Aerial Insights: Earth Observations for Global Flood Risk
By John Galantowicz
Every day, a battery of Earth-observing systems is collecting an astonishing breadth of information globally. Satellites in space as well as gauges, instruments, and sensors across Earth’s surface together are providing a continuously updating data stream. They watch for long-term changes in natural land cover and human land use; rain, wind, and temperature changes in the atmosphere; water in the soil and over land; extreme flows in rivers; and rising tides, tsunamis, and storm surges in the oceans.
In short, these systems paint a complex, valuable, and comprehensive portrait of our ever-changing planet—ideal data sets for assessing flood risk and informing parametric insurance structures anywhere on the globe.
Or are they? For every clear day that affords a satellite’s stunning view of Boston or Ouagadougou, there are cloudy days with no view at all. For every well-gauged river, there are countless rivers that go ungauged. And for every full-color image that leaves no doubt about what is and is not flooded, there are long periods without any clear imagery, leaving ambiguities about true inundation extent, duration, and severity. What’s more, hydrological models can offer an alternative of virtually uninterrupted river discharge simulations with the capacity for generating statistical representations of extreme flood frequency over time spans far exceeding the Earth observation (EO) historical record.
Despite the limitations, there is a strong argument for the use of Earth observations in global flood risk management. In developing economies, EO imagery and EO-derived information may be the best or only choice, because alternatives like hydrological models are not yet available, lack critical inputs such as rainfall rate data, or do not yet have the track record needed for wide acceptance in risk-transfer markets. But work remains for EO to reach its full potential as the basis for parametric insurance policies. The successful use of EO data requires algorithms that turn the raw data into products and clear communication about those products, including their strengths and weaknesses, to risk owners and all parties affected by risk transfer.
Embracing precision, accepting ambiguity
Satellite-based EO data is the most utilitarian for flood risk management in developing economies because of its ubiquity and often free availability. But among satellite systems, there’s a trade-off between precision and timeliness. For example, the most precise flood mapping systems can show that a property is 30 percent flooded at the observation time. However, assessing flood duration or mapping maximum flood extent requires using less precise systems that show, for example, which neighborhoods or ZIP codes experience flooding but not which addresses. The successful use of EO data in parametric insurance necessitates contract structures that accept the ambiguities inherent in the timeliest flood data.
Precise EO data on a global scale is epitomized by commercial, high-resolution satellite imagery. Imagery with meter-scale or better resolution has been commercially available since the late 1990s, albeit for a hefty price. Imagery resolution, costs, and revisit frequency have all improved over the last two decades while data volumes have risen simultaneously. Industry leader DigitalGlobe boasts an imagery catalog of more than 100 petabytes, growing by 100 terabytes a day. Today, the booming SmallSat industry—which already has hundreds of EO satellites in orbit—is further accelerating data acquisition frequency and accessibility.
There’s no doubt that civilian satellite imagery falls short of aerial photography for quality and that SmallSat image quality is still more limited. But what they lack in single-scene accuracy, they more than make up for in global coverage and observation frequency—features critical for understanding changing conditions on the ground and democratizing data access in otherwise data-poor regions. Perhaps most important for flood risk management, meter-scale imagery—coupled with advanced image analysis algorithms—now makes global building-footprint databases feasible, allowing for rapid, large-scale assessment of property and population vulnerabilities to floods. And when clear skies prevail during and immediately after a flood, high-resolution imagery is the gold standard for assessing the extent of flooding and flood damage.
What the meter-scale imagery catalog lacks most is a lengthy time series of maximum flood extents—not to mention additional attributes like flood duration and depth. Yes, there are the occasional stunningly detailed clear-sky images, but where historical records are scant, the ideal data source for flood risk management would be a consistent and accurate EO-based flood mapping system. A historical flood extent catalog would help estimate flood return frequency. In addition, where large floods were observed, it could help quantify the potential impacts of a future worst-case event given current exposure data. The system would preferably link historical and near-real-time flood mapping so that parametric insurance payout triggers tied to the system’s products could be calibrated to its catalog.
Is an EO-based flood mapping system that fits the needs of parametric flood insurance markets feasible today? In my assessment, yes. First, the necessary data is available from a fleet of lower-resolution optical satellite sensors and specialized EO data based on passive microwave sensing.
