Scaling Terrain-Aware Spatial Machine Learning for Flood Mapping on Large Scale Earth Imagery Data

Flood extent mapping plays a crucial role in disaster management and national water forecasting. In recent years, high-resolution optical imagery becomes increasingly available with the deployment of numerous small satellites and drones. However, analysing such imagery data to extract flood extent poses unique challenges due to the rich noise and shadows, obstacles (e.g. tree canopies, clouds), and spectral confusion between pixel classes (flood, dry) due to spatial heterogeneity. Existing machine-learning techniques often focus on spectral and spatial features from raster images without fully incorporating the geographic terrain within classification models. In contrast, we recently proposed a novel machine-learning model called geographical hidden Markov tree (HMT) that integrates spectral features of pixels and topographic constraint from Digital Elevation Model (DEM) data (i.e. water flow directions) in a holistic manner. This paper evaluates the model through case studies on high-resolution aerial imagery from National Oceanic and Atmospheric Administration (NOAA) National Geodetic Survey (NGS) together with DEM. Three scenes are selected in heavily vegetated floodplains near the cities of Grimesland and Kinston in North Carolina during Hurricane Matthew floods in 2016. Results show that the proposed HMT model outperforms several state of the art machine-learning algorithms (e.g. random forests, gradient-boosted model) by an improvement of F-score (the harmonic mean of the user’s accuracy and producer’s accuracy) from around 70% to 80% to over 95% on our datasets.

Remote sensing of discharge and river stage from space provides us with a promising alternative approach to monitor watersheds, no matter if they are ungauged, poorly gauged, or fully gauged. One approach is to estimate river stage from satellite measured inundation area based on the inundation area – river stage relationship (IARSR). However, this approach is not easy to implement because of a lack of data for constructing the IARSR. In this study, an innovative and robust approach to construct the IARSR from digital elevation model (DEM) data was developed and tested. It was shown that the constructed IARSR from DEM data could be used to retrieve water level or river stage from satellite‐measured inundation area. To reduce the uncertainty in the estimated inundation area, a dual‐thresholding method was proposed. The first threshold is the lower limit of pixel value for classifying water body pixels with a relatively high‐level certainty. The second threshold is the upper limit of pixel value for classifying potentially flooded pixels. All pixels with values between the first threshold and the second threshold and adjacent to the classified water body pixels may be partially flooded. A linear interpolation method was used to estimate the wetted area of each partially flooded pixel. In applying the constructed IARSR to the estimated inundation areas from 11 Landsat TM images, 11 water levels were obtained. The root mean square error (RMSE) of the estimated water levels compared with the observed water levels at the US Geological Survey (USGS) gauging station on the Trinity River at Liberty in Liberty County, Texas, is about 0.38 m. Copyright © 2012 John Wiley & Sons, Ltd.

