Crowdsourcing has been proposed on several occasions as a tool to provide a valuable complement to spaceborne data in various types of applications involving geospatial data . Crowdsourcing for damage assessment is also a popular application where “citizen sensors” can support the institutional coverage provided by spaceborne sensors . In this context, we give our modest contribution by analysing the case of hurricane Harvey, which swept across Barbados, Saint Vincent and the Grenadines, the Mexican area of Yucatan, and the US states of Texas and Louisiana, between 17th August and the 1st of September 2017. Harvey was the direct cause for 68 casualties in the US, ranking just second after Sandy (year 2012); damage estimates range in the hundreds of billions of euros. DigitalGlobe, a well-known satellite data provider, offers a catalogue of data related to natural disasters under their Open Data Program, including hurricane Harvey. Pre- and post-event RGB images in .tiff format from the WorldView-3 sensor are distributed together with a vector map where pointers highlight areas of Texas where damage was recorded. Remarkably, some of the worst-hit areas like the city of Victoria, Texas, were imaged soon after the event with acceptably low cloud cover extent. This permitted meaningful comparison between pre- and post-event data, and detection of damaged buildings by visual inspection. At the same time, the twitter archive was scanned for tweets related with Victoria, TX and hurricane Harvey. 53 tweets with pictures attached were identified by keyword search, sent between 26th of August and 5th September. Of these, just 1 was geo-located; of the others, 18 were localized based on the content of the image itself, by e.g. searching online for terms appearing on signs and then confirming the location of the identified businesses through preevent street-level images from other sources (e.g. Google Street View). Overall, about 36% of images could be assigned a location. Results from spaceborne data and tweet pictures confirmed each other, leading us to believe that a deep-learning-based automated system for damage assessment, based on tweet images and the ad-hoc deep learning framework described in  could well integrate a satellitebased damage mapping system. The system in  is being furthered under an European Space Agency (ESA) Kick-Start Activity (KSA - EMITS reference AO8872)  Yifang, B., Gong, P., & Gini, C. (2015). Global land cover mapping using Earth observation satellite data: Recent progresses and challenges. ISPRS journal of photogrammetry and remote sensing (Print), 103(1), 1-6.  Dell’Acqua, F., & De Vecchi, D. (2017). Potentials of Active and Passive Geospatial Crowdsourcing in Complementing Sentinel Data and Supporting Copernicus Service Portfolio. Proceedings of the IEEE, 105(10), 1913-1925.  Yuan, F., & Liu, R. (2018). Feasibility study of using crowdsourcing to identify critical affected areas for rapid damage assessment: Hurricane Matthew case study. International Journal of Disaster Risk Reduction.  Iannelli, G. C., & Dell’Acqua, F. (2017). Extensive Exposure Mapping in Urban Areas through Deep Analysis of Street-Level Pictures for Floor Count Determination. Urban Science, 1(2), 16.
File in questo prodotto:
Non ci sono file associati a questo prodotto.