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

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.

A deep learning technique-based data-driven model for accurate and rapid flood prediction

An accurate and rapid urban flood prediction model is essential to support decision-making on flood management, especially under increasing extreme precipitation conditions driven by climate change and urbanization. This study developed a deep learning technique-based data-driven flood prediction model based on an integration of LSTM network and Bayesian optimization. A case study in north China was applied to test the model performance and the results clearly showed that the model can accurately predict flood maps for various hyetograph inputs, meanwhile with substantial improvements in computation time. The model predicted flood maps 19,585 times faster than the physical-based hydrodynamic model and achieved a mean relative error of 9.5 %. For retrieving the spatial patterns of water depths, the degree of similarity of the flood maps was very high. In a best case, the difference between the ground truth and model prediction was only 0.76 % and the spatial distributions of inundated paths and areas were almost identical. The proposed model showed a robust generalizability and high computational efficiency, and can potentially replace and/or complement the conventional hydrodynamic model for urban flood assessment and management, particularly in applications of real time control, optimization and emergency design and plan.

Validation and Projections of Climate Characteristics in the Saginaw Bay Watershed, MI, for Hydrologic Modeling Applications

Fundamental differences in the nature of climate and hydrologic models make coupling of future climate projections to models of watershed hydrology challenging. This study uses the NCAR Weather Research and Forecast model (WRF) to dynamically downscale climate simulations over the Saginaw Bay Watershed, MI and prepare the results for input into semi-distributed hydrologic models. One realization of the bias-corrected NCAR CESM1 model's RCP 8.5 climate scenario is dynamically downscaled at a spatial resolution of 3 km by 3 km for the end of the twenty-first century and validated based on a downscaled run for the end of the twentieth century in comparison to ASOS and NWS COOP stations. Bias-correction is conducted using Quantile Mapping to correct daily maximum and minimum temperature, precipitation, and relative humidity for use in future hydrologic model experiments. In the Saginaw Bay Watershed the end of the twenty-first century is projected to see maximum and minimum average daily temperatures warming by 5.7 and 6.3°C respectively. Precipitation characteristics over the watershed show an increase in mean annual precipitation (average of +14.3 mm over the watershed), mainly due to increases in precipitation intensity (average of +0.3 mm per precipitation day) despite a decrease in frequency of −10.7 days per year. The projected changes have substantial implications for watershed processes including flood prediction, erosion, mobilization of non-point source and legacy contaminants, and evapotranspirative demand, among others. We present these results in the context of usefulness of the downscaled and bias corrected data for semi-distributed hydrologic modeling.

Simulation of karst floods with a hydrological model improved by meteorological model coupling

Karst basins are prone to rapid flooding because of their geomorphic complexity and exposed karst landforms with low infiltration rates. Accordingly, simulating and forecasting floods in karst regions can provide important technical support for local flood control. The study area, the Liujiang karst river basin, is the most well-developed karst area in South China, and its many mountainous areas lack rainfall gauges, limiting the availability of precipitation information. Quantitative precipitation forecast (QPF) from the Weather Research and Forecasting model (WRF) and quantitative precipitation estimation (QPE) from remote sensing information by an artificial neural network cloud classification system (PERSIANN-CCS) can offer reliable precipitation estimates. Here, the distributed Karst-Liuxihe (KL) model was successfully developed from the terrestrial Liuxihe model, as reflected in improvements to its underground structure and confluence algorithm. Compared with other karst distributed models, the KL model has a relatively simple structure and small modeling data requirements, which are advantageous for flood prediction in karst areas lacking hydrogeological data. Our flood process simulation results suggested that the KL model agrees well with observations and outperforms the Liuxihe model. The average Nash coefficient, correlation coefficient, and water balance coefficient increased by 0.24, 0.19, and 0.20, respectively, and the average flood process error, flood peak error, and peak time error decreased by 13%, 11%, and 2 hours, respectively. Coupling the WRF model and PERSIANN-CCS with the KL model yielded a good performance in karst flood simulation and prediction. Notably, coupling the WRF and KL models effectively predicted the karst flood processes and provided flood prediction results with a lead time of 96 hours, which is important for flood warning and control.