Digital flood-inundation maps for a 4.1-mile reach of the Saddle River from 0.6 miles downstream from the New JerseyNew York State boundary in Upper Saddle River Borough to 0.2 miles downstream from the East Allendale Road bridge in Saddle River Borough, New Jersey, were created by the U.S. Geological Survey (USGS) in cooperation with the New Jersey Department of Environmental Protection (NJDEP). The inundation maps, which can be accessed through the USGS Flood Inundation Mapping Science Web site at http://water. usgs.gov/osw/flood_inundation/, depict estimates of the areal extent and depth of flooding corresponding to select water levels (stages) at the USGS streamgage 01390450, Saddle River at Upper Saddle River, New Jersey. Current conditions for estimating near real-time areas of inundation using USGS streamgage information may be obtained on the Internet at http://waterdata.usgs.gov/nwis/uv?site_no=01390450. The National Weather Service (NWS) forecasts flood hydrographs at many places that are often collocated with USGS streamgages. NWS-forecasted peak-stage information may be used in conjunction with the maps developed in this study to show predicted areas of flood inundation. In this study, flood profiles were computed for the stream reach by means of a one-dimensional step-backwater model. The model was calibrated by using the most current stage-discharge relations (in effect March 2013) at USGS streamgage 01390450, Saddle River at Upper Saddle River, New Jersey, and documented high-water marks from recent floods. The hydraulic model was then used to determine eight water-surface profiles for flood stages at 0.5-foot (ft) intervals referenced to the streamgage datum, North American Vertical Datum of 1988 (NAVD 88), and ranging from bankfull, 0.5 ft below NWS Action Stage, to the upper extent of the stagedischarge rating which is approximately 1 ft higher than the highest recorded water level at the streamgage. Action Stage is the stage which when reached by a rising stream the NWS or a partner needs to take some type of mitigation action in preparation for possible significant hydrologic activity. The simulated water-surface profiles were then combined with a geographic information system 3-meter (9.84 ft) digital elevation model (derived from Light Detection and Ranging (LiDAR) data) in order to delineate the area flooded at each water level. The availability of these maps along with real-time streamflow data and information regarding current stage from USGS streamgages and forecasted stream stages from the NWS provide emergency management personnel and residents with information that is critical for flood response activities, such as evacuations and road closures, as well as for postflood recovery efforts.

A high-resolution (10-meter per pixel) digital elevation model (DEM) was created for the Sacramento-San Joaquin Delta using both bathymetry and topography data. This DEM is the result of collaborative efforts of the U.S. Geological Survey (USGS) and the California Department of Water Resources (DWR). The base of the DEM is from a 10-m DEM released in 2004 and updated in 2005 (Foxgrover and others, 2005) that used Environmental Systems Research Institute(ESRI), ArcGIS Topo to Raster module to interpolate grids from single beam bathymetric surveys collected by DWR, the Army Corp of Engineers (COE), the National Oceanic and Atmospheric Administration (NOAA), and the USGS, into a continuous surface. The Topo to Raster interpolation method was specifically designed to create hydrologically correct DEMs from point, line, and polygon data (Environmental Systems Research Institute, Inc., 2015). Elevation contour lines were digitized based on the single beam point data for control of channel morphology during the interpolation process. Checks were performed to ensure that the interpolated surfaces honored the source bathymetry, and additional contours and(or) point data were added as needed to help constrain the data. The original data were collected in the tidal datum Mean Lower or Low Water (MLLW), or the National Geodetic Vertical Datum of 1929 (NGVD29). All data were converted to NGVD29. The 2005 USGS DEM was updated by DWR, first by converting the DEM to the current modern datum of National Geodetic Vertical Datum of 1988 (NGVD88) and then by following the methodology of the USGS DEM, established for the 2005 DEM (Foxgrover and others, 2005) for adding newly collected single and multibeam bathymetric data. They then included topographic data from lidar surveys, providing the first DEM that included the land/water interface (Wang and Ateljevich, 2012). The USGS further updated and expanded the DWR DEM with the inclusion of USGS interpolated sections of single beam bathymetry data collected by the COE and USGS scientists, expanding the DEM to include the northernmost areas of the Sacramento-San Joaquin Delta, and by making use of a two-meter seamless bathymetric/topographic DEM from the USGS EROS Data Center (2013) of the San Francisco Bay region. The resulting 10-meter USGS DEM encompasses the entirety of Suisun Bay, beginning with the Carquinez Strait in the west, east to California Interstate 5, north following the path of the Yolo Bypass and the Sacramento River up to Knights Landing, and the American River northeast to the Nimbus Dam, and south to areas around Tracy. The DEM incorporates the newest available bathymetry data at the time of release, as well as including, at minimum, a 100-meter band of available topography data adjacent to most shorelines. No data areas within the DEM are areas where no elevation data exists, either due to a gap in the land/water interface, or because lidar was collected over standing water that was then cut out of the DEM.

