Using Convolutional Neural Networks to Build a Lightweight Flood Height Prediction Model with Grad-Cam for the Selection of Key Grid Cells in Radar Echo Maps

Abstract

Recent climate change has brought extremely heavy rains and widescale flooding to many areas around the globe. However, previous flood prediction methods usually require a lot of computation to obtain the prediction results and impose a heavy burden on the unit cost of the prediction. This paper proposes the use of a deep learning model (DLM) to overcome these problems. We alleviated the high computational overhead of this approach by developing a novel framework for the construction of lightweight DLMs. The proposed scheme involves training a convolutional neural network (CNN) by using a radar echo map in conjunction with historical flood records at target sites and using Grad-Cam to extract key grid cells from these maps (representing regions with the greatest impact on flooding) for use as inputs in another DLM. Finally, we used real radar echo maps of five locations and the flood heights record to verify the validity of the method proposed in this paper. The experimental results show that our proposed lightweight model can achieve similar or even better prediction accuracy at all locations with only about 5~15% of the operation time and about 30~35% of the memory space of the CNN.

Real-Time Robotic Mirrored Behavior of Facial Expressions and Head Motions Based on Lightweight Networks

“…The reduction in model size after pruning also facilitates the storage and transfer of deep learning models between mobile applications. Generally speaking, although lightweight models can effectively reduce computational costs and deploy on mobile devices for fast and portable detection [18], [19], it also brings some limitations, such as the reduction in accuracy. Our goal in lightweighting the network is to reduce the storage and computation required for model training and usage, while ensuring the accuracy of the neural network as much as possible.…”

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

“…The hydraulic/hydrodynamic models are comprehensively utilized to numerically exhibit the complicated hydrological process and flood dynamics [3] under consideration of the precipitation observations/forecasts. That is, through the hydraulic numerical models, the inundation depths and the potential flooding zones, as well as the associated area, can be estimated in the case of the various types of rainfalls, such as the design rainfall events of the different return periods and the precipitation forecasts as well as observations [2,[4][5][6][7]. Despite the hydraulic numerical models being applied in the flood simulation, their reliability and accuracy might be affected by the uncertainties in the requirement of sufficient observations, the complex model structures, hydrological/hydraulic features, and extensive computation time [3,4,8].…”

“…That is, through the hydraulic numerical models, the inundation depths and the potential flooding zones, as well as the associated area, can be estimated in the case of the various types of rainfalls, such as the design rainfall events of the different return periods and the precipitation forecasts as well as observations [2,[4][5][6][7]. Despite the hydraulic numerical models being applied in the flood simulation, their reliability and accuracy might be affected by the uncertainties in the requirement of sufficient observations, the complex model structures, hydrological/hydraulic features, and extensive computation time [3,4,8]. Recently, the artificial intelligence (AI) models have been comprehensively employed in flood-induced inundation based on machine learning (ML) techniques [9].…”

“…With computation power increasing, the CNN-based and ANN-derived models are widely applied to the relevant hydrological/hydraulic analysis, especially in flood-related simulations/forecasts [1,3,4,[9][10][11][12][13][14]. In detail, the CNN-based model is derived via the neutral network, comprising the convolution, pooling, and fully connected layers, requiring extensive 2D data (i.e., gridded data) as datasets (e.g., the image and videos) for the model training and application [3,4,15]; accordingly, the CNN model can efficiently provide the single model output from the grid-format model inputs. That is to say, the CNN-based model is advantageous concerning the 2D flood simulations for predicting or estimating the spatiotemporal flood-related variates (e.g., the inundation depths and corresponding area) [9].…”

Mitigation Techniques for Water-Induced Natural Disasters: The State of the Art

“…Chen et al [5] proposed using a deep learning model (DLM) to overcome these problems. We alleviated the high computational overhead of this approach by developing a novel framework for the construction of lightweight DLMs.…”