For insurance companies to accurately model flood risk, primarily two sources of information are needed. First is information about flood severity, which can be derived from current and historic water location (rivers, coastlines, lakes), stores (glaciers, snowmelt), and levels (topology). Second is an analysis of vulnerability, which is based on the properties within these regions.
Property insurance firms require the ability to combine such information about natural structures with high geographic resolution. Historic information is of limited value due to limited detail of existing records (in both a spatial and temporal sense) and the rarity of flood events. Additionally climate change causes local change in weather patterns and surface modifications lead to altering water run-off pathways.
Despite a multitude of data sources, there has been a drought in actionable information and as a result flood modeling has been both uncertain and ambiguous. The uncertainty and ambiguity has tremendous financial implications as in the year 2013 alone, water-hazard-related losses amounted to nearly one trillion USD 1.
For an evidence-based risk modeling to be a reality, it is essential to have access to high-resolution data collected over the entire area of insured properties on a regular basis. With this dynamic information, it would be possible to couple the models for distribution of the flood severity with the vulnerability of the insurance portfolio and arrive transparently at a concrete, traceable value for total loss prediction. The modeled information can further be used in various analytical areas of general insurance firms such as in pricing, reserving, and capital modeling for solvency purposes.
In parallel, the latest satellite developments from NASA, Google, Airbus Defense EADS and others provide frequent (weekly) high resolution (10m) imaging of the entire planet. These measurements result in petabytes of data generation.
Our tools enable large, complicated satellite imaging datasets to be processed easily and scaled across hundreds of machines.
Here we locate and segment the water for a single day in the following images.
SELECT feature FROM ( SELECT SEGMENT_WATER(image.blue) FROM SatelliteImages ) WHERE AREA > 1000m^2 AND DATE IS 4-8-2015