LIDAR and TDS in GIS: Evaluating LIDAR Data Accuracy Using Ground-Surface Total Station Surveys, Lodore Canyon of the Green River, Dinosaur National Monument, Colorado

Jason S. Alexander

Graduate Student

Department of Aquatic, Watershed, and Earth Resources

Utah State University, Logan, Utah

Introduction and Background

Narrowing and simplification of channel form on the Green River below Flaming Gorge Dam (Figure 1) has been well documented by several authors. Climate change, river regulation and invasive riparian shrubs have all been implicated as agents altering sedimentation patterns and causing degradation to both riparian and aquatic habitats along the Green River. Recent work by Grams and Schmidt (2005) has shown that the Green River is in sediment deficit between the Flaming Gorge Dam and Red Creek, the first major tributary below the dam, but in relative balance or sediment surplus below Red Creek. Simplification of channel form has occurred through the filling of side channels and backwaters. The deposition of inset floodplains within the pre-dam river banks has reduced the channel of the Green River to less than 30% of its pre-dam width. Modern bankfull width of the Green River above the Yampa River reflects that of the annual powerplant flood of 125 m³/s. Six bypass floods have occurred since operations began at Flaming Gorge in 1963 (Figure 2). All but one of these flows occurred in the surplus water years of 1983, 1984, 1986, 1997, and 1999, when Flaming Gorge Reservoir was at full capacity and upstream inflows exceeded powerplant capacity. The establishment of invasive riparian vegetation on the inset floodplains, dominated by Salt Cedar (tamarisk), has precluded mobilization of the banks and subsequent restoration of aquatic and riparian habitat by the bypass floods.

The Green River through Dinosaur National Monument (Figure 1) flows alternately through steep canyons, alluvial parks, and incised meander reaches. The canyon reaches are further distinguished by their rapids-forming debris fans, similar to those of the Grand Canyon of the Colorado River. Termed the “fan-eddy complex” by Schmidt and Rubin (1995), reaches spanning debris fans invariably consist of ponded backwaters, steep rapids or riffles in the constricted waters adjacent to the debris fan, and an expansion reach below the fan. These reaches are the dominant storage sites for fine sediment in the canyon segments of the Green River. As a result, these reaches are also sensitive to changes in sediment regimes. Utah State University Geomorphology Lab (USUGL) established monitoring cross sections every river kilometer on the Green River through Dinosaur National Monument in 1994. Paired experimental reaches were created in 2001 within the fan-eddy reaches of Wade and Curtis, Winnies, Mile 233, and Triplet Falls (Figure 1). Experimental pairs consist of two sub-reaches, a control and a treatment, each spanning the length of a fan-eddy complex. The treatment reaches have had all invasive tamarisk shrubs removed from the floodplain; no riparian treatment has been applied in the control reaches. In May of 2005, the Bureau of Reclamation (BOR), in cooperation with the Upper Basin Recovery Program (UBRP), released bypass flows from Flaming Gorge Dam to meet experimental target flows on the lower Green River for monitoring of aquatic habitat in the Uinta Basin. Because this release was the first flow to exceed powerplant capacity (125 m³/s) since the inception of the experimental reaches in Lodore Canyon, USUGL used the flow as a real-time experiment, monitoring channel form before, during and after the flood.

Channel form in the experimental reaches is monitored through repeat surveys of channel cross-sections and sandbar topography using a total station laser survey instrument (TDS) coupled with sonar data (for subaqueus topography below wadeable depths). Measures of change include channel top width at powerplant capacity, mean channel depth at baseflow (23 m³/second) and powerplant capacity, sand area above baseflow and powerplant capacity, and sandbar volume above baseflow and powerplant capacity (Figures 3, 4, and 5). Volumetric analysis of total alluvial sediments above baseflow or powerplant capacity would be a robust measure of sedimentation patterns (net deposition = narrowing, net scour = restoration) in the experimental reaches, however there is both physical and logistical difficulty in obtaining a relevant number of survey points efficiently for an entire reach.

LIght Detection And Ranging (LIDAR) surveys allow for rapid acquisition of millions of topographic survey points including ground points below forest canopy. The potential for the evaluation of long-term sedimentation patterns in Lodore Canyon through repeat LIDAR flights is great. Common LIDAR accuracies report a root-mean squared error (RMSE) of 15 cm in the vertical and 1-meter in the horizontal. Recent evaluations have shown this accuracy to be achievable only under ideal topographic circumstances (flat, non-vegetated topography and low-altitude flights) (Hodgson and Bresnahan 2004). This degree of spatial error presents significant challenges to volumetric analysis at the reach scale. This study seeks to evaluate the accuracy of a recent LIDAR survey data taken on April 1, 2005 of the Lodore Canyon experimental reaches.

