SpatialKat is KatRisk’s event based portfolio analysis software capturing correlations in space and time for flood, wind and storm surge events.
Model Development Overview
Model Development Components
Global Correlations with SST
- Catastrophic events are globally correlated with sea surface temperature (SST) as the main driver. We can establish this through linking teleconnections to large scale droughts and floods.
- As an example, the figure below shows the maximum cross correlation of 3 month precipitation anomalies with ENSO. Through this teleconnection pattern between ENSO and precipitation we drive our stochastic non-tropical cyclone precipitation model based on an EOF based stochastic VARMAX model.
Tropical Cyclone Modeling
A 50,000 year event set has been developed for the Atlantic Basin. Wind, storm surge, and precipitation have been modeled for all events.
Defining Inland Flood Events
Modeled downscaled precipitation is used to drive a high resolution land surface and river routing model. The output of this stochastic model together with the high resolution flood maps is then used as input to our probabilistic loss model. The relevant input to the loss model are event flood footprints clustered in space and time as shown in the figure below.
Highlights of SpatialKat Software
- Event sets consist of 50,000 years of simulation
- In the US, tropical cyclone events include wind, storm surge, and precipitation driven inland flooding
- Model hazard data and vulnerability are open and can be customized
- As with the KatRisk hazard maps, inland flood modeling includes both pluvial and fluvial flooding
- Output can be generated from the location to portfolio level by event
- Sampled losses at the location/coverage level are propagated all the way up to the portfolio level considering financial terms, resulting in straightforward and transparent financial model calculations
- Loss computations are repeatable, transparent, and extendable. The technology stack is based on powerful open source technologies R (graphics and statistics) and C++ (fast code execution)
- Model run parameters such as number of samples and correlation are flexible and user specified
- Software is scalable, allowing for almost linear reductions in runtime with additional processors