[M-GI26] Data assimilation: A fundamental approach in geosciences

[MGI26-02] Ensemble Data Assimilation with Binary Observations: Theory and Application to Urban Flood Monitoring

Naoki Hiura¹, Duc Le¹, *Yohei Sawada¹ (1.The University of Tokyo)

Keywords:ensemble data assimilation, binary observation, urban flooding

In the era of citizen science, a vast number of observations can be collected from Social Network Services (SNSs) and Internet of Things (IoT) devices. For instance, in urban flood modeling and monitoring, numerous Closed-Circuit TeleVision (CCTV) cameras and eyewitness reports of flooding from SNSs have the potential to enhance numerical simulations of complex water flow in urbanized areas. However, these observations are often binary (e.g., flooding or no flooding) and difficult to integrate into numerical simulation using conventional data assimilation methods for high-dimensional problems, such as Ensemble Kalman Filter (EnKF). Here we present a new ensemble data assimilation method which can effectively assimilate binary observations into models. We show that the discontinuity of likelihood functions for binary observation can be effectively addressed by introducing observation errors that follow a logistic distribution. With this modification, the posterior distribution can be approximated by Maximum Likelihood Ensemble Filter (MLEF). We applied the proposed data assimilation method to the Storm Water Management Model (SWMM), which is widely used in the urban flooding research community, and demonstrated that binary water level observation can be stably and effectively integrated into SWMM.