Classical Big-Data problems can be summarized (from a very simple and schematic perspective) according to two main goals: it is mainly about (i) increasing the data storage capacity (to be able to collect and store as much data as possible, including historical consideration) and (ii) improving pattern recognition feature (to be able to parse the obtained gigantic data fields and find quickly and correctly specific patterns). Most technological progresses made these years in Big-Data domain are dedicated to one or both of these expectations (more storage or more parsing). Actually, this approach is perfectly adapted to most of the domains in which Big-Data is one perspective of improvement because: (i) data sources are well known and (ii) exploitation of data field through pattern recognition can serve a lot of simple objectives. However, as presented in the first section, this is exactly what DM is not.
In  the authors particularly describes how big-data can be considered as a new challenge in DM, but not in a “classical” way (mainly considering the increasing amount of data and dataflows): the way this large amount of data could be treated to become intelligible to exploit it relevantly (automatically or not) is at the heart of this new challenge. This article considers that Big-Data for DM should perform the following specific features.
3.1 Gathering: From Data Sources to Data Storage
Considering that data sources on disaster site might be unknown, randomly distributed and potentially doubtful in terms of veracity, the first layer of the Big-Data framework is in charge of identifying the data environment to be considered. This objective is a very innovative one and really arduous on a technological point of view; actually, Big-Data approaches are usually based on predefined data background with selected data sources.
The expectation is to perform discovery (of unknown data sources), understanding (of the emitted data), trust and veracity evaluation (of the understood data) and subscription (to discovered, understood and accepted data sources) mechanisms. It is absolutely necessary, in the connected world of DM, to make the gathering layer open, i.e. to allow Big-Data for DM to deal with a dynamical data environment involving incoming and outgoing unknown and non-dedicated data sources.
3.2 Interpretation: From Data to Information
The main stake for Big-Data in DM is probably the ability to interpret gathered data in order to automatically build situational models (such situational model might be considered as a common operational picture). This step is really the corner stone of Big-Data for DM. Actually, while classical Big-Data aims at exploring gigantic fields of data to detect patterns and thus at providing some direct results, DM Big-Data must insert an abstraction level between raw data and exploitation results. This intermediary level is in charge of combining and transforming collected data into intelligible concepts. This level is mandatory in the case of DM because data sources and data types may be unknown and consequently not directly exploitable.
The theory defended in this article states that this information level, to perform interpretation requires a situational metamodel (presenting concepts and relations inherent in DM domain). Such a metamodel (see the one described in ) provides the theoretical background to create a situation model by instantiating concepts and relations with gathered data.
3.3 Exploitation: From Information to Knowledge
Finally, the last level for Big-Data for DM is about the ability to exploit fruitfully the generated and formalized information, to support the coordination of responders. This level concerns mainly exploitation features to perform decisions and actions. Actually, there are a lot of DM tools covering coordination functionalities (such as first-responders alert, collaborative tasking, monitoring of coordinated actions, or orchestration of collaborative workflows), which would strongly benefit from such information base: these tools require real inputs to perform their service; consequently, providing these tools with consolidated information basis would remove this constraint and furthermore would ensure a higher quality of service considering that the collected and interpreted data would certainly supply robust and trustable situation models.
3.4 Big Picture of the Presented Big-Data Framework for Disaster Management
On the following figure, the three levels presented above are conceptually presented.
The most noticeable elements of Fig. 1 are the following:
The abstraction path from data to information to knowledge inherent in the Big-Data for DM.
Data level and Knowledge level are supposed to be open (allowing the use of emerging data sources and the exploitation of new DM tools).
The central role of the Metamodel to perform data interpretation and information formalization.