Hydro-Flattened Digital Elevation Models (DEMs)

                                The DEMs are derived from the source lidar and 3D breaklines created from the lidar.  Horizontal accuracy is not performed on the DEMs or breaklines.

Only checkpoints photo-identifiable in the intensity imagery can be used to test the horizontal accuracy of the lidar.  Photo-identifiable checkpoints in intensity imagery typically include checkpoints located at the ends of paint stripes on concrete or asphalt surfaces or checkpoints located at 90 degree corners of different reflectivity, e.g. a sidewalk corner adjoining a grass surface.  The xy coordinates of checkpoints, as defined in the intensity imagery, are compared to surveyed xy coordinates for each photo-identifiable checkpoint.  These differences are used to compute the tested horizontal accuracy of the lidar.  As not all projects contain photo-identifiable checkpoints, the horizontal accuracy of the lidar cannot always be tested.
                              

                                The DEMs are derived from the source lidar and 3D breaklines created from the lidar.  The DEMs are created using controlled and tested methods to limit the amount of error introduced during DEM production so that any differences identified between the source lidar and final DEMs can be attributed to interpolation differences.  DEMs are created by averaging several lidar points within each pixel which may result in slightly different elevation values at a given location when compared to the source LAS, which is tested by comparing survey checkpoints to a triangulated irregular network (TIN) that is created from the lidar ground points.  TINs do not average several lidar points together but interpolate (linearly) between two or three points to derive an elevation value.

The vertical accuracy of the final bare earth DEMs was tested by Dewberry with 36 independent checkpoints. The same checkpoints that were used to test the source lidar data were used to validate the vertical accuracy of the final DEM products.  The survey checkpoints are evenly distributed throughout the project area and are located in areas of non-vegetated terrain (26 checkpoints), including bare earth, open terrain, and urban terrain, and vegetated terrain (10 checkpoints), including forest, brush, tall weeds, crops, and high grass.  The vertical accuracy is tested by extracting the elevation of the pixel that contains the x/y coordinates of the checkpoint and comparing these DEM elevations to the surveyed elevations.

All checkpoints located in non-vegetated terrain were used to compute the Non-vegetated Vertical Accuracy (NVA).  Project specifications required a NVA of 0.64 ft (19.6 cm) at the 95% confidence level based on RMSEz (0.33 ft/10 cm) x 1.9600.  All checkpoints located in vegetated terrain were used to compute the Vegetated Vertical Accuracy (VVA).  Project specifications required a VVA of 0.96 ft (29.4 cm) based on the 95th percentile.
                              

                                Data for St. Johns River Water Management District QL2 Lidar Project was acquired by Digital Aerial Solutions LLC.

The project area included approximately 185.5 contiguous square miles or 480.44 square kilometers for the City of Palm Coast in Florida.
Lidar sensor data were collected with the Leica ALS80 HP lidar system.  The data was delivered in NAD83(2011) State Plane Florida East, U.S. Survey Feet and NAVD88 (Geoid12B), U.S. Survey Feet. Deliverables for the project included a raw (unclassified) calibrated lidar point cloud, survey control, and a final acquisition/calibration report.

The calibration process considered all errors inherent with the equipment including errors in GPS, IMU, and sensor specific parameters. Adjustments were made to achieve a flight line to flight line data match (relative calibration) and subsequently adjusted to control for absolute accuracy. Process steps to achieve this are as follows:
Rigorous lidar calibration: all sources of error such as the sensor's ranging and torsion parameters, atmospheric variables, GPS conditions, and IMU offsets were analyzed and removed to the highest level possible. This method addresses all errors, both vertical and horizontal in nature. Ranging, atmospheric variables, and GPS conditions affect the vertical position of the surface, whereas IMU offsets and torsion parameters affect the data horizontally. The horizontal accuracy is proven through repeatability: when the position of features remains constant no matter what direction the plane was flying and no matter where the feature is positioned within the swath, relative horizontal accuracy is achieved.
Absolute horizontal accuracy is achieved through the use of differential GPS with base lines shorter than 25 miles. The base station is set at a temporary monument that is 'tied-in' to the CORS network. The same position is used for every lift, ensuring that any errors in its position will affect all data equally and can therefore be removed equally.

Vertical accuracy is achieved through the adjustment to ground control survey points within the finished product. Although the base station has absolute vertical accuracy, adjustments to sensor parameters introduces vertical error that must be normalized in the final (mean) adjustment.

The withheld and overlap bits are set and all headers, appropriate point data records, and variable length records, including spatial reference information, are updated in GeoCue software and then verified using proprietary Dewberry tools.
                              

                                Dewberry utilizes a variety of software suites for inventory management, classification, and data processing. All lidar related processes begin by importing the data into the GeoCue task management software.  The swath data is tiled according to project specifications (5,000 ft x 5,000 ft). The tiled data is then opened in Terrascan where Dewberry identifies edge of flight line points that may be geometrically unusable with the withheld bit.  These points are separated from the main point cloud so that they are not used in the ground algorithms.  Overage points are then identified with the overlap bit.  Dewberry then uses proprietary ground classification routines to remove any non-ground points and generate an accurate ground surface. The ground routine consists of three main parameters (building size, iteration angle, and iteration distance); by adjusting these parameters and running several iterations of this routine an initial ground surface is developed. The building size parameter sets a roaming window size. Each tile is loaded with neighboring points from adjacent tiles and the routine classifies the data section by section based on this roaming window size. The second most important parameter is the maximum terrain angle, which sets the highest allowed terrain angle within the model. As part of the ground routine, low noise points are classified to class 7 and high noise points are classified to class 18.  Once the ground routine has been completed, bridge decks are classified to class 17 using bridge breaklines compiled by Dewberry.  A manual quality control routine is then performed using hillshades, cross-sections, and profiles within the Terrasolid software suite. After this QC step, a peer review is performed on all tiles and a supervisor manual inspection is completed on a percentage of the classified tiles based on the project size and variability of the terrain. After the ground classification and bridge deck corrections are completed, the dataset is processed through a water classification routine that utilizes breaklines compiled by Dewberry to automatically classify hydrographic features.  The water classification routine selects ground points within the breakline polygons and automatically classifies them as class 9, water. During this water classification routine, points that are within 1x NPS or less of the hydrographic features are moved to class 10, an ignored ground due to breakline proximity.  A final QC is performed on the data.  All headers, appropriate point data records, and variable length records, including spatial reference information, are updated in GeoCue software and then verified using proprietary Dewberry tools.


