PRAFOR: Probabilistic drought Risk Analysis for FORested landscapes
Lead Research Organisation:
James Hutton Institute
Department Name: BIOSS
Abstract
This research aims to extend theory for probabilistic risk analysis of continuous systems, test its use against forest data, use process models to predict future risks, and develop decision-support tools.
Risk is commonly defined as the expectation value for loss. Most risk theory is developed for discrete hazards such as accidents, disasters and other forms of sudden system failure. Less theory has been developed for systems where the hazard variable is always present and continuously varying, with matching continuous system response. We can think of dynamic systems whose performance varies with ever-changing resource availability or other dynamic constraints, e.g. crop growth depending on water supply, or urban health as a function of air pollutant concentration. Risks from such continuous hazards (levels of water, pollutants) are not associated with sudden discrete events, but with extended periods of time during which the hazard variable exceeds a threshold. To manage such risks, we need to know whether we should aim to reduce the probability of hazard threshold exceedance or the vulnerability of the system. In earlier work (Van Oijen et al. 2013, http://iopscience.iop.org/1748-9326/8/1/015032), we showed that there is only one possible definition of vulnerability that allows formal decomposition of risk as the product of hazard probability and system vulnerability (R = p[H] V). We have used this approach to analyse risks from summer droughts to the productivity of vegetation across Europe under current and future climatic conditions (Van Oijen et al. 2014, http://www.biogeosciences.net/11/6357/2014/bg-11-6357- 2014.html). This showed that climate change will likely lead to greatest drought risks in southern Europe, primarily because of increased hazard probability rather than significant changes in vulnerability. We plan to improve on this preliminary theoretical work in different ways:
- Add one more major risk component to the analysis: exposure to the hazard, so that risk becomes the product of three terms. That will allow distinguishing between hazards that only affect few individuals or points in space to those that affect larger populations and areas.
- Derive equations for quantifying the uncertainties in our estimates for risk and its components. Only with quantified uncertainties can the estimates play a legitimate role in decision-support.
- Relax assumptions underlying previous work and develop the theory for any type of joint probability distribution for hazard, exposure and vulnerability. This will likely require the use of extreme value theory and numerical estimation using Bayesian hierarchical modelling.
- Test our equations and numerical algorithms on both observed and simulated data in this research. Observational data will be from forests in the U.K., Spain and Finland. Simulated data will be generated by process-based modelling of forest response to climate change.
- Analyse the underlying causes of vulnerability, as represented by the parameters and processes of the process-based forest model.
- Show the wider implications of the risk decomposition and the uncertainty quantification, by embedding the equations in Bayesian decision theory to allow identification of optimal drought management measures.
- Develop an interactive web application as a tool for preliminary exploration of risk and its components to support decision-making.
The work will be carried out by CEH-Edinburgh in close collaboration with Biomathematics and Statistics Scotland (BioSS, part of the James Hutton Institute, Aberdeen) and Forest Research UK (Alice Holt, Aberdeen, Edinburgh). Data and expertise from Spain and Finland will be provided by two Project Partners: the University of Alcalá (Madrid, Spain) and the Natural Resources Institute (Luke-Helsinki, Finland).
Risk is commonly defined as the expectation value for loss. Most risk theory is developed for discrete hazards such as accidents, disasters and other forms of sudden system failure. Less theory has been developed for systems where the hazard variable is always present and continuously varying, with matching continuous system response. We can think of dynamic systems whose performance varies with ever-changing resource availability or other dynamic constraints, e.g. crop growth depending on water supply, or urban health as a function of air pollutant concentration. Risks from such continuous hazards (levels of water, pollutants) are not associated with sudden discrete events, but with extended periods of time during which the hazard variable exceeds a threshold. To manage such risks, we need to know whether we should aim to reduce the probability of hazard threshold exceedance or the vulnerability of the system. In earlier work (Van Oijen et al. 2013, http://iopscience.iop.org/1748-9326/8/1/015032), we showed that there is only one possible definition of vulnerability that allows formal decomposition of risk as the product of hazard probability and system vulnerability (R = p[H] V). We have used this approach to analyse risks from summer droughts to the productivity of vegetation across Europe under current and future climatic conditions (Van Oijen et al. 2014, http://www.biogeosciences.net/11/6357/2014/bg-11-6357- 2014.html). This showed that climate change will likely lead to greatest drought risks in southern Europe, primarily because of increased hazard probability rather than significant changes in vulnerability. We plan to improve on this preliminary theoretical work in different ways:
- Add one more major risk component to the analysis: exposure to the hazard, so that risk becomes the product of three terms. That will allow distinguishing between hazards that only affect few individuals or points in space to those that affect larger populations and areas.
- Derive equations for quantifying the uncertainties in our estimates for risk and its components. Only with quantified uncertainties can the estimates play a legitimate role in decision-support.
- Relax assumptions underlying previous work and develop the theory for any type of joint probability distribution for hazard, exposure and vulnerability. This will likely require the use of extreme value theory and numerical estimation using Bayesian hierarchical modelling.