Using combined SAR image data acquired by the Japanese JERS-1 (L-band, hh-polarization) and the European ERS-1 (C-band, vv-polarization) yields excellent classification results. An important preprocessing step prior to classification is the geometric and radiometric calibration of the image data from each instrument. In order to match and radiometrically calibrate JERS-1 and ERS-1 radar images the geocoding of the scenes using terrain height information is indispensible. Therefore the generation of digital elevation models (DEM) and vector/point data for the search of ground control points is necessary. An invaluable data resource is made available on-line on the Internet through the US Geological Survey (USGS). Using DEM data, the sensor/target geometries are analyzed to derive the local incidence angle and determine layover and shadow regions from the DEM data. Layover and shadow regions need to be determined for both JERS-1 and ERS-1 scenes, and than excluded from the classification procedure. The radiometric information in both scenes needs to be calibrated using area correction terms derived from the local angle of incidence. This paper discusses the generation of DEM's from USGS data the generation of layover/shadow mask and local incidence angle maps, as well as the radiometric correction for ERS-1/JERS-1 SAR data at the University of Michigan.

Continuous topography from Digital Elevation Model (DEM) data is frequently segmented into terrain classes based on local morphological characteristics of terrain elevation, e.g., local slope gradient and convexity. The resulting classes are often used as proxies for the average shear wave velocity up to 30 m, and the determination of ground types as required by the Eurocode (EC8) for computing elastic design spectra. In this work, we investigate the links between terrain related variables, particularly slope gradient, extracted for the area of Greece from the Shuttle Radar Topography Mission (SRTM) 30 arc second global topographic data available from the United States Geological Survey (USGS), with: (a) the global terrain classification product of Iwahashi and Pike (2007) in which 16 terrain types are identified for the same spatial resolution, and (b) information on geological units extracted at the same resolution from the geological map of Greece at a scale of 1/500000 as published from the Institute of Geology and Mineral Exploration (IGME). An interpretation of these links is presented within the context of understanding the reliability of using geology, slope and terrain classes for site characterizations of earthquake risk in a high seismicity area like Greece. Our results indicate that slope is a somewhat biased proxy for solid rocks, whereas in Alluvial deposits the distance to and type of the nearest geological formation appears to provide qualitative information on the size of the sedimentary deposit.

Since hurricane Katrina, developer and planners are looking at a tools and sustainable ways to minimize vulnerabilities in events of disasters especially along the coast. In this setting, remote sensing and Geographic information systems (GIS) have the capacity to provide valuable and timely information about coastal resources, and form an important basis for sustainable planning for land management and decision making. This paper focuses on the design of appropriate coastal databases in six counties in Southern Mississippi using remote sensing and GIS technology. The intent is to enable planners and policy-makers contribute to improved land administration and coastal resources management in order to enhance their competence in decision-making. In order to achieve these objectives, satellite image and digital elevation models (DEMs) data were downloaded from the United States Geological Survey (USGS) seamless data warehouse National Elevation Dataset (NED). From there, the DEM data was co-registered to the satellite image to facilitate draping of the image over the DEM to create terrain models. Result reveals greater part of the three counties along the coast lies less than 10 meters above mean sea level with exposure to coastal disaster vulnerability. In the context of the objectives of the research anchored on the applications of GIS and remote sensing towards efficient land administration and coastal resource management in six coastal counties in an ecologically fragile area already dubbed the epicenter of coastal disasters. Considering the changes that took place in the six counties after Hurricane Katrina debacle, the findings in this study not only stand out, but they remain highly beneficial to decision makers made up of planners and policy makers in several ways. Just as the study injected elements of novelty by identifying digital elevation model information for the six counties in low lying areas, revealing the steeper areas in the north portion of the study stands out as a major finding and study contribution in a way beneficial to decision makers in the region. With that they are now better informed in sharing and cautioning and pinpointing to the public the hidden critical pathways to coastal vulnerability that were previously unknown to ordinary people.