Flood Discharge Prediction Based on Remote-Sensed Spatiotemporal Features Fusion and Graph Attention

Floods have brought a great threat to the life and property of human beings. Under the premise of strengthening flood control engineering measures and following the strategic thinking of sustainable development, many achievements have been made in flood forecasting recently. However, due to the complexity of the traditional lumped model and distributed model, the hydrologic parameter calibration process is full of difficulties, leading to a long development cycle of a reasonable hydrologic prediction model. Even for modern data-driven models, the spatial distribution characteristics of the rainfall data are also not fully mined. Based on this situation, this paper abstracts the rainfall data into the graph structure data, uses remote sensing images to extract the elevation information, introduces the graph attention mechanism to extract the spatial characteristics of rainfall, and employs long-term and short-term memory (LSTM) network to fuse the spatial and temporal characteristics for flood prediction. Through well-designed experiments, the forecasting effect of flood peak value and flood arrival time is verified. Furthermore, compared with the LSTM model and BIGRU model without spatial feature extraction, the advantages of spatiotemporal feature fusion are highlighted. The specific performance is that the RMSE (the root means square error) and R2(coefficient of determination) of the GA-RNN model have been significantly improved. Finally, we conduct experiments on the observed ten rainfall events in the history of the target watershed. According to the hydrological prediction specifications, the model can be evaluated as a Class B flood forecasting model.

Flood Forecasting in Large River Basins Using FOSS Tool and HPC

The Indian subcontinent is annually affected by floods that cause profound irreversible damage to crops and livelihoods. With increased incidences of floods and their related catastrophes, the design, development, and deployment of an Early Warning System for Flood Prediction (EWS-FP) for the river basins of India is needed, along with timely dissemination of flood-related information for mitigation of disaster impacts. Accurately drafted and disseminated early warnings/advisories may significantly reduce economic losses incurred due to floods. This study describes the design and development of an EWS-FP using advanced computational tools/methods, viz. HPC, remote sensing, GIS technologies, and open-source tools for the Mahanadi River Basin of India. The flood prediction is based on a robust 2D hydrodynamic model, which solves shallow water equations using the finite volume method. The model is open-source, supports geographic file formats, and is capable of simulating rainfall run-off, river routing, and tidal forcing, simultaneously. The model was tested for a part of the Mahanadi River Basin (Mahanadi Delta, 9225 sq km) with actual and predicted discharge, rainfall, and tide data. The simulated flood inundation spread and stage were compared with SAR data and CWC Observed Gauge data, respectively. The system shows good accuracy and better lead time suitable for flood forecasting in near-real-time.

Identification of Flood-prone Areas using HEC-HMS and HEC-RAS, the Case of Ciberang River Basin, Lebak District of Banten Province

It is common practice that flood hydrograph simulations help to provide better flood prediction and flood damage reduction planning. These efforts require information on flood-prone areas identification from the hydrological and hydraulic analysis results. Historically, the Ciberang River Basin has experienced floods. Those floods cause the loss of human life and damage some houses along the river’s channels, especially in Lebak District, Banten Province, Indonesia. The main objective of this study is to identify flood-prone areas based on the simulation result of a hydrologic and hydraulic model of catchment response due to several extreme rainfall events using HEC-HMS and HEC-RAS software. Rainfall and discharge data measured at the Ciberang-Sabagi water level gauge on 10 January 2013 were used to calibrate hydrological watershed parameters. The hydraulics channel routing is started from the planned location of the Sabo dam to the downstream control point. The next stage was the simulation of rainfall-runoff transformation and 1D unsteady flow channel routing for the 2, 5, and 10-years floods return periods. The main result of this study is a flood hazards map that shows the spatial distribution of the area and inundation depth for each return period of the flood.

An Integrated Framework of GRU Based on Improved Whale Optimization Algorithm for Flood Prediction

Accurate prediction of floods is the first step in formulating flood control strategies and reducing flood disasters. This research proposes a deep learning model based on Gate Recurrent Unit (GRU), Random Forest Algorithm (RF), Whale Optimization Algorithm (WOA) and Optimal Variational Mode Decomposition (OVMD) for flood prediction. First, the random historical time series is decomposed using OVMD. Secondly, combined with the RF feature importance measurement, select features with high importance to obtain the optimal input set. Third, use the GRU model to predict all sub-models, and use the WOA algorithm to optimize the hyperparameters in the GRU model. This study also proposes a hybrid strategy to improve the traditional WOA algorithm and enhance the optimization ability of the WOA algorithm. Finally, the prediction results of all sub-modes were aggregated to generate the final prediction result. The model was validated using data from three hydrological stations in the upper, middle and lower reaches of the Minjiang river basin in China. Through the results of the experiment, it can be seen that the proposed prediction model can effectively predict the flood time series, and has better accuracy than other models.