- Lower-resolution optical observing systems, like the 30-meter Landsat jointly run by the U.S. Geological Survey and NASA and 250- to 500-m-resolution satellites operated by NASA and the National Oceanic Atmospheric and Administration (NOAA), have produced imagery of many large-scale floods since 1973. Although these systems are as obscured by cloud cover as the more precise meter-scale optical systems, when used together they can capture the extent of a large portion of flood events.
- Passive microwave sensors see through clouds and some vegetation, making them useful for filling the observation gap in the optical sensor record. In operation since 1998, these sensors make measurements across 15- to 30-kilometer footprints. Two algorithm types—developed separately by researchers at the Dartmouth Flood Observatory (DFO), now at the University of Colorado, and Atmospheric and Environmental Research (AER)—estimate flooding at smaller scales. The DFO approach estimates river discharge at selected points globally. The AER approach estimates flood extent daily over entire continents at 90-meter or better resolution through a downscaling process based on relative floodability scoring.
By reconciling the differences between microwave- and optical-derived flood maps, the two can be used interchangeably to underlie appropriately scaled parametric risk-transfer programs, provided that the flood detection limitations are well understood and policies are structured to minimize basis risk.
Flood mapping to financial risk transfer
The sovereign risk-transfer programs run by the African Risk Capacity (ARC) are a case study in the value of satellite EO in parametric insurance for developing economies. ARC is a specialized agency of the African Union (AU) charged with helping AU member states improve their self-capacity to respond to extreme weather events and protect vulnerable populations. ARC has coordinated index-based drought insurance policies for African governments since 2014 and is now testing river flood insurance in a pilot program. The indices for both programs are underpinned by EO data: the drought index incorporates satellite-based rainfall data, and the flood index uses a version of AER’s flood maps derived from microwave sensor data. The indices tie EO to proportionate national response costs through population vulnerability factors, including household resilience and exposure to losses (for example, farm income and displacement).
ARC’s approach has subverted the traditional disaster assistance paradigm in two ways. First, by structuring risk transfer as parametric insurance, payouts can be made quickly in the wake of an event and put to use by governments through a prepared action plan. Second, the sovereign risk pool allows countries to manage risk as a group and efficiently share financial resources.
ARC’s programs require index EO data to have at least 20-years of historical time-series length. In addition, the data must be consistent over time and across countries, transparently produced, algorithmically objective, and reliable in near real time. Leveraging EO data that meets these requirements, ARC works with member states to design and validate an index matched to in-country records of past flood events and their impacts. For ARC’s programs to succeed in the long term, these indices must be coupled appropriately with vulnerability and resilience assessments and supported by effective governmental management of disasters both above and below payout attachment levels. In the near term, ARC’s experience suggests that proponents of similar parametric insurance programs for flood should expect a high standard of evidence before parties agree on a flood index and triggering thresholds.
The challenges ahead
Nowhere is the need for innovative flood risk assessment approaches more apparent than in the urban areas of developing countries. Urban areas are messy, sometimes with dense pockets of unplanned, improvised housing in lowlands prone to frequent flooding. Natural risk factors are amplified by impermeable surfaces that prevent rainfall from soaking into the soil and poorly maintained storm sewer systems that are easily overwhelmed by heavy rain. The most vulnerable urban populations are also the most exposed.
High-resolution EO can provide urban area building footprints, but advanced technologies (for example, lidar) are needed for high-accuracy elevation mapping. Lower-cost weather radar systems, if feasible, are needed to achieve rainfall data accurate enough for flash flood warnings. Many other factors, such as urban drainage network topologies and day/night population movements, may be out of reach for current EO systems.
The best way forward is likely through systems that can combine flood EO, simulation and prediction models, and unconventional data. For example, analytics are now available that assess flooding based on social media posts. While each data type alone may be highly ambiguous, together they may triangulate to an objective flood index that’s verifiable, acceptable to interested parties, and workable in a parametric insurance framework.
John Galantowicz, Ph.D., is principal scientist at Atmospheric and Environmental Research, Inc. (AER), a Verisk (Nasdaq:VRSK) business.