News — 26 April, 2017
HOT Research Partnership on Crowdsourced Damage Assessment
HOT is partnering with the Stanford Urban Resilience Initiative (SURI), the World Bank’s Global Facility for Disaster Reduction and Recovery (GFDRR), Heidelberg University, and the University of Colorado, Boulder to find better methods to provide information on the impacts of natural disasters.
During nearly every large activation, HOT’s partners ask us if our community can help identify collapsed buildings, damage to roads and bridges, or the condition of other important assets. This information is vital to guiding both immediate response and longer-term recovery work, but due to limitations of post-disaster imagery and the difficulties in accurately assessing damage from above, it’s been very difficult to meet these demands.
This new research effort is aimed at exploring several ways to resolve these limitations. Our approach starts from the recognition that most users of damage data don’t require building-level precision to meet their information needs. The research team is in the process of interviewing experienced practitioners about the ways that damage assessment informs their work during disaster response and recovery activities. We’re early in the process, but it’s clear that for many purposes, damage and impact data delivered in a timely manner is more useful than extremely precise information that arrives too late.
Our current experiments are oriented toward informing the Post Disaster Needs Assessments (PDNA).
PDNA’s are government-led processes that seek to create harmonized understanding of the impacts of disaster between the affected governments and donor countries that can help guide overall recovery planning. The PDNA typically takes place about a month after the disaster and needs to be accurate to general orders of magnitude in order to gauge the extent and characteristics of disaster impacts. With these constraints in mind, we are looking at:
- Statistical methods of aggregation - while accuracy to the level of individual buildings might be out of reach given current methods, our team is looking at options for aggregating results to ward or district levels, including multiple passes per building and weighting user responses based on various factors, such as contributor accuracy and experience.
- Area based assessment - Rather than ask participants to rate the level of damage to individual buildings, we are testing if scoring the amount of damage in a given area (tile at zoom level 18) on scales of 1-5 and 1-10 could produce estimates that are useful to the level of precision required by PDNA processes.
- Comparative damage rankings - We are also testing a method by which participants would simply be given a satellite photo of two separate areas and asked which image shows more damage. By repeating this, over thousands of images that make up disaster-affected areas, we hope to be able to develop a relative understanding of damage that could be calibrated for use with the PDNA using targeted field assessment.
Yesterday was the second anniversary of the devastating 2015 Nepal earthquake that killed nearly 9,000 people and injured nearly 22,000. The tools and methodology that we are developing are aimed at improving the way that online volunteers can play a key role in responding to such disasters. More accurate and rapid estimates of damage can help guide responders on the ground as well as to determine immediate emergency needs and inform long term recovery strategies.
We are launching these experiments towards the end of May and will send out a call for participation to the HOT community then. We hope to involve as many of you as possible in the experiments and in discussions of the results through on-line meetings and in person events such as the HOT Summit in September. Stay tuned!
This project is supported by the National Science Foundation (NSF) grant 1645335/EAGER - “A dynamic, reliability-weighted, multi-pass probabilistic framework to reduce uncertainty in crowd-sourced post-disaster damage assessments”. Satellite imagery is kindly provided by DigitalGlobe through the Open Data Program.
This update was written by Robert Soden, visiting fellow at Stanford University and HOT community member.