- Test our equations and numerical algorithms on both observed and simulated data in this research. Observational data will be from forests in the U.K., Spain and Finland. Simulated data will be generated by process-based modelling of forest response to climate change.
- Analyse the underlying causes of vulnerability, as represented by the parameters and processes of the process-based forest model.
- Show the wider implications of the risk decomposition and the uncertainty quantification, by embedding the equations in Bayesian decision theory to allow identification of optimal drought management measures.
- Develop an interactive web application as a tool for preliminary exploration of risk and its components to support decision-making.
The work will be carried out by CEH-Edinburgh in close collaboration with Biomathematics and Statistics Scotland (BioSS, part of the James Hutton Institute, Aberdeen) and Forest Research UK (Alice Holt, Aberdeen, Edinburgh). Data and expertise from Spain and Finland will be provided by two Project Partners: the University of Alcalá (Madrid, Spain) and the Natural Resources Institute (Luke-Helsinki, Finland).
Planned Impact
Our risk analyses will inform future national climate change risk assessments (CCRA), and future IPCC reports. The project outcomes will benefit a variety of stakeholders whom we are in contact with:
- The Forestry Commission (England), Natural Resources Wales, Scottish Forestry and NI Forest Service in the evolved administrations through evidence on future climate suitability, forest adaptation measures and conservation strategies for two major conifer species.
- DEFRA, Natural England and Scottish Natural Heritage and other agencies through evidence on likely changes in risk and impacts on forest ecosystem structure and function.
- Public and private sector forestry practitioners, producers and processors (via UK Forestry Societies, Institute of Chartered Foresters and Confederation of Forest Industries through the DSS developed and through the exploration of the potential of management options to reduce risk. Our work will also provide valuable evidence to specialist woodland groups e.g. the Native Woodland Discussion Group, Woodland Trust, RSPB.
- European groups seeking evidence for policy-making and national and EU long-term forest management planning such as European Forest Institute, European State Forest Association and national Forest Owners Associations through the Confederation of European Forest Owners.
- This research will also be of substantial interest to the broad swath of the public concerned about the sustainability of our way of life, woodland resource, woodland resilience and their links to climate security.
- Development of this new risk assessment framework will have many possible applications across environmental science, in addition to the immediate examples developed in this proposal. In forestry, it will help develop further FR's work on wind risk (UK forestry's biggest present risk to production) and in risks to tree health from various pest and pathogens, where there is considerable concern. The improved assessment of current and future drought risk will be particularly useful. The combined risk assessment and modelling framework will lend itself to application to other novel species to help define 'climate-smart' forestry, and strengthen sector resilience through species diversification. At the European scale, the methodologies developed will allow extensive assessment of Scots pine vulnerability.
- The Forestry Commission (England), Natural Resources Wales, Scottish Forestry and NI Forest Service in the evolved administrations through evidence on future climate suitability, forest adaptation measures and conservation strategies for two major conifer species.
- DEFRA, Natural England and Scottish Natural Heritage and other agencies through evidence on likely changes in risk and impacts on forest ecosystem structure and function.
- Public and private sector forestry practitioners, producers and processors (via UK Forestry Societies, Institute of Chartered Foresters and Confederation of Forest Industries through the DSS developed and through the exploration of the potential of management options to reduce risk. Our work will also provide valuable evidence to specialist woodland groups e.g. the Native Woodland Discussion Group, Woodland Trust, RSPB.
- European groups seeking evidence for policy-making and national and EU long-term forest management planning such as European Forest Institute, European State Forest Association and national Forest Owners Associations through the Confederation of European Forest Owners.
- This research will also be of substantial interest to the broad swath of the public concerned about the sustainability of our way of life, woodland resource, woodland resilience and their links to climate security.
- Development of this new risk assessment framework will have many possible applications across environmental science, in addition to the immediate examples developed in this proposal. In forestry, it will help develop further FR's work on wind risk (UK forestry's biggest present risk to production) and in risks to tree health from various pest and pathogens, where there is considerable concern. The improved assessment of current and future drought risk will be particularly useful. The combined risk assessment and modelling framework will lend itself to application to other novel species to help define 'climate-smart' forestry, and strengthen sector resilience through species diversification. At the European scale, the methodologies developed will allow extensive assessment of Scots pine vulnerability.
