Bridging the information gap of disaster responders by optimizing data selection using cost and quality

Abstract

Natural disasters are chaotic and disruptive events, with compressed timelines and high levels of uncertainty. Comprehensive data on the impact becomes only available well into the response phase and data is scattered across organizations. Data heterogeneity issues are common. Consequently, responding organizations have difficulties finding data that match their information needs. We investigated the information needs of and the disaster management data available to both national and local decision makers during the 2014 floods in Bangladesh. We conducted 13 semi-structured interviews and three focus group discussions, collecting in this way input from 51 people, transcribed and coded them so that themes of information needs emerged. We mapped the information needs on the available data sets and determined which needs were not, partially or completely covered. We identified seven themes of in total 71 information needs and 15 data sets. The mapping revealed a significant information gap of timely and location-based data. Only 40% of the information needs are covered in time and 75% if no time constraints are considered. Instead of using all data sets, we optimized for coverage -with Integer Linear Programming-combinations of data sets against the costs of extracting data from structured versus unstructured data and against the quality in terms of timeliness, source and content rating and granularity. Without time constraints, three data sets yield already a coverage of 68%, whereas adding five extra data sets only gives an improvement of 7%. We recommend executing identification and mapping of available data sets on the information needs as part of Data Preparedness. Determination of the optimal combination of data sets can be used to extract data on information needs more efficiently. Currently, we did this manually, but future research will investigate automatic matching of information needs on data sets, by applying intelligent querying and semantic data matching.