The data was classified as follows:
Class 1 = Unclassified.  This class includes vegetation, buildings, noise etc.
Class 2 = Ground 
Class 7 = Low Noise
Class 9 = Water
Class 10 = Ignored Ground due to breakline proximity
Class 17 = Bridge Decks
Class 18 = High Noise

The LAS header information was verified to contain the following:
Class (Integer)
Adjusted GPS Time (0.0001 seconds)
Easting (0.003 m)
Northing (0.003 m)
Elevation (0.003 m)
Echo Number (Integer)
Echo (Integer)
Intensity (16 bit integer)
Flight Line (Integer)
Scan Angle (degree)
                              

                                Dewberry used GeoCue software to produce intensity imagery and raster stereo models from the source lidar for use in lidargrammetry techniques.  Dewberry then produced full point cloud intensity imagery, bare earth ground models, density models, and slope models.  These files were ingested into eCognition software, segmented into polygons, and training samples were created to identify water.  eCognition used the training samples and defined parameters to identify water segments throughout the project area.  Water segments were then reviewed for completeness, separated into project defined feature classes, merged, and smoothed.  Elevations derived from a bare earth lidar terrain were applied to each feature for 3D attribution. 

The delineation of lakes and ponds and tidal waters, or other water bodies at a constant elevation, was achieved using eCognition software. Lidargrammetry was used to monotonically collect streams and rivers, or features that have gradient 3D elevations.  All breaklines were collected according to specifications for the project.
                              

                                Dewberry digitzed 2D bridge deck polygons from the intensity imagery and used these polygons to classify bridge deck points in the LAS to class 17.  As some bridges are hard to identify in intensity imagery, Dewberry then used ESRI software to generate bare earth elevation rasters.  Bare earth elevation rasters do not contain bridges.  As bridges are removed from bare earth DEMs but DEMs are continuous surfaces, the area between bridge abutments must be interpolated.  The rasters are reviewed to ensure all locations where the interpolation in a DEM indicates a bridge have been collected in the 2D bridge deck polygons.
                              

                                The bridge deck polygons are loaded into Terrascan software.  Lidar points and surface models created from ground lidar points are reviewed and 3D bridge breaklines are compiled in Terrascan.  Typically, two breaklines are compiled for each bridge deck-one breakline along the ground of each abutment.  The bridge breaklines are placed perpendicular to the bridge deck and extend just beyond the extents of the bridge deck.  Extending the bridge breaklines beyond the extent of the bridge deck allows the compiler to use ground elevations from the ground lidar data for each endpoint of the breakline.
                              

                                Breaklines are reviewed against lidar intensity imagery to verify completeness of capture.  All breaklines are then compared to ESRI terrains created from ground only points prior to water classification.  The horizontal placement of breaklines is compared to terrain features and the breakline elevations are compared to lidar elevations to ensure all breaklines match the lidar within acceptable tolerances.  Some deviation is expected between hydrographic breakline and lidar elevations due to monotonicity, connectivity, and flattening rules that are enforced on the hydrographic breaklines.  Once completeness, horizontal placement, and vertical variance is reviewed, all breaklines are reviewed for topological consistency and data integrity using a combination of ESRI Data Reviewer tools and proprietary tools.  Corrections are performed within the QC workflow and re-validated.
                              

                                Class 2, ground, and Class 8, model key points, lidar points are exported from the LAS files into an Arc Geodatabase (GDB) in multipoint format.  The 3D breaklines,  Inland Lakes and Ponds and Streams and Tidal are imported into the same GDB.  An ESRI Terrain is generated from these inputs.  The surface type of each input is as follows:

Ground Multipoint:  Masspoints
Inland Lakes and Ponds:  Hard Replace
Streams and Rivers : Hard Line
Tidal : Hard Replace
                              

                                The ESRI Terrain is converted to a raster.  The raster is created using linear interpolation with a 2.5 foot cell size.  The DEM is reviewed with hillshades in both ArcGIS and Global Mapper.  Hillshades allow the analyst to view the DEMs in 3D and to more efficiently locate and identify potential issues.  Analysts review the DEM for missed lidar classification issues, incorrect  breakline elevations, incorrect hydro-flattening, and artifacts that are introduced during the raster creation process.
                              

                                The corrected and final DEM is clipped to individual tiles.  Dewberry uses a proprietary tool that clips the DEM to each tile located within the final Tile Grid, names the clipped DEM to the Tile Grid Cell name, and verifies that final extents are correct.  All individual tiles are loaded into Global Mapper for the last review.  During this last review, an analsyt checks to ensure full, complete coverage, no issues along tile boundaries, tiles seamlessly edge-match, and that there are no remaining processing artifacts in the dataset.