Fusing Social Media Posts into Daily Floodplain Delineation of Hurricane Harvey

Keywords: flood analytics, big data fusion, riverine and floodplain, HAND, flood impact
Abstract Type: Paper Abstract

Abstract

At times of flooding, there has been increasing availability of locational-based social media (LBSM) posts about the hazard. In fact, about 95 percent of geotagged tweets and crowdsourced data relevant to Hurricane Harvey were multimedia (i.e. images or videos). Using computer vision algorithms, prior research has shown that the accuracy of water depth (WD) extracted from LBSM images is about 66%-70%. These flood-scene images were located along the river channel as well as afar in the floodplain. Comparing to conventional data, the continuous records of stream gages are limited to the channel whereas the scattered High Water Marks (HWM) do not have a timestamp. Hence, these observations from social media provide valuable information to supplement existing data sources in both space and time. The primary objective of this research is to incorporate those water depths to reconstruct the daily flood landscape in both riverine and floodplain during Hurricane Harvey in 2017. This research proposed a framework that integrates heterogeneous data sources to expand the riverine WD and delineate the floodplain. The discrete observations from LBSM and HWM are queried by each day and incorporated with the continuous records of USGS stream gage to derive the daily maximum WD. Using the Height Above the Nearest Drainage (HAND) concepts, the daily maximum WD were then interpolated to infer the extent of possible floodplain. The results revealed uncertainties from the data sources and subsequent modeling. Nevertheless, the findings are insightful to evaluate the potential of LBSM-based flood analytics in near-real time.