Soil erosion in the agricultural area of a hill slope is a fundamental issue for crop productivity and environmental sustainability. Building terrace is a very popular way to control soil erosion, and accurate assessment of the soil erosion rate is important for sustainable agriculture and environmental management. Currently, many soil erosion estimations are mainly based on the freely available medium or coarse resolution digital elevation model (DEM) data that neglect micro topographic modification of the agriculture terraces. The development of unmanned aerial vehicle (UAV) technology enables the development of high-resolution (centimeter level) DEM to present accurate topographic features. To demonstrate the sensitivity of soil erosion estimates to DEM resolution at this high-resolution level, this study tries to evaluate soil erosion estimation in the Middle Hill agriculture terraces in Nepal based on UAV derived high-resolution (5 × 5 cm) DEM data and make a comparative study for the estimates by using the DEM data aggregated into different spatial resolutions (5 × 5 cm to 10 × 10 m). Firstly, slope gradient, slope length, and topographic factors were calculated at different resolutions. Then, the revised universal soil loss estimation (RUSLE) model was applied to estimate soil erosion rates with the derived LS factor at different resolutions. The results indicated that there was higher change rate in slope gradient, slope length, LS factor, and soil erosion rate when using DEM data with resolution from 5 × 5 cm to 2 × 2 m than using coarser DEM data. A power trend line was effectively used to present the relationship between soil erosion rate and DEM resolution. The findings indicated that soil erosion estimates are highly sensitive to DEM resolution (from 5 × 5 cm to 2 × 2 m), and the changes become relatively stable from 2 × 2 m. The use of DEM data with pixel size larger than 2 × 2 m cannot detect the micro topography. With the insights about the influencing mechanism of DEM resolution on soil erosion estimates, this study provides important suggestions for appropriate DEM data selection that should be investigated first for accurate soil erosion estimation.

High-resolution Digital Elevation Model (DEM) data provides essential information for pluvial flood simulation. Although the increased accessibility and quality of publicly available DEM datasets can facilitate geospatial analysis at various scales, existing DEM datasets with global coverage mostly lack sufficient spatial resolution for pluvial flood simulations, which require detailed topographic information to be included in the simulation. Simulating flood scenarios with low-resolution DEMs (>30m) can result in substantial deviations from real cases. This issue becomes even more severe for flood-prone areas in data-scarce developing countries.Image super-resolution is a technique for reconstructing low-resolution information into high-resolution data. Various deep-learning models have been employed for this task, primarily focusing on generating high-resolution natural-colour images. However, the effects of these deep learning models on enhancing the resolution of DEM data have not been extensively investigated. One of the state-of-the-art super-resolution models, the Residual Channel Attention Network (RCAN), has gained popularity due to its accuracy and efficiency. Leveraging publicly available low-resolution global DEM data and high-resolution regional DEM data, this study assesses the performance of RCAN models in a DEM super-resolution task. The experimental results suggest that, compared to conventional interpolation methods, the tested RCAN model exhibits superior performance in constructing high-resolution DEM data. The generated super-resolution DEM data were then tested in pluvial flood simulations and achieved substantially higher realism in modelling floodwater distribution. The proposed method for constructing super-resolution DEMs opens up the possibility of simulating flooding at hyper-resolution globally.

Digital elevation model (DEM) data describe the information of ground elevation. So it is important to protect the copyright of digital elevation model data. A lossless visible 3-D watermarking algorithm to protect DEM data is proposed in this paper. The copyright watermarking is embedded in the DEM data by an visible way. The original data, blocked by 3-D visible watermarking, are hidden in the watermarked DEM data by a generalized histogram algorithm. Because the visible watermark blocks a part of DEM data, illegal users are restricted to retrieve. At the same time, 3-D visible watermark can identify the copyright. Without original 3-D watermark data, authorized users can eliminate the 3-D visible watermark and restore the original DEM data lossless by applying the proposed algorithm. It is a blind watermarking algorithm. Experiments demonstrate that the proposed algorithm has satisfactory security and can effectively protect the copyright of DEM data.

