The deployment of resources during a natural disaster, whether it be water, food, machines, people or something else, requires complex planning and scheduling and the need to adapt to constantly changing scenarios, often involving large number of resources, unique requirements based on location and the varying staffing levels associated with each resource. Government agencies use different systems to estimate their program needs, including preparedness resource planning, yet no one system has been able to adapt to the increasing complexity of natural disaster management.
These challenges resulted in IBM developing a large-scale strategic budgeting framework for managing natural disaster events, with a focus on better preparedness for future uncertain disaster scenarios. The underlying stochastic optimization models and algorithms were initially prototyped on a large U.S. Government program, where the key problem was how to efficiently deploy a large number of critical resources to a range of disaster event scenarios. That system generated a single solution for each disaster scenario. The current enhancements to the budgeting system include the development of simulation models to assess and evaluate the impact of alternative strategies based upon criteria selected by the client. Stochastic optimization allows the client to trade off among multiple priorities to understand the impacts to performance measures.
The same models can be explored to manage floods or famines in India, or natural disasters anywhere in the world. A fully developed, customized and implemented model can significantly help the country's approach for disaster risk reduction and disaster management.
"We are creating a set of intellectual properties and software assets that can be employed to gauge and improve levels of preparedness to tackle unforeseen natural disasters," says Dr. Gyana Parija, senior researcher and optimization expert at IBM India Research Laboratory, New Delhi. "Most real-world problems involve uncertainty, and this has been the inspiration for us to tackle challenges in natural disaster management."
In the case of flooding, for example, the stochastic programming model developed by IBM researchers would use various flood scenarios, resource supply capabilities at different dispatch locations, and fixed and variable costs associated with deployment of various flood-management resources to manage various risk measures. By assigning probabilities to the factors driving outcomes, the model outlines how limited resources can meet tomorrow's unknown demands or liabilities. In this way, the risks and rewards of various tradeoffs can be explored.
Stochastic programming offers greater modeling power and flexibility, but it comes at a cost-premium processing time. However, recently, stochastic programming has benefited from the development of more efficient algorithms and faster computer processors. This means that rather than predicting a limited future using forecasting, decisions supporting a wide range of probable scenarios can be taken. The model allows all unforeseen challenges to be solved, mostly within an hour, and has very good scalability that promises to gracefully manage even larger models in the future.
"What we have been able to accomplish is to make such innovative optimization solutions accessible and affordable to a wide spectrum of clients operating in diverse socio-economic environments," says Tarun Kumar, an optimization researcher at IBM's T.J. Watson Research Center in Yorktown Heights, New York.
As stochastic models become more sophisticated, researchers like Gyana Parija have been able to infuse the models with "human" factors, such as politics, custom and culture. As researchers factor in human behavior in the models, the results grow less uncertain and more accurate and acceptable.