What Opportunities does Uber’s H3 Index offer for Weather Analysis and the Insurance Industry?
Uber has developed a grid system to analyse large spatial data sets through partitioning
areas of the Earth into a grid system by using hexagonal, hierarchical indexes styled as ‘H3’. This development was driven by the need to optimize ride pricing and dispatch and to bucket events into discreet areas, with the ability to conduct analysis at different levels of detail.
There are 16 different H3 resolutions for covering the globe, with each increment in
resolution producing a finer granularity. A top 0 Level Resolution will cover the globe with 122 unique hexagons, whilst a Level 16 Resolution would be represented by more than 550 trillion hexagons enabling a precision down to 0.0000009 km². Truncation of child cells can readily occur to find ancestor coarser cells and conduct indexing.
Figure 1: Subdividing areas into smaller and smaller hexagons (Copyright Uber)
Administrative and political boundaries such as the UK’s Postcode District to Postcode Point and the UN’s Global Administrative Unit Layers (GAUL) are examples of arbitrary definitions of zones, defined by natural features or socio-economic characteristics, such as numbers of population. These zones can change over time and the absence of uniformity causes substantial analytical difficulties. In the insurance industry, the use of administrative boundaries for analysis is common. For example, PERILS is a Swiss based independent reporting agency providing industry wide catastrophe insurance data. For the United Kingdom, PERILS takes data for events related to extratropical windstorm and flood events impacting property and causing losses greater than Euro 200 million and presents to the geo-resolution of ‘Low-res CRESTA’.
Figure 2: CRESTA Screenshot of UK Low Resolution Boundaries (Copyright CRESTA)
Understanding property exposures and different types of risk is fundamental to the insurance industry. Aggregation of data is a prerequisite for the needs of the property insurance industry to achieve a suitable level of organisational knowledge. There will always be a clustering effect of exposure, for example a business park where advanced manufacturing factories are located or a defined area of high value residential property. Comparing data such as the financial value at risk using different spatial boundaries can determine the appropriate H3 Resolution to be used and mitigate the potential for coarse representation of different levels of exposure. Loss events related to human activity such as theft and arson, tend to have a temporal factor and applying temporal granularities can help develop risk models for criminal activity.
Balkerne sees several opportunities for applying the H3 Index in our work. We believe there is value in a spatiotemporal approach to describing exposure and losses and to experiment with predicting short-term and long-term future losses. Furthermore, the emergence of the regulatory requirement for climate related financial disclosures potentially creates opportunities for the application of the H3 Index to UK Climate Projections 2018 data. The highest available resolution data is at a 2.2km spatial resolution for forecasted weather, whilst other data sets, such as projected sea level rise and storm surges are at a 12km resolution.
Figure 3: Balkerne use of H3 Index for UK Historical Rainfall Observations on 01/01/2000
Rainfall in mm per 1km2: Green <2mm, Yellow 2.01mm to 5mm, Red >5.01mm
In the coming months, the Balkerne Engineering team will continue to experiment with the H3 index to seek to improve assessments of client exposure, the modelling of value at risk and the production of loss forecasts for extreme weather events. We welcome engagement with any insurance stakeholders or insured customers wishing to discuss further.