Fusing stacked autoencoder and long short-term memory for regional multistep-ahead flood inundation forecasts

Recurring annual floods creates havoc to the life and property of millions of people around the world. Along with damaging the natural habitat of the living beings, floods terminally affects the growth and development of crops. Accurate flood inundation forecasting plays a crucial role in analysing the risk associated with crop damage due to flooding. Keeping this in view, this study attempts to simulate the flood inundation extent and depth with a daily lead-time of up to 3 days. The Mahanadi River delta in eastern India, which is one of the highly flood prone river deltas in the world is considered as the study area. The rainfall forecasts for the river basin are first bias-corrected using a newly developed bias-correction technique employing copula functions and self-organizing maps. Forcing the hydro-meteorological inputs to a conceptual hydrological model, the discharge forecasts up to 3 days lead-time are obtained. For further improvement, the errors in the discharge forecasts are updated using the state-of-the-art deep learning model, Long-Short Term Memory. Finally, the forecasted inundation depth and extent are simulated by forcing the hydrodynamic 1D-2D MIKE FLOOD model with the improved daily discharge forecasts as the upstream inflow boundary conditions. The hydrological model-simulated discharges after performing error updation are found to be reasonably accurate with a Nash-Sutcliffe Efficiency of >0.90. More than 50% of the observed flood inundated area are found to coincide with the model simulated inundations.

This study aimed at mapping flood inundation areas in the Tana River Delta. The 30 m DEM and digitized streams and road networks data were collected through remote sensing techniques followed by ground truthing. The data used also included Tana River discharge data at Garsen gauging station. Probability distributions were fitted on the annual maximum discharges. The 4 parameter generalized Gamma probability distribution was the best-fitting distribution according to goodness-of-fit criteria and the Q-Q plots and was used to generate the flood quantiles for different return periods. The generated flood quantiles were; 286 m3 /s, 369 m3 /s, 400 m3 /s, 405 m3 /s, 439 m3 /s, and 449 m3 /s for 2-year, 5-year, 10-year, 20-year 50-year, and 100-year return periods respectively. GIS Flood Tool in ArcGIS used flood quantiles to produce flood extents and inundation depths. The largest part of the Tana River delta was predicted to be flooded with inundation depths of upto 1.8 m causing inundation of settlements and leading to the death of people and livestock, as well as the destruction of properties and infrastructure including roads and buildings, farms, crops, among others. The study established that 31% of the schools located along the Tana River in the Garsen sub-catchment are exposed to inundation depths above 0.5 m. Different flood management options were identified for Tana River Delta and were subjected to Multicriteria Analysis for a decision to be arrived at in choosing the best. Advanced technology in recession agriculture, early warning systems, and forecasting were ranked as the optimal options for managing floods in Tana River Delta. The outcomes of this study contribute to preparedness for floods and guide development decisions within the mapped areas.

Problems and prospects in long and medium range weather forecasting

Flooding is a major natural hazard that can cause significant damage to socioeconomic and ecological systems. This study presents an approach to producing the maximum flood inundation and flood duration maps over the Chao Phraya River Basin (CPRB), Thailand. An integrated numerical model and spatial analysis tool were utilized in this study. The Rainfall-Runoff-Inundation (RRI) model was first used to simulate both river discharge and inundation depth. Then, the maximum flood inundation and flood duration maps with different return periods were estimated using a Geographical Information System (GIS) tool. The results illustrate that the flood inundation areas were spread out, starting from Nakhon Sawan Province, which is located in the central part of the basin. The maximum flood inundation depth could reach up to approximately 7.71, 8.28, and 8.78 m for the flood return periods of 50, 100, and 200 years, respectively. The results also indicate that the inundation areas over the CPRB could cover approximately 21,837, 23,392, and 24,533 km2 for flood return periods of 50, 100, and 200 years, respectively. The longest flood durations for return periods of 50, 100, and 200 years were approximately 159, 177, and 198 days, respectively. The longest flood duration occurred in the vicinity of the Nakhon Sawan. This study suggests that flood inundation areas and duration mapping could provide supporting information regarding the impacts caused by varying degrees of flood hazards and can be used to enhance comprehensive disaster risk management planning.