Method of Evaluation

This study compares two survey datasets of the Lodore Canyon experimental reaches: (A) LIDAR data taken on April 1, 2005 and (B) TDS data taken May 13 – 15, 2005. Because TDS data is commonly found to be accurate to less than 2 centimeters in repeat surveys, the TDS data is used in this study as the control dataset. The following steps were taken to evaluate the accuracy of the LIDAR data: (1) Precisely translate and rotate TDS control points from local coordinate systems, into the LIDAR coordinate system (UTM Zone 12, NAD 83) (2) Resolve TDS elevation data for comparison with the LIDAR elevation model (Geoid03 NAVD88) (3) Create a triangulated irregular network (TIN) from LIDAR data points (4) Use EZ Profiler software to create cross section profiles from the TIN (5) Export EZ Profiler TIN data and overlay with TDS profile data in Excel and WinXSPRO software to evaluate spatial and geometrical differences.

Step 1:

The link between the TDS data and the LIDAR data is the presence of LIDAR ground targets installed in some of the experimental reaches (Figure 6). Each ground target consists of a 1.5 x 1.5 m² ground cloth with a precisely surveyed monument pin at the center. Ground targets were located and surveyed into each TDS control network. Database files for the TDS control networks were then brought into ArcGIS as XY events. Using the editor, the XY events were translated by moving the point group as a whole, using the LIDAR control point to “lock in” the ground targets visible on .15 meter resolution aerial orthophotographs (taken during the LIDAR flight) (Figure 7). The LIDAR control point was placed on the center of the “X” of the ground target while zoomed into the four center pixels of the ground target (Figure 8). Using the LIDAR control point as the center, the “rotate” tool was used from the ArcGIS editor to rotate the control points until the “waters edge” points matched up with the waters edge visible in the aerial photographs (Figures 9 and 10). Using the waters edge as a guide for the rotation was reasonable because the discharge on April 1, 2005 was within 5m³/ second of the May 13-15^th TDS survey (stage changes of less than 10 cm). The control points were then cross-checked with existing maps to ensure the cross-section endpoints were in their proper positions. After the TDS survey control points were translated and rotated into place, an empty point feature class was added. This feature class was then edited and points were added by snapping to the TDS XY event themes, creating a TDS control file in UTM Zone 12, NAD 83.

Step 2:

Elevations of the TDS survey data were resolved for comparison with profiles from the LIDAR TIN. Elevations of each LIDAR ground target pin were obtained from Rocky Mountain Surveys, of Vernal, Utah. These elevations were used to convert the local elevations of the TDS survey points into the elevations of the TIN (Geoid03 NAVD88). Rocky Mountain Surveys installed and surveyed the ground control pins using a differentially corrected GPS system (Figure 6). Reported error in elevations varied from 4.0 to 35.0 cm for the three ground targets used. The combined error of the LIDAR survey points and ground target pins yields a maximum error range of 15.0 to 50.0 cm in the vertical dimension between surveys. Each TDS point elevation was adjusted to the TIN elevation by the difference in elevation between the LIDAR ground control pin in the TDS survey and the GPS survey.

Step 3:

A point feature class was created within the LIDAR feature dataset from the LIDAR survey database. Using ArcGIS 3D analyst, a TIN of each of the experimental reaches was created from the LIDAR point feature dataset (reaches without ground targets were excluded) (Figures 11 and 12). Contract specifications for the LIDAR survey demanded an RMSE of no more than .15 meters in the vertical and .5 meters in the horizontal. RMSE values for the LIDAR data were reported as .10 meters in the vertical and .21 meters in the horizontal.

