On-demand data assimilation of large-scale spatial temporal systems using sequential Monte Carlo methods

The proliferation of sensors is generating rapidly increasing quantities of data like never before. These extensive amounts of data can provide useful information for more accurate state inference of large-scale spatial temporal systems. Sequential Monte Carlo methods are used to assimilate the observed data from sensors to improve the state estimation of large-scale spatial temporal systems, which highly rely on the available real time observation data. In many scenarios, the real time data are limited in space and time. Therefore, it is important to effectively obtain critical sensor data in real time and then dynamically feed them into the running model. In this paper, we propose the on-demand data assimilation method for large-scale spatial temporal systems, in which we quantify the spatial states using run-time state quantification methods and decide if we need to trigger data assimilation on demand and obtain more relevant real time data when the state uncertainty is high. The effectiveness of the developed framework is evaluated based on large-scale wildfire spread simulations.