Real-time decision support in the face of emerging natural hazard events

Engineering structures are designed to resist a certain range of intensities of natural hazards. However, they are not designed to resist the entire range of possible intensities due to technical and economic constraints. Instead, in cases where they are most likely to fail as a result of emerging hazard events, several actions are undertaken to minimize possible consequences in real-time.

For example, a dike is built to protect inhabitants and properties against flood events up to a certain return period. In extreme rain storm events where dike failure is likely, persons and movable property can be evacuated or temporary physical protections can be built. Such measures, when deemed prudent or necessary, are recommended or ordered by public authorities but also voluntarily undertaken by individuals.

Other examples where private sector agents are in charge include engineering facilities such as wind turbines, agricultural facilities and offshore platforms. Operators of these facilities are often required to make decisions regarding the continued operations of their facilities in extreme storm events.

These decisions, which in the present thesis are called real-time decisions, are often made by a small number of people in extremely stressful situations, ad-hoc relying on personal experiences of decision makers. On the other hand, recent advancements of information technology potentially make it possible for decision makers to access various types of information in real-time.

Remarkable examples that facilitate real-time decision making in emerging natural hazard events are weather observation systems at the global scale, observation data processing systems, provision of best estimates of current atmospheric states and weather forecasts. However, the information provided is in most cases limited to the estimate of the current intensity of the emerging hazard event and the forecast thereof, and includes, in very limited cases, the prediction of risks.

Yet, none of the cases seem to systematically utilize such information for the decision optimization of the choice and commencement of risk reduction measures in real-time. Consequently, unnecessary costs and losses may occur. However, systematic use of such information on a decision support system would not only alleviate the stress of decision makers but also facilitate the identification of optimal decisions, thereby avoiding unnecessary costs and losses.

Motivated by these factors, the present thesis aims at developing a framework for the decision support system for real-time decision making in emerging natural hazard events. The thesis also demonstrates the implementation of the developed framework to illustrate its use and advantages. The developed framework is based on the work by Nishijima et al. (2009).

They formulate the general framework concept; however, it lacks an algorithm that solves the optimization problem with sufficient speed so that it can be utilized in practice. The difficulty lies in the sequential nature of the optimization problem, which requires backward induction. Respecting the analogy between the considered decision problem and the American option pricing, the present work proposes a very efficient algorithm on the basis of the Least Squares Monte Carlo method (LSM), which has been developed as an algorithm for pricing American options.

The main contribution of the present work is the development of the efficient algorithm based on LSM, which is called enhanced LSM (eLSM). As shown in the examples the efficiency of the proposed algorithm is up to the order of 100. Due to its efficiency it becomes possible to utilize decision support systems for variety of real-time decision problems.

Moreover, whereas the algorithm is developed primarily aiming at applications to the real-time decision making in emerging natural hazard events, the algorithm can be straightforwardly applied for other types of decision problems that share the same decision problem characteristics. These include decision problems in quality control and structural health monitoring.