Organisations
| Description | We have developed an approach for assessing drought risk using a Bayesian Decision Theory approach in forested landscapes. The approach accounts for uncertainty in probabilistic terms and can make use of expert knowledge when data are not available. We have concentrated on data sets from the UK and Spain, and applied the approach to produce exemplar "case studies" to illustrate use of a software package. A book has been published describing the approach with examples, and this will also acts as a "manual" for the software, the latter being downloadable from the (e-book/online) versions of the book. |
| Exploitation Route | We provided case study analyses showing the use of the approach, but implementation in other regions will take time and money. For example, gathering the necessary data sets is not straightforward. Funding could be used to extend the application of the approach, in particular to consider other types of risk, such as flooding or wind damage (as seen in large parts of Scotland during the winter 2021/22, for example). The approach we have developed could be used in such contexts, given appropriate data. |
| Sectors | Agriculture, Food and Drink,Environment |
| Title | Bayesian Decision Theory for Drought Risk Analysis in Forested Catchments |
| Description | We have now derived a decision theory modelling approach to use in conjunction with the risk analysis tool produced last year. There are two versions, one which uses approximations to the fully Bayesian model, and an alternative which makes fewer assumptions and uses Markov Chain Monte Carlo methods. This has involved a novel comparison of decision theory and risk analysis with clear identification of which approach is appropriate in which context. The models have been completed in R and using the "Nimble" package for simulation, to allow full integration with forest models and weather generators or emulators. The approach has been described in a draft book which will be published in due course by Springer Nature. A Github repository containing code will also be made available at the conclusion of the project, now delayed (with approval) to May 2022. |
| Type Of Material | Data analysis technique |
| Year Produced | 2022 |
| Provided To Others? | No |
| Impact | We are at present creating usable case studies using actual drought and forestry data, in co-operation with Forest Research in the UK and with the University of Alcala in Spain. |
| Title | Fully Bayesian Probabilistic Risk Analysis for Decision Theory |
| Description | This project will ultimately provide tools for decision makers, primarily with regard to drought risk in forested catchments, based upon a Bayesian Decision Theory analysis building on a form of Probabilistic Risk Analysis. My work this year has been to develop a Bayesian implementation, using Markov Chain Monte Carlo methods, of risk analysis, where regions of a key metric (e.g. rainfall or river flow) determine discrete "drought" states. This has been implemented in R using the "Nimble" package, and as such should allow full integration with forest models and weather generators. The R implementation will be placed on GitHub after further testing, which is being conducted now in tandem with development of connections to the Bayesian Decision Theory modelling. At this point, the output will be freely available to others. |
| Type Of Material | Data analysis technique |
| Year Produced | 2021 |
| Provided To Others? | No |
| Impact | Within this project, the approach has enabled a full quantification of the uncertainty in the distributions of formal quantities of Risk and Vulnerability, for example. This will expand further to enable similar robust uncertainty quantification within the final product, the decision support tool. |
| Title | Probabilistic Risk Analysis and Decision Theory for Forested Landscapes |
| Description | We have now published the code behind a risk analysis and decision theory modelling approach to use in conjunction with the book "Probabilistic Risk Analysis and Bayesian Decision Theory" published by SpringerBriefs in Statistics in 2022. The book consists of four parts: chapters 1-11 focus on probabilistic risk analysis (PRA); chapters 12-15 introduce Bayesian decision theory (BDT); chapters 16-18 examine the link between PRA and BDT; and chapter 19 contains a concluding discussion. Each part of the book at https://link.springer.com/book/10.1007/978-3-031-16333-3 contains links to the code for download and application by the reader. A Github repository containing code will also be made available at the conclusion of the project, now delayed (with approval) to May 2022. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2022 |
| Provided To Others? | Yes |
| Impact | We have completed the model development work, and this will be used by UKCEH and partners in assessing drought risk in forested catchments in future (my part of the work was assisting with the model development, the entirety of this Award which supplemented the main submission by UKCEH). |
| URL | https://link.springer.com/book/10.1007/978-3-031-16333-3 |
| Description | Departmental Seminar at the University of Glasgow |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Professional Practitioners |
| Results and Impact | This was a Departmental Seminar at the University of Glasgow in the School of Mathematics and Statistics, where I presented the preliminary work on the Probabilistic Risk Assessment related to PRAFOR. This was both to publicise the work itself and to obtain the input of peers on the general approach. It led to one potential collaboration in the area of emulators, related to the inclusion of complex forestry models in the forthcoming Bayesian Decision Theory component. |
| Year(s) Of Engagement Activity | 2020 |
| Description | Presentation at Isaac Newton Institute workshop connecting researchers with policymakers |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Study participants or study members |
| Results and Impact | This was a workshop intended to link research progress from the UKRI Strategic Priority Fund (SPF) "Landscape Decisions: Toward a new framework for using land assets" programme with stakeholders working on land-related research and policy questions. A key extra theme was connecting the mathematical/statistical research with state-of-the-art social modelling approaches. |
| Year(s) Of Engagement Activity | 2020 |
| URL | https://www.newton.ac.uk/event/ebdw03 |
| Description | Presentation at international conference - RSS 2021 |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | This was a specially organised session of the funding stream for the international Conference of the Royal Statistical Society, held in Manchester, UK, in September 2021. I presented the latest update of the theoretical methodology for accounting for uncertainty in Risk Analysis and linking with Bayesian Decision Theory. There were around 60 participants, in person and online. There was lively group discussion during all the (related) talks and ideas were suggested for taking the work beyond the framework of the current project. |
| Year(s) Of Engagement Activity | 2021 |
| URL | https://rss.org.uk/training-events/events/events-2021/conferences/rss-2021-international-conference |