have a wide range of applications.

In conservation planning, different sources of evidence need to be integrated to support decision making. However, one might find it hard and complicated to combine the evidences in a classical statistical way. In addition, a proper interpretation of and treatment of uncertainty from different sources usually help the decision maker produce better decisions. Thus, uncertainty and risk is another thing one should take into account when doing decision analysis. The fact that Bayesian approaches incorporate uncertainty in scientific evidence with decision analysis in a transparent way, makes them useful in model based approaches in evidence based management decision. Bayesian methods are more and more adopted in scientific research.

I illustrated different Bayesian methods with examples from PVA (population viability analysis). Then I applied Bayesian modelling on a case-study where I searched for the optimal management of bees for strawberry production in different landscapes given the evidence found in a field experiment of strawberry yields and pollinating bees. A Bayesian model was built and applied to simulate the profits of different strawberry farm under different management options. The optimal management was found through simulated annealing using a genetic algorithm in which the farmers’ attitudes towards risk and knowledge-based uncertainty is taken into account.

What is the difference between Bayesian methods and classical statistical methods?
Bayesian methods differ from classical (frequentist) statistics in their fundamental philosophical bases. In classical statistics, uncertainty is presented by confidence interval interpreted as the relative frequency of an event over time. Bayesian methods describe uncertainty in the form of probability distributions, interpreted as subjective degrees of beliefs.

Bayesian Evidence Synthesis
In this thesis, I compare three Bayesian methods to calibrate models. I conclude that Bayesian Evidence Synthesis (BES, see figure) is a suitable framework for conservation planning. A BES starts from a cost-efficiency problem. The cost-efficiency analysis is based on the predictions about the future states of a system, generated by a simulator (a model representing the relevant system processes). Assigning values to the parameters in the simulator (i.e. to calibrate) is an important step. Parameters are informed by different sources of evidence, including field observations, expert knowledge, and an understanding of the underlying mechanisms.

Advisor: Ullrika Sahlin
Master´s Degree Project 30 credits in Conservation Biology 2014

• author: Jiang, Yifei
• supervisor: Ullrika Sahlin
• organization: Degree Projects in Biology
• course: BIOM24 20141
• year: 2014
• type: H2 - Master's Degree (Two Years)
• subject: Biology and Life Sciences
• language: English
• id: 4467804
• date added to LUP: 2014-06-17 11:28:18
• date last changed: 2014-06-17 15:52:47