Step 4:

Creation of cross section profiles in ArcGIS with the standard tools presented some difficulties. The first step in this process is to add an empty line feature class. This feature class is then edited and new cross section lines are drawn between cross section endpoints. After all cross section lines are drawn between points, the feature class is converted to 3D using the “convert features to 3D” tool in 3D analyst. Profiles can then be extracted but points from the lines CANNOT be exported with the standard tools. A search of the ESRI support center produced many solutions, mainly Visual Basic codes. However, a GUI module, the EZ Profiler, written by Min-Lang Huang, was found to be the most powerful and user-friendly. The EZ Profiler is an independent extension written for ArcGIS 9.0 and does not need 3D Analyst or Spatial Analyst to perform. EZ Profiler V8.3 can be downloaded at http://arcscripts.esri.com/details.asp?dbid=13688. All that is needed for the creation of a cross section profile with EZ Profiler is a TIN surface. EZ profiler provides many options for creating profiles (Figure 13): (A) the user can draw a profile line, although this does not allow snapping (B) the user can get a profile from an existing polyline (C) the user can rapidly obtain multiple profiles from several polylines and (D) the user can obtain a profile from a 3D polyline. Once an option for a profile is chosen and executed, a shapefile and associated attribute table are created and added to the map (Figure 14). The attribute table for each profile can then be exported to Microsoft Excel with the touch of a button! The profile attributes include a unique line identifier, X, Y, and Z coordinates, and distance from the starting point along the profile (Figure 15). No reduction of the exported data was required for comparison with the resolved TDS cross sections! Each of the profiles extracted for this study were created from line feature classes converted to 3D polylines.

Step 5:

Three parameters were evaluated for each profile comparison: (A) elevation (B) water surface variation and (3) profile shape. For each cross section overlay, the total area between cross sections was calculated in WinXSPRO (Figure 16). Attributes of each profile created in EZ Profiler were exported to Excel and formatted for analysis in WinXSPRO. WinXSPRO is free software developed by the USDS Forest Service, Stream Systems Technology Center for the public domain (download at http://www.stream.fs.fed.us/publications/winxspro.html). WinXSPRO is a cross section analytical tool which allows the user to estimate hydraulic, geometric and sediment transport parameters. Average elevation differences for each profile overlay were calculated by dividing the total incremental area between cross sections by the total incremental horizontal length.

D_avg = (A1 + A2 + Ai….) / (L1 + L2 + Li….)

Where:

D_avg = total average elevation difference between LIDAR and TDS profiles

Ai = total area of i increments

Li = total horizontal length of i increments

Water surface variation was calculated by simply taking the difference between the maximum and minimum elevation of water surface profile points. In order to minimize potential inclusion of ground surface points, water surface datasets did not include waters edge data points. Cross section shape was qualitatively evaluated by visual comparison of profile plots in Excel.

Results and Discussion – Elevation Differences Between LIDAR and TDS Data:

Magnitudes of elevation differences were generally not consistent between reaches or between profiles (Figure 17). Average elevation differences varied from a low of 10cm to a high of 28cm, consistent with or greater than the error reported by the LIDAR contractor (~10cm vertical error). Lower Mile 233 was the only reach to show consistent variation in both profiles. The error reported for the LIDAR control targets (surveyed using differentially corrected GPS) varied from as large as 35.0cm in Winnies reach to as low as 4.0cm in the lower Mile 233 reach. The combined LIDAR and GPS elevational reported error is equivalent or, in some cases, greater than the elevational error found in this study. Ironically, the reach with the least vertical error in the GPS ground control measurement, Lower Mile 233, produced the largest error in elevation between surveys. Nevertheless, errors found within this study are nested closely within the error reported for both the LIDAR and GPS ground control data.

Consistency of error within a reach would suggest error in translation, rotation, or resolution of TDS elevations consistent between cross sections. For example, an erroneous rotation or translation would create LIDAR profiles oblique to the TDS profiles in a generally uniform pattern (i.e. Lower Mile 233). Inconsistent error within a reach would suggest error within the LIDAR data because discrete LIDAR survey points each have inherent error independent of the surrounding points. For example, a TIN created with LIDAR points, each with inconsistent error, should produce profiles whose topography is both above and below that of the TDS survey (i.e. Upper Winnies and Upper Mile 233).

For this study, error in translation, rotation and elevation was minimized though the identification of significant inconsistencies in shape or magnitude of elevation between profiles. These inconsistencies were checked, in some cases many times over, by complete repetition of the analysis. Errors in TDS data were found, resolved and the analysis was performed no less than a third time to minimize humanistic error wherever possible. While the error within this study was found to be consistent with the reported error, the magnitude of elevation differences presents significant challenges to data interchange. For example, a 30.0 cm average elevational difference between the TDS and LIDAR data would suggest significant scour or deposition in environments such as river left of Lower Mile 233 where the combined eddy bar and beach are nearly 40 meters wide (Figure 19). While local stage discharge relationships allow for us to infer what could be changing and at what elevation, significant and non-uniform (i.e. left and right bank) alteration of the cross sections would be necessary to “fit” the data. This makes TDS survey data, a necessary component to couple with any LIDAR floodplain analysis.

