“All models are wrong, but some are useful” - Using risk models to rapidly estimate post disaster impacts
By Emma Phillips
Usually the first questions after a disaster are “How many people are affected?” and “What’s the damage?” We want to know the hard numbers on how many people were affected and the potential impact on the economy – difficult information to ascertain in the chaotic aftermath of a disaster. Understanding the situation on the ground takes coordination, data, and time – exactly what you’re often missing during a disaster. Using catastrophe risk models before a disaster occurs can improve coordination, provide critical data, and be done without time constraints.
Thinking in terms of risk provides decision makers with data before the disaster occurs. Catastrophe risk models are a useful analytical tool to assess the likelihood and the cost of a potential future hazard event. Probabilistic risk models calculate risk by translating the effects of multiple hazard occurrences (for example, an earthquake or a tsunami), into probable damage, and damage into cost. With this information, decisions can be taken to manage future risk.
For example, the recently completed South West Indian Ocean Islands Risk Assessment and Financing Initiative that was conducted in The Comoros, Mauritius, Madagascar, Seychelles, and Zanzibar provides the Island States with their disaster risk profile. Generated through probabilistic risk modelling, the profiles provide an overview of the countries’ risk from tropical cyclones, earthquakes and flooding. Risk is quantified as the likelihood of losses at a variety of return periods, and the average loss per year caused by the hazards.
For instance, in Madagascar, modeling suggests that the annual average loss from tropical cyclones is $87 million. The model used to generate this loss incorporates potential tropical cyclone events over a period of 10,000 years, the economic exposure of the country (the replacement value of infrastructure and assets) and vulnerability functions (the infrastructure or asset’s response to the tropical cyclone events). In any given year loss from tropical cyclones can vary widely. In some years Madagascar could be fortunate and experience no tropical cyclones. In other years, Madagascar could experience one or multiple severe tropical cyclones. Further, the model allows for the losses to be disaggregated by sectors. In this case, the data indicates that the residential sector is most vulnerable.
Giving governments hard numbers on the cost, and the probability of occurrence, of a disaster enables important discussions on which measures should be taken to better prepare for this risk, and in which priority sectors. But catastrophe risk models can be taken further. They can also be used to quickly estimate economic losses in a post-disaster situation when information is scarce, and having some information is better than no information at all. This approach was recently deployed in Madagascar.
On March 7, 2017, Tropical Cyclone Enawo made landfall in northeast Madagascar. The humanitarian impact was expected to be high based on the storm’s strength and population densities within the projected path of the storm. Having access to the modelled data from the South West Indian Ocean Risk Assessment provided useful risk information to our team ahead of time. But we wanted more details: we wanted to get an estimate of the losses caused by the cyclone, without having to do an on-the-ground assessment.
By re-running the model, this time with local data from the Madagascan Meteorological agency, we could get some estimates of the projected impact of Tropical Cyclone Enawo on infrastructure (buildings, critical facilities, transport networks). Direct losses were estimated at approximately $208 million (2015 currency), corresponding to a return period of approximately 11 years. This number is comparable to Tropical Cyclone Gafilo in Madagascar in 2004, which caused damage of more than $250 million to both agricultural production and capital stock.
What was exciting about this approach was that our team was able to provide the Government with an estimate within days of the disaster. Although a detailed ground assessment is still necessary to design a robust response and recovery plan for the country, what I learnt from the process was that while the numbers from an in-depth assessment may differ from the initial modelled estimates, this approach still enabled the government to understand the big picture impacts immediately post disaster. To paraphrase the statistician George Box: although the model might not simulate the exact reality, it was still useful.