Understanding the Extent of Flooding in Houston from Hurricane Harvey

For over a week now, we’ve seen evidence of how Hurricane Harvey wreaked havoc on Texas, with surreal snapshots of buildings, roads, and other structures underwater. But with such a large area affected, it’s difficult to gain a comprehensive understanding of the extent of the flooding and its severity. Eye-witness reports or photos from certain areas provide a glimpse, but what is needed is a complete overview, rich in detail and scope. It’s a big-picture problem — the kind that’s perfect for a macroscope to address.

Capturing the fullest extent of flooding in a specific area is difficult. It takes time for floodwaters to reach their peak level, which means we need frequent updates for a given location to accurately track its inundation. Finding frequent, clear images can be challenging, though, when flooding is the result of a storm system, as cloud cover might not clear for days. Finally, even though satellites are achieving higher revisit rates, we’re a long way from covering every square mile, every hour. This means that we need to be able to intelligently infer which areas are underwater based off of available imagery and our knowledge of regional topography.

At Orbital Insight, our vision is to help increase global transparency — our understanding of what we’re doing on and to the Earth. As part of that vision, we want to apply our technology to help solve humanitarian issues like deforestation, poverty, and natural disaster response. From the moment we began receiving imagery of the areas impacted by Harvey from our satellite providers, we started applying our proprietary flood-detection algorithms so that we might contribute to the understanding of the scope of this disaster.

Our first step was to map out the flood peak for the Houston area. To get the most comprehensive understanding possible, we processed an enormous amount and diverse variety of satellite imagery from our many imagery provider partners. Optical imagery, synthetic aperture radar (SAR) imagery, high-resolution, medium-resolution — we took everything we could get in order to have the most pixels possible showing flooding. Our framework was able to quickly process all of these diverse inputs.

Even with all this imagery, however, we knew that we hadn’t captured every pixel of flooding in the Houston area. Because of this, we needed to be able to apply the information we did have in order to fill in the pieces that were missing. To do this, we used Digital Elevation Maps (DEMs) from the State of Texas, along with topography maps from the U.S. Geological Survey (USGS), to define and map out watersheds. A watershed is an area that possesses similar hydro-properties, meaning that any water that rains down in that area drains to the same lowest point, which is typically a lake or river bed. Once we overlaid a map of where we had detected flooding, we used the DEMs and watershed outlines to determine the complete “flood fill” for each area. This is where our mathematical models and data science worked their magic: if we knew from our imagery that water was at a certain elevation in a given watershed, we could then calculate that any area below that elevation would be flooded, too. It became a simple physics problem.

Map of observed flood extent (source: DigitalGlobe)

Top left: Map of observed flood extent (source: DigitalGlobe); Top right: Digital Elevation Map (source: National Elevation Dataset); Bottom right: Cross-sectional view of flood depth (source: Wolfram Demonstrations Project); Bottom left: Digital elevation points for observed flood extent regions (source: Texas Natural Resources Information System).

Now, we had a model for the full extent of flooding in an area based on both observable imagery and our knowledge of the area’s topography. This builds a more comprehensive understanding than traditional models show because it takes into account multiple data sources and combines them to create a more accurate estimate. We’ve also begun benchmarking our results, utilizing crowdsourced geo-tagged imagery to confirm that locations, where we estimated water to be, were indeed flooded. Through this process, we’ve seen very positive results thus far for our approach.

We believe this type of analysis will be useful for policymakers, first responders, insurance companies, and others. In fact, our current preliminary analysis shows a total flood extent that is more than three times larger than previously reported. As we continue to refine this understanding in the coming days, our hope is that this focus on geospatial analysis can be used to benefit people at times when they most need help. We’re proud to be able to employ our technology for good and look forward to continuing to explore these capabilities further.

Output of processed SAR imagery indicating observed flood extents

Left: Output of processed SAR imagery indicating observed flood extents (Source: Google Street Map, Sentinel 1 with Orbital Insight processing); Right: Actual flood maps after applying Orbital Insight’s geospatial interpolation across observation points and DEM (Source: Orbital Insight, Google Street Map).

If you’re interested in hearing more about our Harvey flooding results or think they might be useful to your organization, please reach out to us at