Results and Discussion – Water Surface Variation:

Water surface elevation for LIDAR data varied widely and inconsistently within and between reaches (Figure 18). The lowest magnitude variation water surface variation within the LIDAR data was 23.0cm. Maximum water surface variation for LIDAR was 71.0cm. If the interpreted water’s edge were incorporated into the analysis, variance of LIDAR water surface elevation would have exceeded 1 meter in some cases. TDS data was found to have a maximum water surface variation of 4.0cm with the rest of the cross sections showing no more than 2.0cm of variation (some of this variation may be due to ramping discharges from Flaming Gorge).

Error in water surface elevation for LIDAR data was assumed to be related to two factors (1) surface turbulence due to wind and (2) surface turbulence due to wave pulsation. High resolution aerial photographs taken during the LIDAR survey show significant water surface disturbance (in normally placid waters) due to wind. TIN surfaces in all reaches were also generally more horizontal the farther downstream of riffles or constrictions. In all reaches for this study, profiles were in expansion waters or ponded backwaters downstream of riffles and constrictions. In these locations wave pulsation is significant and contributes to significant surface disturbance. The significant error associated with the LIDAR water surface elevation data makes the usefulness of the data limited. For example, a stage change of .5 meters at Lower Mile 233 would result in a decrease or increase of 63 m³/s, a significant difference considering the typical annual peak discharge is 125 m³/s.

Results and Discussion – Cross Section Shape:

LIDAR data generally produced cross section shapes that closely followed those of the TDS survey. In some case, the higher resolution LIDAR data produced superior detail of shape locally. However, large differences in shape were encountered near overhanging canyon walls where LIDAR scanning could not reach and steep cutbanks where sharp edges were rounded or sloped (Figures 19 and 20). While the LIDAR profile shape typically closely followed that of the TDS survey data, some bank features represent deposits from specific discharges and may provide important insights for scour and depositional patterns (i.e. from a bypass flood or local deposition around riparian vegetation).

Conclusions:

1. Significant error in elevation data was found between LIDAR and TDS data.
2. Elevational error was within or just outside the magnitudes of error reported for the LIDAR and GPS data and is therefore within reported precision limits.
3. In the absence of TDS survey data, LIDAR data presents challenges in error resolution for floodplain analysis in any dimension (1, 2 or 3D)
4. Variation of water surface elevation data from LIDAR severely limits the usefulness of the data.
5. LIDAR data can produce an accurate general representation of surface topography, however misses important features where surface topography is steep or vertical.

Improvements:

While local GPS ground control targets provided elevational control on TDS survey data, the presence of only one ground target in any given reach requires that TDS data must be rotated by eye around the ground control target. While this may be done carefully using waters edge and topographic features, some error must remain and therefore some measure of obliqueness surely exists between profiles from TDS and LIDAR. Quality control could be improved through the installation of a second ground control station in each reach, eliminating the need to rotate and translate TDS data.

References Cited:

Publication:

Grams, P.E., Schmidt, J.C., In press. Equilibrium or indeterminate? Where sediment budgets fail: Sediment mass balance and adjustment of channel form, Green River downstream from Flaming Gorge Dam, Utah and Colorado. Geomorphology.

Hodgson, M.E., P., Bresnahan, 2004. Accuracy of airborne lidar-derived elevation: empirical assessment and error budget. Photogrammetric Engineering and Remote Sensing, 70: 331-339.

Schmidt, J.C., Rubin, D.M., 1995. Regulated streamflow, fine-grained deposits, and effective discharge in canyons with abundant debris fans, in Natural and anthropogenic influences in fluvial geomorphology: AGU Geophysical Monograph. P. 177-195. American Geophysical Union.

Web References (also referenced above):

EZ Profiler, written and published by Min Lang Huang. Information and download at http://arcscripts.esri.com/details.asp?dbid=13688

WinXSPRO, developed and published by the United States Department of Agriculture, U.S. Forest Service, Stream Systems Technology Center. Information and download at http://www.stream.fs.fed.us/publications/winxspro.html.