Can Geo-Simulations Provide A Roadmap to Better Disaster Response?

Monday, January 29th, 2018 Cory RandJanuary 29th, 2018 • Bookmarks: 69

Bangkok, Thailand - November 4, 2011: An army boat ferries flood victims to safety on a flooded road in Pinklao district.

The 2017 Atlantic hurricane season was one of the priciest on record, mirroring a global trend of natural disasters becoming more expensive. Understanding the impacts of these natural disasters has become increasingly important. Researchers often model these impacts by creating relationships between indicators such as hurricane wind speed, predicted costs in lives and property damage. Natural disaster impacts can be harder to model in poor countries as dependence on subsistence agriculture, weak institutional capacity, and lack of insurance can catalyze secondary indirect crises far from the initial emergency, such as mass migration and starvation. Thus, policymakers need models to swiftly anticipate where secondary crises will occur without relying on immediate post-disaster data, which can be difficult to collect.

A recent article published in World Development introduces an innovative modeling technique that uses geo-simulation to quickly project indirect effects without post-disaster data. Unlike previous natural disaster response models, geo-simulation uses satellite data to map a disaster-affected region’s cities, villages, and roads. The model is then populated with agents representing low-income workers that must maintain a minimum level of food consumption to avoid starvation. When a natural disaster triggers a localized output shock, the model will reduce the workers’ incomes in that area. Consequently, workers may migrate to unaffected areas, depending on distance. If migration does not stabilize incomes, workers will dip into their savings to purchase food; if savings run out before incomes rebound, the workers will starve. The model shows how localized shocks can ripple between decentralized interconnected markets and tracks indicators that signal secondary crises, such as changes in regional incomes, food prices and savings rates.

Dr. Naqvi, the author of the study, applied this geo-simulation to the 2005 Kashmir Earthquake. The earthquake killed almost 75,000 people in Northern Pakistan and injured another 100,000. Among other indirect effects, the quake displaced more than 3.5 million people and left approximately 2.3 million people starving or food insecure. To examine where and when these indirect effects developed, Dr. Naqvi calibrated the model to the region’s pre-disaster economic trends and then subjected it to a localized shock corresponding to the earthquake’s fault line, with damage to villages and cities declining exponentially as distance from the fault increased.

The model simulated how the region would adjust over the course of one year following the earthquake and tracked how the region’s economy responded to the shock. The modeled earthquake immediately reduced economic output (55 percent decline) much more than labor (13 percent decline), and predicted that newly unemployed workers would migrate to unaffected areas in search of jobs and food. As migration distributed the labor surplus away from the localized shock, the model showed that regional incomes adjusted to a lower equilibrium and food prices rose from lost output. At the three-month mark, early signs of food insecurity appeared when worker savings rates began falling. Though the region’s pre-disaster savings rate was ten percent, the modeled savings rates in the region’s three cities all stabilized around zero percent. In some remote villages farther from the fault line, the savings rates dropped thirty percent; these workers quickly gained the model’s “starving” label.

The model showed that remote villages poorly connected to regional markets could not cope with an increase in both population and food prices. Yet in actuality, immediately after the 2005 earthquake aid efforts focused on the region’s three major cities, operating under the assumption that populations were likely to migrate toward urban areas in a post-shock scenario. The model’s contrary findings could have major implications for disaster relief.

Further work is needed in exploring applications of geo-simulations to natural disaster models. Dr. Naqvi concluded his article with suggestions for how his model could be made more robust: include more detailed spatial information (such as altitude and weather variations), add more complex agent behavioral decisions to incorporate multimember households, and differentiate agents’ skills and information. Dr. Naqvi’s early work shows promise that such a model could pinpoint where indirect effects occur long before reliable post-disaster data becomes available, enabling policymakers to determine where best to direct immediate vital resources.