Modelling the impact of Hurricane Maria on Dominica (what can we simulate and what can we predict?) (Speaker: Victor Jetten (University of Twente - faculty ITC))

Speaker: Victor Jetten (University of Twente - faculty ITC, with contributions of: Cees van Westen, Bastian van de Bout and Sobhan Emtehani

In September 2016, hurricane Maria hit Dominica a few weeks after hurricane Irma devastated St Maarten. There were 65 casualties and approximately 1.37 Billion US$ damage. Because the island is an independent country with 70000 inhabitants, it basically has to cope with this alone, and uses World Bank loans for cleaning and rebuilding. There were several types of damages: wind surges destroyed most of the roofs (currently being rebuild by the Red Cross), the coastal zone suffered sea surges, there were flash floods, landslides and debris flows. The wind also stripped large swaths of forest from its leaves. When dealing with such disaster, stakeholders generally want to know three things: where does it happen, how bad will it be and when can we expect the next one. In scientific terms: what is the multi-hazard spatial variability, what is the risk (potential damage) and what is the recurrence interval or probability.

To answer these questions, the ITC has been doing research on Dominica and other islands for several years with staff and students, focusing on flash flood and landslide hazard assessment and assisting in the damage assessment after Maria. We developed a multi-hazard simulation model, that models runoff, erosion, flash floods, landslides and debrisflows simultaneously, because these processes influence each other in reality, and most models predict only one of these processes. However such a model is very hard to calibrate and validate, for instance river water level measurements are missing. Currently we are attempting to compare simulated sediment load to observed sediment deposits after the hurricane, and flood traces on the walls of buildings, and there is a possibility to use smartphone pictures and movies taken during the event. With this multi-hazard model we can make estimates of where and how much.

The probability of occurrence however is difficult. Extreme Value Statistics show that hurricanes do not always follow a frequency magnitude relation when we look at the amount of rainfall. Also, the multi-hazard simulation is done with design events instead of with real rainfall data, and design storm methodology is not particularly fit for hurricanes. Lastly, even with a known recurrence interval of a hurricane with a given magnitude, the antecedent circumstances are very important. A hurricane later in the season when the island has already received a lot of rain has a much bigger impact than when the island is ‘drier’. Moreover, hazard models need various other data layers that have their own uncertainty. Basically, we can simulate a lot but predict very little.