Assimilation of Real-Time Satellite And Human Sensor Networks for Modeling Natural Disasters

We describe the development of underlying technologies needed to address the merging of a web of real time satellite sensor Web (SSW) and Human Sensor Web (HSW) needed to augment the US response to extreme events. As an initial prototyping step and use case scenario, we consider the development of two major system tools that can be transitioned from research to the responding operational agency for mitigating coastal oil spills. These tools consist of the capture of Situation Aware (SA) Social Media (SM) Data, and assimilation of the processed information into forecasting models to provide incident decision managers with interactive virtual spatial temporal animations superimposed with probabilistic data estimates. The system methodologies are equally applicable to the wider class of extreme events such as plume dispersions from volcanoes or massive fires, major floods, hurricane impacts, radioactive isotope dispersions from nuclear accidents, etc. A successful feasibility demonstration of this technology has been shown in the case of the Deepwater Horizon Oil Spill where Human Sensor Networks have been combined with a geophysical model to perform parameter assessments. Flickr images of beached oil were mined from the spill area, geolocated and timestamped and converted into geophysical data. This data was incorporated into General NOAA Operational Modeling Environment (GNOME), a Lagrangian forecast model that uses near real-time surface winds, ocean currents, and satellite shape profiles of oil to generate a forecast of plume movement. As a result, improved estimates of diffusive coefficients and rates of oil spill were determined. Current approaches for providing satellite derived oil distributions are collected from a satellite sensor web of operational and research sensors from many countries, and a manual analysis is performed by NESDIS. A real time SA HSW processing system based on geolocated SM data from sources such as Twitter, Flickr, YouTube etc., greatly supplements the current operational practice of sending out teams of humans to gather samples of tarballs reaching coastal locations. We show that ensemble Kalman filter assimilation of the combination of SM data with model forecast background data fields can minimize the false positive cases of satellite observations alone. Our future framework consists of two parts, a real time SA HSW processing system and an on-demand SSW processing system. HSW processing system uses a geolocated SM data to provide observations of coastal oil contact. SSW system is composed of selected instruments from NASA EOS, NPP and available Decadal Survey mission satellites along with other in situ data to form a real time regional oil spill observing system. We will automate the NESDIS manual process of providing oil spill maps by using Self Organizing Feature Map (SOFM) algorithm. We use the LETKF scheme for assimilating the satellite sensor web and HSW observations into the GNOME model to reduce the uncertainty of the observations. We intend to infuse these developments in an SOA implementation for execution of event driven model forecast assimilation cycles in a dedicated HPC cloud.