DEM Explorer: An online interoperable DEM data sharing and analysis system

Digital flood-inundation maps were created by the U.S. Geological Survey (USGS) in cooperation with the U.S. Army Corps of Engineers, New York District for a 25-mile reach of the Ottauquechee River and a 2-mile reach of Reservoir Brook in Vermont. The reach of the Ottauquechee River that was studied extends from River Road Bridge in Killington, Vt., to the Taftsville Dam in the village of Taftsville, in the town of Woodstock, Vt., and the reach of Reservoir Brook extends from a location downstream from the Woodward Reservoir in Plymouth, Vt., to its confluence with the Ottauquechee River in Bridgewater, Vt. The inundation maps depict estimates of the areal extent of flooding corresponding to the 1-percent annual exceedance probability (AEP) flood (also referred to as the 100-year flood) and the peak of the tropical storm Irene flood of August 28, 2011, which was greater than the 0.2-percent AEP flood (also referred to as the 500-year flood), as referenced to the USGS Ottauquechee River near West Bridgewater, Vt. streamgage (station 01150900). In addition to the two digital flood inundation maps, flood profiles were created that depict the study reach flood elevation of tropical storm Irene of August 2011 and the 10-, 2-, 1-, and 0.2-percent AEP floods, also known as the 10-, 50-, 100-, and 500-year floods, respectively. The 10-, 2-, 1-, and 0.2-percent AEP flood discharges were determined using annual peak flow data from the USGS Ottauquechee River near West Bridgewater, Vt. streamgage (station 01150900). Flood profiles were computed for the Ottauquechee River and Reservoir Brook by means of a one-dimensional stepbackwater model. The model was calibrated using documented high-water marks of the peak of the tropical storm Irene flood of August 2011 as well as stage discharge data as determined for USGS Ottauquechee River near West Bridgewater, Vt. streamgage (station 01150900). The simulated water-surface profiles were combined with a digital elevation model within a geographic information system to delineate the areas flooded during tropical storm Irene and for the 1-percent AEP water-surface profile. The digital elevation model data were derived from light detection and ranging (lidar) data obtained for a 3,281-foot (1,000-meter) corridor along the Ottauquechee River study reach and were augmented with 33-foot (10meter) contour interval data in the modeled flood-inundation areas outside the lidar corridor. The 33-foot (10-meter) contour interval USGS 15-minute quadrangle topographic digital raster graphics map used to augment lidar data was produced at a scale of 1:24,000. The digital flood inundation maps and flood profiles along with information regarding current stage from USGS streamgages on the Internet provide emergency management personnel and residents with information that is critical for flood response activities, such as evacuations and road closures, as well as for post-flood recovery efforts.

The U.S. Geological Survey (USGS), in cooperation with the Indiana Office of Community and Rural Affairs, created digital flood-inundation maps for an 8.3-mile reach of the Elkhart River at Goshen, Indiana, extending from downstream of the Goshen Dam to downstream from County Road 17. The inundation maps, which can be accessed through the USGS Flood Inundation Mapping Science Web site at http://water. usgs.gov/osw/flood_inundation/, depict estimates of the areal extent and depth of flooding corresponding to nine selected water levels (stages) at the USGS streamgage at Elkhart River at Goshen (station number 04100500). Current conditions for the USGS streamgages in Indiana may be obtained on the Internet at http://waterdata.usgs.gov/. In addition, streamstage data have been provided to the National Weather Service (NWS) for incorporation into their Advanced Hydrologic Prediction Service (AHPS) flood-warning system (http://water. weather.gov/ahps/). The NWS forecasts flood hydrographs at many places that are often collocated with USGS streamgages. NWS-forecasted peak-stage information may be used in conjunction with the maps developed in this study to show predicted areas of flood inundation. In this study, flood profiles were computed for the stream reach by means of a one-dimensional step-backwater model. The model was calibrated using the most current stagedischarge relation at the Elkhart River at Goshen streamgage. The hydraulic model was then used to compute nine watersurface profiles for flood stages at 1-foot (ft) intervals referenced to the streamgage datum and ranging from approximately bankfull (5 ft) to greater than the highest recorded water level (13 ft). The simulated water-surface profiles were then combined with a geographic information system (GIS) digital-elevation model (DEM), derived from Light Detection and Ranging (LiDAR) data having a 0.37-ft vertical accuracy and 3.9-ft horizontal resolution in order to delineate the area flooded at each water level. The availability of these maps, along with Internet information regarding current stage from USGS streamgages and forecasted stream stages from the NWS, provide emergency management personnel and residents with information that is critical for flood response activities such as evacuations and road closures as well as for postflood recovery efforts.

