Synergizing Point-Based CCTV and Wide-Area Remote Sensing Intelligence for Adaptive Flood Monitoring in Bandung Bayulodie Vallianto (a,b*), Masahiko Nagai (a), Yusuf Cahyadi (c)
a) Graduate School of Sciences and Technology for Innovation, Yamaguchi University, Japan
b) National Research and Innovation Agency (BRIN), Jakarta, Indonesia
c) Bandung Command Center, Bandung, Indonesia
*bayulodie.val[at]gmail.com
Abstract
Urban flood monitoring remains a challenge for cities where rapid inundation during the rainy season disrupts transportation and risks public safety. This study addresses the critical gap in urban flood monitoring with the aid of combining real-time point-based closed-circuit television (CCTV) streams with wide-area remote sensing intelligence for Bandung City, Indonesia. Our framework continuously processes feeds from 10 flood-prone geotagged CCTV locations, extracting frames every 20 seconds through OpenCV and classifying water severity (dry, wet, or flood) using a pre-trained MobileNetV2 and fine-tuned on 3,058 actual frames that achieve promising performance with 94% classification accuracy in controlled tests. When floods are detected, the pipeline triggers elevation-guided spatial interpolation using a Digital Elevation Model (DEM), modeling flood spread along low-lying roads through Inverse Distance Weighting. This method estimates water surface elevation across a 500-meter radius around each flood point. The interpolated output traces probable inundation extent by comparing water elevations against ground-level DEM values, providing emergency responders with actionable flood spread forecasts. This integration strategically bridges the temporal gain of CCTV (real-time point data) and the spatial intelligence of remote sensing (DEM terrain analysis), overcoming their individual shortcomings. As an ongoing studies initiative, the framework is being refined for operational assessment by means of the Bandung Command Center, with future work focusing on field deployment.
Keywords: Flood severity classification, Geospatial fusion, MobileNetV2, Real-time interpolation, Bandung
