Modeling the 2D Inundation Simulation Based on the ANN-Derived Model with Real-Time Measurements at Roadside IoT Sensors

Abstract

This study aims to develop a smart model for the two-dimensional (2D) inundation simulation based on the derived artificial neural network (ANN) model with real-time measurements at the roadside IoT (Internet of Things) sensors; in detail, the flooding zones and associated area can be quantified by combining the inundation-depth estimates at the ungauged locations (defined by the virtual IoT sensor, VIOT) via the corresponding inundation-estimation equations, established using the ANN-derived model with the measurements at the IoT sensors (named SM_EID_VIOT model). Moreover, the resulting inundation-depth estimates at the ungauged locations from the proposed SM_EID_VIOT model can be improved by means of the real-time error-correction approach for the 2D inundation simulation. To demonstrate the reliability of the results from the proposed SM_EID_VIOT model, 1000 simulations of the rainfall-induced flood events within the study area of the Miaoli City of Northern Taiwan are generated as the model-training and validation datasets. Consequently, the proposed SM_EID_VIOT could estimate the inundation depths with an acceptable accuracy at the ungauged locations in time and space based on a low root mean square error (RMSE) of under 0.01 m and a high coefficient of determination (R2) of over 0.8; and it also can delineate the flooding zone to quantify the corresponding area in high reliability in terms of the precision ratio of about 0.7.