Announcing SpaceNet 8: Flood Detection Challenge Using Multiclass Segmentation
Editor’s note: SpaceNet is an initiative dedicated to accelerating open-source, artificial intelligence applied research for geospatial applications, specifically foundational mapping (i.e., building footprint and road network detection). SpaceNet is run by co-founder Maxar and our partners Amazon Web Services (AWS), IEEE-GRSS, Oak Ridge National Laboratory and Topcoder.
Each year, natural disasters such as hurricanes, tornadoes, earthquakes and floods significantly damage infrastructure and result in loss of life, property and billions of dollars. As these events become more frequent and severe, there is an increasing need to rapidly develop maps and analyze the scale of destruction to better direct resources and first responders.
With this need in mind, the SpaceNet 8 Flood Detection Challenge will focus on infrastructure and flood mapping related to hurricanes and heavy rains that cause route obstructions and significant damage. The goal of SpaceNet 8 is to leverage both existing datasets and algorithms from SpaceNet Challenges 1–7 as well as new training data and a baseline algorithm, then apply them to a real-world disaster response scenario. This challenge also expands the task to a multiclass feature extraction and characterization problem.
Since its launch in 2016, SpaceNet has made significant progress advancing open-source building footprint and road extraction algorithms. During SpaceNet 8, challenge participants will train algorithms on imagery and labels from previous challenges — as well as newly created labeled training datasets from Maxar — to rapidly map an area affected by flooding. Any winning open-source algorithm from SpaceNet 1–7 may also be used. New areas of interest (AOIs) will include New Orleans, Louisiana, following Hurricane Ida in August 2021; Dernau, Germany, during the June 2021 floods across Western Europe; and a new “mystery city” for blind testing of the algorithms.
SpaceNet 8 aims to answer these questions:
- How have algorithms that extract buildings and roads improved since SpaceNet was launched, and how can top algorithms from previous challenges be leveraged?
- What is the impact on performance for a multiclass feature extraction challenge — i.e., buildings and roads?
- How accurately can roads obstructed by flood waters be characterized from pre-event road detections and post-event satellite imagery?
Algorithms submitted as part of SpaceNet 8 will be measured to determine accuracy in detecting roads covered by flood water and buildings with water present in the proximity.
The challenge will close August 23 and winners will be announced in early September. A total of $50,000 in prizes will be awarded to the teams that submit the top five algorithms, as well as to the top undergraduate and graduate teams.
Like previous SpaceNet Challenges, the winning algorithms will be released as open source and could have applications for disaster response and recovery, rapid mapping of remote regions and potentially also to assess the impact of man-made disasters.
In future posts, we will dive deeper into the SpaceNet 8 imagery and labels and provide further details about what makes this dataset unique.