RESCUE Tsunami Portal Overview

The Tsunami Portal is a GIS desktop client-based application consisting of a situation dashboard and analysis toolkit. The application facilites awareness from multi-modal data and enables end-users to apply situational awareness information in decision making and analysis. It provides integrated information from multiple sources as well as a suite of analytical tools, query languages, triaging and filtering functionality. Such a system can be used by emergency managers for planning analysis and logistics tasks, and to continuously monitor events as they unfold. For instance information extracted can be used for evacuation and route planning, and clusters of information can be used for damage and impact assessment. For initial damage assessment, loss estmation tools have been integrated into the system.

The Tsunami Portal was developed and deployed during the 2004 Boxing Day Tsunami disaster in Southeast Asia. During the first three days following the incident, more than 4000 reports were gathered and within the next month over 200,000 news reports were gathered from a variety of online sources. The content obtained was diverse and heterogeneous in nature including text, multimedia (images, video), and GIS data (satellite imagery, maps). The Tsunami Portal provided a single, integrated view of all of this data.

The Tsunami Portal was created by Project RESCUE as part of its mission to significantly enhance the decision-making capabilities of first responders, response organizations, and the public in the context of a crisis by providing timely and seamless access to accurate, reliable, and actionable situational awareness information derived from multiple sources and modalities. In addition to the Tsunami Portal, which provides for the monitoring of a single, large-scale incident, the RESCUE team has also developed a portal framework which targets local governments and their communities. More information about this project can be found at the Disaster Portal project website.

Major Features of the Tsunami Portal

Data Collection

Relevant information is collected and stored from multiple online information sources (news reports, blogs, satellite images, etc.) The Tsunami Portal also provides real-time access to sources when appropriate (e.g. weather data). The data collected is stored in a relational database along with pertinent metadata.

Topic-Based Browse and Search

Data is organized by topic using automatic topic modeling tools, such that the user can easily browse or query the large number of documents which are contained by the system. This improves retrieval, and also enables an analyst to gain insights at a semantically enriched level.

     

Data Visualization

Users can perceive the spatial, temporal, and topical contexts of the data in a GIS-based summary. The analyst can interactively examine the data returned based on the geographic information stored in the documents. Other georeferenced data including still images and reconnaissance video can also be displayed via appropriate interface extensions.

                 

Research Incorporated into the Tsunami Portal

Topic Modeling & Information Extraction - Data is first organized by topics, which are formed by features such as sets of keywords. However, topic modeling alone is not sufficient to be able to capture knowledge and concepts embedded in the data. Semantic concepts like causality require deeper information extraction techniques. We have developed a next-generation extraction framework which exploits existing lower level text extractors such as GATE for identifying tokens or entities, and provides further deductive rule-based capabilities on top for slot filling required for relation or event extraction.

Modeling Uncertainty - Information extracted from raw data might not always be precise due to incomplete knowledge representation or extraction errors. In order to model the uncertainty involved in such information and efficiently support analysis, we utilize a probabilistic modeling framework defining the semantics of the uncertain spatial dimensions. The spatial properties of the event are projected onto a 2-dimensional domain. Uncertain event locations are modeled as random variables with probability density functions (PDFs) associated with them. These PDFs are are then mapped back to the 2-dimentional domain using GIS and probabilistic modeling tools.

Profile-based Personalization - User filters and profiles capture user characteristics in order to achieve personalized information dissemination during disaster situations. When new data is obtained, it is scored by the individual filters and if the relevance score is beyond a certain threshold, it will be disseminated to the user. Relevance scores can also be used to prioritize the information provided to the user. A similarity retrieval and refinement framework allows the user to express the initial profile using similarity semantics, and then automatically enhances the profile based on continuous user feedback.