Abstract. River bathymetry is important for accurate flood inundation modelling but is often unavailable due to the time-intensive and expensive nature of its acquisition. This leads to several proposed and implemented approaches for its estimation. However, the errors in estimations inherent in these methods and how they affect the accuracy of the flood inundation modelling outputs, has not been extensively researched. Hence, to contribute, we investigate the sensitivity of flood predictions to the errors in river slope, width, and bank-full flow used in two formulas – the Uniform Flow and the Conceptual Multivariate Regression – for estimating river bathymetry. In this study, we employed a Monte Carlo framework to introduce random errors into these parameters drawn from a normal distribution with zero mean and a standard deviation set to 10 % of their best estimates. Using this process, we generated 50 simulated river bathymetries for each parameter along with an additional 50 where the errors were applied to all parameters simultaneously. The riverbeds generated from these bathymetries were combined with topographic LiDAR data to create model grids. Each grid was used in the hydrodynamic model LISFLOOD-FP to simulate the 2005 flood event in the Waikanae River area of New Zealand. We assessed the resulting flood inundation predictions for their variability and sensitivity. The results indicate that between two methods, the errors in the parameters in the Uniform Flow formula are associated with greater uncertainty in flood inundation depths and extents compared to the Conceptual Multivariate Regression. Among the parameters, the width errors correspond to the highest uncertainty, while the slope errors correspond to the lowest.

The study proposes a methodology for producing probabilistic flood hazard maps based on an ensemble of 2-D hydrodynamic model and frequency analysis of discharge data. With available floodplain topography and limited river bathymetric data, a 2-D hydrodynamic model has been set-up to predict flood inundation area for recently breaching location of the embankment of the river Brahmaputra for a given flood hydrograph. Frequency analysis of the flood waves in nearby gauged records has been conducted and the flood inundation at the breaching location for flood waves of different return period and average monsoonal condition (dry, normal and wet) has been simulated. After that, spatial analysis of the simulated flood inundation depth and maximum velocity is conducted to categorize flood hazard zones. The methodology was applied over a 600 km2 of flood prone area, close to the northern bank of a 64 km reach of the river Brahmaputra (Assam, India). In this study, probabilistic flood hazard mapping approach for embankment breaching at Matamara, Lakhimpur district, Assam was evaluated with the flood inundation obtained from satellite imagery. The spatial variation of simulated flood inundation of 20 years return period closely matches with the observed inundation. The most vulnerable zone within the floodplain area is identified considering random breaching locations. This study demonstrates that probabilistic flood hazard maps in a river valley can be prepared by this approach with limited topographic dataset.

The most frequent natural hazards and the most costly in terms of property damage in study area is Flood. The purpose of this work is use Geographic Information Systems (GIS) Hydrological tools to model and map flood event in Ifo Local Government Area of Ogun state, Nigeria using Shuttle Radar Topographic Mission ( SRTM) data and land use maps as input data layers. In this study, flood inundation depth is calculated with GIS in order to understand the temporal, spatial scale and transition. Flood inundation map and 3D views of the hydrological modelling results are provided for a better understanding of the severity of flooding in the location. It is recommended that the processed data from this study be used as input for future research and the development of flood warning systems for the area. It will also be a base study for future study when rainfall data will be available.

As in the rest of the world, floods have devastating socio-economic effects in Turkey as well. Especially, highly urbanized areas do not enable the implementation of structural measures, properly. In addition, in recent years, structural measures have been replaced by nature-based and eco-friendly approaches. Therefore, it is essential to investigate the effectiveness of nature-based solutions (NBSs) in such places in order to eliminate the negative effects of floods. This study covers the area of Esenyurt District in Istanbul, which is highly urbanized and frequently affected by floods. The main channel of the Haramidere Creek and its six branches were studied with a total length of approximately 25 km. Firstly, 2, 5, 10, 50, 100, 500 and 1000 year recurrence interval of flow rates were calculated by using a hydrological model and extreme value analysis. Next, flood inundation area and depths were determined using 1D and 2D hydrodynamic models. Social and economic risks were estimated by combining the related studies mentioned above with the road, vehicle, building etc. inventory. Following this, the basin was divided into regions according to its NBS characteristics. The approaches such as rainwater harvesting in areas with insufficient infrastructure and permeable pavement in suitable areas having gentle slopes were modeled both individually and in combinations. In the literature, the effects of NBS have been revealed by examining the flood volume and peak flow values. In some similar studies, there is more decrease in peak flow rates while implementing a combination of NBSs rather than applied alone as a solution. In our study, in addition to the changes in flood volume and peak values, changes in flood inundation boundary, depth, social and economic risk will also be quantitatively revealed for the relevant recurrence interval of flow rates.

