Flood Mapping with Convolutional Neural Networks Using Spatio-Contextual Pixel Information

A novel change detection and threshold-based ensemble of scenarios pyramid for flood extent mapping using Sentinel-1 data

Flood disasters destroy infrastructure, disrupt ecosystem processes, adversely affect social and economic activities and cause human fatalities. As such, flood extent mapping (FEM) is critical to mitigate these impacts. Specifically, FEM is essential to mitigate adverse impacts through early warning, efficient response during evacuation, search, rescue and recovery. Furthermore, accurate FEM is crucial for policy formulation, planning and management, rehabilitation, and promoting community resilience for sustainable occupation and use of floodplains. Recently, remote sensing has become valuable in flood studies. However, whereas free passive remote sensing images have been common input into predictive models, damage assessment and FEM, their utility is constrained by clouds during flooding events. Conversely, microwave-based data is unconstrained by clouds, hence is important for FEM. Hence, to increase the reliability and accuracy of FEM using Sentinel-1 radar data, we propose a three-step process that builds an ensemble of scenarios pyramid (ESP) based on change detection and thresholding technique. We deployed the ESP technique and tested it on a use-case based on two, five and 10 images. The use-case calculated three co-polarized Vertical-Vertical (VV) and three cross-polarized Vertical-Horizontal (VH) normalized difference flood index scenarios to form six binary classified FEMs at the base. We ensembled the base scenarios to three dual-polarized centre FEMs, and likewise the centre scenarios to a final pinnacle flood extent map. The base, centre and pinnacle scenarios were validated using six binary classification performance metrics. The results show that the ESP increased the base-to-pinnacle minimum classification performance metrics with overall accuracy, Cohen's Kappa, intersect over union, recall, F1-score, and Matthews Correlation coefficient of 93.204%, 0.864, 0.865, 0.870, 0.927, and 0.871 respectively. The study also established that the VV channels were superior in FEM than VH at the ESP base. Overall, this study demonstrates the efficacy of the ESP for operational flood disaster management.

A novel framework for addressing uncertainties in machine learning-based geospatial approaches for flood prediction

Globally, many studies on machine learning (ML)-based flood susceptibility modeling have been carried out in recent years. While majority of those models produce reasonably accurate flood predictions, the outcomes are subject to uncertainty since flood susceptibility models (FSMs) may produce varying spatial predictions. However, there have not been many attempts to address these uncertainties because identifying spatial agreement in flood projections is a complex process. This study presents a framework for reducing spatial disagreement among four standalone and hybridized ML-based FSMs: random forest (RF), k-nearest neighbor (KNN), multilayer perceptron (MLP), and hybridized genetic algorithm-gaussian radial basis function-support vector regression (GA-RBF-SVR). Besides, an optimized model was developed combining the outcomes of those four models. The southwest coastal region of Bangladesh was selected as the case area. A comparable percentage of flood potential area (approximately 60% of the total land areas) was produced by all ML-based models. Despite achieving high prediction accuracy, spatial discrepancy in the model outcomes was observed, with pixel-wise correlation coefficients across different models ranging from 0.62 to 0.91. The optimized model exhibited high prediction accuracy and improved spatial agreement by reducing the number of classification errors. The framework presented in this study might aid in the formulation of risk-based development plans and enhancement of current early warning systems.

To Expedite Roadway Identification and Damage Assessment in LiDAR 3D Imagery for Disaster Relief Public Assistance

Aerial surveys using LiDAR systems can play a vital role in the quantitative assessment of infrastructure damage caused by hurricanes, floods, and other natural disasters. GmAPD LiDAR provides high-resolution 3D point-cloud data which enables the surveyor to take accurate measurements of damages to roads, buildings, communication towers, power lines, etc. Due to the high point cloud density, a very large volume of data is generated during an aerial survey. The data collected during the airborne imaging is post-processed with calibration, geo-registration, and segmentation. Albeit very accurate, extracting useful information from this data is a slow and laborious process. For disaster response, methods of automating this process have spurred the development of simple, fast algorithms that can be used to recognize physical structures from the point-cloud data that can later be assessed for structural damage. In this paper, we describe an efficient algorithm to extract roadways from a massive Lidar data-set to assist the Federal Emergency Management Agency (FEMA) in assessing road conditions as a step toward helping surveyors expedite a quantitative assessment of road damages for providing and distributing public assistance for disaster relief.