Digital flood-inundation maps for a 3.2-mile reach of North Fork Salt Creek at Nashville, Indiana, were created by the U.S. Geological Survey (USGS) in cooperation with the Indiana Department of Transportation. The flood-inundation maps, which can be accessed through the USGS Flood Inundation Mapping Science Web site at http://water.usgs.gov/osw/flood_inundation/, depict estimates of the areal extent and depth of flooding corresponding to selected water levels (stages) at the USGS streamgage 03371650, North Fork Salt Creek at Nashville, Ind. Real-time stages at this streamgage may be obtained on the Internet from the USGS National Water Information System at http://waterdata.usgs.gov/nwis or the National Weather Service (NWS) Advanced Hydrologic Prediction Service at http:/water.weather.gov/ahps/ (NWS site NFSI3). Flood profiles were computed for the stream reach by means of a one-dimensional, step-backwater hydraulic modeling software developed by the U.S. Army Corps of Engineers. The hydraulic model was calibrated using the current stage-discharge rating at the USGS streamgage 03371650, North Fork Salt Creek at Nashville, Ind. The hydraulic model was then used to compute 12 water-surface profiles for flood stages at 1-foot (ft) intervals, except for the highest profile of 22.9 ft, referenced to the streamgage datum ranging from 12.0 ft (the NWS action stage ) to 22.9 ft, which is the highest stage of the current USGS stage-discharge rating curve and 1.9 ft higher than the NWS major flood stage. The simulated water-surface profiles were then combined with a Geographic Information System digital elevation model (derived from light detection and ranging [lidar] data having a 0.98-foot vertical accuracy and 4.9-foot horizontal resolution) to delineate the area flooded at each stage. The availability of these maps, along with information regarding current stage from the USGS streamgage, will provide emergency management personnel and residents with information that is critical for flood response activities such as evacuations and road closures, as well as for post-flood recovery efforts.

ABSTRACTThe calculation of surface area is meaningful for a variety of space-filling phenomena, e.g., the packing of plants or animals within an area of land. With Digital Elevation Model (DEM) data we can calculate the surface area by using a continuous surface model, such as by the Triangulated Irregular Network (TIN). However, just as the triangle-based surface area discussed in this paper, the surface area is generally biased because it is a nonlinear mapping about the DEM data which contain measurement errors. To reduce the bias in the surface area, we propose a second-order bias correction by applying nonlinear error propagation to the triangle-based surface area. This process reveals that the random errors in the DEM data result in a bias in the triangle-based surface area while the systematic errors in the DEM data can be reduced by using the height differences. The bias is theoretically given by a probability integral which can be approximated by numerical approaches including the numerical integral and the Monte Carlo method; but these approaches need a theoretical distribution assumption about the DEM measurement errors, and have a very high computational cost. In most cases, we only have variance information on the measurement errors; thus, a bias estimation based on nonlinear error propagation is proposed. Based on the second-order bias estimation proposed, the variance of the surface area can be improved immediately by removing the bias from the original variance estimation. The main results are verified by the Monte Carlo method and by the numerical integral. They show that an unbiased surface area can be obtained by removing the proposed bias estimation from the triangle-based surface area originally calculated from the DEM data.