Multi-Hazard Risk Assessment for the Comoros

Weather Impact has joined a consortium with Haskoning and the University of Twente to develop nationwide climate-driven multi-hazard maps for the Union of the Comoros, a small island state off the coast of Mozambique. By using the latest meteorological and hydrological methods and combine it with a high-resolution Digital Elevation Model (DEM), these maps will provide a strategic tool to strengthen the operational and analytical capacities of institutions responsible for managing climate and natural risks in the Comoros. The mapping will support the identification of high-risk areas for landslides, floods, coastal submersion, and even seismic and volcanic hazards.

Cyclone Kenneth

Following the destructive passage of Cyclone Kenneth over the Comoros in 2019, the urgent need for detailed and reliable meteorological risk information became clear. Such information is essential both for understanding current risks and for preparing for the impacts of ongoing climate change.

Fig.1 Rainfall simulated with WRF during the passage of Kenneth on 24th April 2019. The top left figure shows the input data (ERA5) and the other figure the WRF simulation at increasingly high resolution. The bottom-right figure is at 1x1km resolution.

Return period for rainfall event

Weather Impact’s specific contribution within this PRRC project focuses on producing high-resolution (1×1 km) precipitation and wind data. These datasets make it possible to calculate return periods for both the present climate and for future climate scenarios. The return period of a meteorological variable, such as rainfall or wind speed, is the average time between events of a given magnitude or greater. For instance, a “100-year rainfall event” does not mean it occurs exactly once every century, but rather that in any given year there is a 1% chance of such an event happening. These data form the backbone of the multi-hazard mapping work carried out by our partners.

How do we obtain high resolution weather data?

To create hazard maps for the current climate, we are using a hybrid approach that combines two methods, because running detailed climate simulations for every past weather event would be really demanding on system resources. Instead, we focus on about twenty of the most extreme rainfall and wind events that have affected the islands. For these cases, we use the Weather Research and Forecasting (WRF) model to recreate the events at a resolution of 1x1km, giving us a highly detailed view of local conditions. At the same time, we apply a statistical technique known as Random Forest, a type of supervised machine learning method, to the full historical weather record from 1980 to 2020. This allows us to “downscale” the dataset to the same fine 1x1km scale without the heavy processing of simulating every event with WRF. By combining these two approaches, we can estimate how often very intense rainfall and wind are likely to occur at each location on the islands, expressed as return periods, which describe the likelihood of rare events. These estimates then form the basis for the hazard maps.

Fig. 2 Downscaled wind data during Kenneth passage on 24th April 2019. On the left, the input data at about 30x30km resolution (ERA5), on the right the statistically downscaled wind at 1x1km resolution (only over land).