Three-Dimensional Inundation Mapping Using UAV Image Segmentation and Digital Surface Model

Flood occurrence is increasing due to the expansion of urbanization and extreme weather like hurricanes; hence, research on methods of inundation monitoring and mapping has increased to reduce the severe impacts of flood disasters. This research studies and compares two methods for inundation depth estimation using UAV images and topographic data. The methods consist of three main stages: (1) extracting flooded areas and create 2D inundation polygons using deep learning; (2) reconstructing 3D water surface using the polygons and topographic data; and (3) deriving a water depth map using the 3D reconstructed water surface and a pre-flood DEM. The two methods are different at reconstructing the 3D water surface (stage 2). The first method uses structure from motion (SfM) for creating a point cloud of the area from overlapping UAV images, and the water polygons resulted from stage 1 is applied for water point cloud classification. While the second method reconstructs the water surface by intersecting the water polygons and a pre-flood DEM created using the pre-flood LiDAR data. We evaluate the proposed methods for inundation depth mapping over the Town of Princeville during a flooding event during Hurricane Matthew. The methods are compared and validated using the USGS gauge water level data acquired during the flood event. The RMSEs for water depth using the SfM method and integrated method based on deep learning and DEM were 0.34m and 0.26m, respectively.

A Quantitative Framework for Analyzing Spatial Dynamics of Flood Events: A Case Study of Super Cyclone Amphan

Identifying the flooding risk hotspot is crucial for aiding a rapid response and prioritizes mitigation efforts over large disaster impacted regions. While climate change is increasing the risk of floods in many vulnerable regions of the world, the commonly used crisis map is inefficient and cannot rapidly determine the spatial variation and intensity of flooding extension across the affected areas. In such cases, the Local Indicators of Spatial Association (LISA) statistic can detect heterogeneity or the flooding hotspot at a local spatial scale beyond routine mapping. This area, however, has not yet been studied in the context of the magnitude of the floods. The present study incorporates the LISA methodology including Moran’s I and Getis–Ord Gi* to identify the spatial and temporal heterogeneity of the occurrence of flooding from super cyclone Amphan across 16 coastal districts of Bangladesh. Using the Synthetic Aperture Radar (SAR) data from Sentinel-1 and a Support Vector Machine (SVM) classification, “water” and “land” were classified for the pre-event (16 May 2020) and post-events (22 May, 28 May, and 7 June 2020) of the area under study. A Modified Normalized Difference Water Index (MNDWI), and visual comparison were used to evaluate the flood maps. A compelling agreement was accomplished between the observed and predicted flood maps, with an overall precision of above 95% for all SAR classified images. As per this study, 2233 km2 (8%) of the region is estimated to have been inundated on 22 May. After this point, the intensity and aerial expansion of flood decreased to 1490 km2 by 28 May before it increased slightly to 1520 km2 (2.1% of the study area) on 7 June. The results from LISA indicated that the main flooding hotspots were located in the central part, particularly in the region off the north-east of the mangrove forest. A total of 238 Unions (smallest administrative units) were identified as high flooding hotspots (p < 0.05) on 22 May, but the number of flooding hotspots dropped to 166 in the second week (28 May) after Amphan subsided before it increased to a further 208 hotspots (p < 0.05) on 7 June due to incessant rainfall and riverbank failure in the south-west part of the study area. As such, an appropriate, timely, and cost-effective strategy would be to assess existing flooding management policies through the identified flooding hotspot regions. This identification would then allow for the creation of an improved policy to help curtail the destructive effects of flooding in the future.

Flood Detection Using Multi-Modal and Multi-Temporal Images: A Comparative Study

Natural disasters such as flooding can severely affect human life and property. To provide rescue through an emergency response team, we need an accurate flooding assessment of the affected area after the event. Traditionally, it requires a lot of human resources to obtain an accurate estimation of a flooded area. In this paper, we compared several traditional machine-learning approaches for flood detection including multi-layer perceptron (MLP), support vector machine (SVM), deep convolutional neural network (DCNN) with recent domain adaptation-based approaches, based on a multi-modal and multi-temporal image dataset. Specifically, we used SPOT-5 and RADAR images from the flood event that occurred in November 2000 in Gloucester, UK. Experimental results show that the domain adaptation-based approach, semi-supervised domain adaptation (SSDA) with 20 labeled data samples, achieved slightly better values of the area under the precision-recall (PR) curve (AUC) of 0.9173 and F1 score of 0.8846 than those by traditional machine approaches. However, SSDA required much less labor for ground-truth labeling and should be recommended in practice.