Today, many governments use national riskassessment methods (NRA) as well as capability-based planning or capability analysis methods(CA). Many recently developed NRAs are all-hazard, scenario-based, andmulti-dimensional: scenarios of accidental, natural and man-maderisks are assessed by means of the same broad set of criteria,  makingvery different risks comparable in terms of multi-dimensional impact and likelihood. All-hazard CAs help to identify the most important capabilities, both generic (for all risks) and specific (for each particular risk).

For example, the Netherlands have an all-hazard NRA approach (see Figure 1) with a recurring process to develop, assess, plot and compare all-hazard risk scenarios, as well as a distributed, expert-based CA approach to identify and select capabilities that require further improvement.

Figure 1: Integrated All-hazard Risk-Capability (not under deep uncertainty)

Many RCA approaches werefirst and foremost developed for –at most– moderately uncertain and moderately complexincident-typerisks – not for very uncertain and/or very complex risks, nor for other types of risks(e.g. slumbering risks).And in spite of all good intentions, many approaches were not developed as truly integrated approaches in view of the final purpose of capability-based planning. Mostly, NRAswere developed first, being supplemented later with CAs, resulting in CAs designed for NRAs, not NRAs designed for CAs as ought to be the case.

So, in spite of the fact that most approaches attempt to be all-hazard and fully integrated, few really are. But truly all-hazard and integrated RCA methodologies that allow dealing with increasing degrees of complexity and deep uncertainty are –in our ever more complex and uncertain world– needed more than ever.

One of the underlying causes for the lack of truly appropriate approaches, was the state of science just a few years ago. However, since then, scientists developed the necessary methods and techniques. For example, our research team which focuses on developingmethods and tools for dealing with deeply uncertain dynamically complex issues, developed an integrated model-based RCA approach for dealingsimultaneously with all sorts of risks,from incident to slumbering, from moderately to deeply uncertain, and from simple to dynamically complex.

Our quantitative approach consists of six phases (see Figure 2):

• ESD Modeling:First, as many plausible underlying mechanismsand uncertainties that may significantly influence the evolution of a risk as possible need to be identified. These mechanisms and uncertainties need to be modeled, for example using exploratory System Dynamics (ESD) modeling.

• EMA Scenario Generation phase:These models and the widest possible uncertainty ranges are combined using Exploratory Modeling and Analysis(EMA) in order to generate tens of thousands of plausible time-evolutionary scenarios.

• Scenario Discovery and Set Selection phase:The smallest sets of scenarios that are representative (in terms of impacts, time-evolutionary behavior, and origin in the multi-dimensional uncertainty space) for the full set are identified using scenario discovery methods. Note that the plot of the NRA scores of the smallest sets of representative scenariosin a risk-envelope diagram ought to correspond to the plot of the full set.

• Single-hazard Capability Analysis phase:The deeply uncertain mediating effects of different capability policies on scenarios are then calculated for all representative scenarios of all risksusing risk-specific versions of a Capability Analysis model under deep uncertainty (CAuDU), resulting in thousands of simulations per representative risk scenario per risk, or tens of thousands of runs per risk, and hundreds of thousands of runs in total. Each risk requires different settings of the CAuDU model, which is why risks need to be treated separately until after this phase.

• Integration  of CAs over all representative scenarios of all risks:The deeply uncertain mediating effects of different capability policies on all different risks are then calculated, integrated and compared. Potential all-hazard capability sets could then be designed, tested, and compared, providing insight into the appropriateness of different sets of capabilities.

• Automated All-Hazard CA:Optimization could then be used to obtain the most robust capability setof capabilities given a particular investment level, starting from all promising sets, for all representative scenarios of all risks simultaneously.

Figure 2: Integrated All-hazard Risk-Capability under deep uncertainty

Although the description of this purely quantitative approach is very complicated, it is less complicated in practice. Risk-specific experts are required to provide insights in plausible underlying mechanisms and uncertainties, for example in a risk-related workshop. Based on these insights, modelers build preferably differentfast-to-build simulation models. Then, risk-capability-specific experts are required to provide insights in plausible mediating effects, and (representations of) policy makers are required provide insights in politically acceptable sets of capabilities in risk-capability workshops in which the consequences of hypotheses and capability sets can be simulated on the spot (few hours per risk).Automated all-hazard CA cannot be performed on the spot though, because these analyses are computationally very intensive.

Compared to qualitative approaches, this purely quantitative approach allows dealing with deep uncertainty as well as with dynamic complexity, and allows distilling the most robust sets of capabilities over a plethora of risks. As such, it is complementary to innovative qualitative approaches.

The condition sine qua nonfor real-world implementation of this integrated RCA under deep uncertainty is the existence of software to generate large ensembles of scenarios, of software to select representative ensembles of scenarios, generic risk-capabilities models that can be adapted to different risks, and software to optimize robustnessof capability sets simultaneously over many risks – all of which were developed and are available. Hence, all ingredients are available to turn the current state of science into the new state of the art.

Dr. Erik Pruyt is Assistant Professor of Policy Analysis and System Dynamics at Delft University of Technology, where his team develops methods and models for dealing with deeply uncertain dynamically complex issues. Since 2007, he is methodological advisor to the Dutch government regarding national safety and security. He co-developed the Dutch National Risk Assessment and Capability Analysis. He also developed methods and software tools for regional risk analysis, and for integrated risk-capability analysis for event-related risks.

Dr. ir. Jan Kwakkel received a Ph.D. from Delft University of Technology. His research focuses on the treatment of uncertainty in model based decision support. His Ph.D research focused on handling uncertainty in long term airport planning. He currently works as a postdoc at Delft University of Technology. He is the lead developer of the exploratory modeling and analysis workbench; a python library for exploratory modeling and analysis for supporting model based decision making under deep uncertainty.