Prediction concept needs human-like thinking and ideas. ANN helps make a program for describing the solutions to the prediction problems. To gain accurate prediction, there is a need for deep thinking in the knowledge system discovered. Deep learning helps form such next-level type of ANN thinking. Deep learning models do not require humans for programming the problems. It can determine the prediction by itself by learning the dataset. This paper describes the prediction by considering sequential model for continuous accessing of data with the help of multi-layers. Deep learning is very helpful for predicting larger dataset. This paper proposes an algorithmic model known as a Comprehensive Deep Recurrent Artificial Neural Networks (CDRANN) by hybridizing LSTM (long short-term memory) with RNN (recurrent neural network) backtrack solver for better future stock prediction. This paper also explains the need for hybridizing evolutionary model CDRANN to create a novel change in future prediction.KeywordsCDRANN (Comprehensive Deep Recurrent Artificial Neural Networks)LSTM (long short-term memory)RNN (recurrent neural network)Stock marketPrediction

Deep learning (also known as deep structured learning, hierarchical learning, or deep machine learning) is a branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data by using a deep graph with multiple processing layers, composed of multiple linear and nonlinear transformations. Deep learning has been characterized as a class of machine learning algorithms with the following characteristics [257]: • They use a cascade of many layers of nonlinear processing units for feature extraction and transformation. Each successive layer uses the output from the previous layer as input. The algorithms may be supervised or unsupervised and applications include pattern analysis (unsupervised) and classification (supervised). • They are based on the (unsupervised) learning of multiple levels of features or representations of the data. Higher-level features are derived from lower-level features to form a hierarchical representation. • They are part of the broader machine learning field of learning representations of data. • They learn multiple levels of representations that correspond to different levels of abstraction; the levels form a hierarchy of concepts. These definitions have in common: multiple layers of nonlinear processing units and the supervised or unsupervised learning of feature representations in each layer, with the layers forming a hierarchy from low-level to high-level features. Various deep learning architectures such as deep neural networks, convolutional deep neural networks, deep belief networks (DBN), and recurrent neural networks have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition, and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. In this chapter we start in Section 7.1 with an introduction, giving a brief history of this field, the relevant literature, and its applications. Then we study some basic concepts of deep learning such as convolutional neural networks, recurrent neural networks, backpropagation algorithm, restricted Boltzmann machines, and deep learning networks in Section 7.2. Then we illustrate three examples for Apache Spark implementation for mobile big data (MBD), user moving pattern extraction, and combination with nonparametric Bayesian learning, respectively, in Sections 7.3 through 7.5. Finally, we have summary in Section 7.6.

The simulation speed of two‐dimensional hydrodynamic flood models is a limiting factor when catchments are large, a considerable number of simulations is required (e.g., exploratory modeling, Monte‐Carlo flood simulations, or predicting probabilistic flood maps), or when there is a need for real‐time flood emergency management. Rapid Flood Models (RFMs) that rely only on topographic depressions and the water balance equation have been successfully implemented to predict maximum urban flood inundation depths within seconds to a few minutes. However, the preprocessing step (identification of depressions and their attributes) and the postprocessing step (marking up possible flow paths of flood water in between flooded depressions) of RFMs is time consuming. In this study, we developed a new fast flood inundation model based on the cellular automata (CA) approach. The new model does not require the preprocessing and postprocessing steps of RFMs and therefore can provide more simulation speed. The performance of our new model, referred to as Cellular Automata fast flood evaluation (CA‐ffé), was compared to two well‐known hydrodynamic flood models (HEC‐RAS and TUFLOW) in 20 simulation experiments conducted in five different urban subcatchments. CA‐ffé predicted maximum inundation depth with reasonable accuracy in a matter of seconds to a few minutes for a single rainfall event simulation. The CA‐ffé model performed exceptionally well in areas with low‐lying depressions. However, in areas where floodwaters had higher momentum and velocity, the model usually was not able to estimate inundation depths calculated by HEC‐RAS or TUFLOW. CA‐ffé's key drawback is also its inability to represent the temporal evolution of flooding and flow velocities. Nevertheless, its ability to provide spatial flood extents and depths in a fraction of the time compared to its hydrodynamic counterparts is a significant advancement toward exploratory approaches for water systems planning, model‐based predictive control, and real‐time flood management.

<p>Hydrological extremes can be particularly impactful in populated areas where they are modulated by human intervention such as reservoir regulation. However, little is known about how reservoir regulation affects hydrological extremes, particularly at a regional scale. Here, I present a large data set of natural and regulated catchment pairs in the United States and assess how reservoir regulation affects local and regional drought and flood characteristics. I show that (1) reservoir regulation affects drought and flood hazard at a local scale by reducing severity (i.e. intensity/magnitude and deficit/volume) but increasing duration; (2) regulation affects regional hazard by reducing spatial flood connectedness in winter and by increasing spatial drought connectedness in summer; (3) the local alleviation effect is only weakly affected by reservoir purpose for both droughts and floods. I conclude that both local and regional flood and drought characteristics are substantially modulated by reservoir regulation, an aspect that should neither be neglected in hazard nor climate impact assessments.</p>

Do catastrophic floods change the gender division of labor? Panel data evidence from Pakistan

Effects of Dam Decommissioning on Flood Inundation in an Urban Area

Impacts of rainstorm characteristics on flood inundation mitigation performance of LID measures throughout an urban catchment