Accuracy Assessment Project

The purpose of this project is to transform the vertical datum of LiDAR points. This an especially difficult task because of the processing power required to translate every point.

LiDAR data is usually collected using the WGS84 datum. Horizontal points are typically transformed from lat and long to UTM during kinematic processing. The vertical information is often not transformed and is instead left referenced to the datum’s ellipsoid (NAD83 CSRS) and stored as ellipsoidal height. To use ground control points collected using real time kinematics it is important that the LiDAR point cloud be transformed into orthometric heights (CGVD28 datum).

After this shift has been applied it is important to assess the accuracy of the new translated points. In order to do this transformation ground control points are required. These points must be converted into TerraScan compatible files. Check point files typically consist of Fundamental Vertical Accuracy (FVA). These points should have the least errors. There are also Supplemental Vertical Accuracy (SVA), Experimental Vertical Accuracy (EVA) and Consolidated Vertical Accuracy (CVA) whose check points have less desirable, varying land cover types. The consolidated vertical accuracy is a file containing both FVA and SVA control points.

Case Study

The LiDAR survey was flown over the town of Middleton, Nova Scotia by the Applied Geomatics Group on August 18th, 2010 at a height of 1200m, scan angle of 18 degrees in strips with a resolution of 0.829m. The system used was an integrated applanix POS-AV 510, ALTM 3100 and Rollei digital camera. The data was provided in NAD83 UTM Zone 20 . The MicroStation and TerraScan projects have already been created and the points have been classified. The next step is to correct the data to orthometric heights and assess for accuracy. This workflow and instruction was taught by Rob Hodder with the Centre of Geomatic Sciences

LiDAR : Orthometric Heights and Accuracy Assessment

This image shows the study area in Middleton, Nova Scotia, Canada



Transforming to Orthometric Heights

To transform the vertical points the first step is to create reference points. This is accomplished by reading the block points and exporting a lattice model. Only every 100th point was loaded for processing speed and the grid spacing for the lattice was set to 500 m. A closer grid spacing is not required because geoid models are no more accurate than this. Gaps of up to 3 pixels were set to be filled automatically.

In order to retain precision, decimal places were kept up to 3 places resulting in mm precision. The data was exported as an XYZ text file.

In order to transform elevations from the NAD83 CSRS98 ellipsoid to the CGVD28 datum (which is the orthometric height), the Canadian Height Transformation 2.0 model (HT 2.0) must be used. The simplest way to do this is by using the Natural Resources Canada free tool GPS-H to do the conversion.

This tool was downloaded from Set your projection and create input settings using the input edit method.

To calculate the orthometric height the geoidal height is subtracted from the ellipsoidal height

Orthometric = ellipsoidal height – geoidal height.

Microstation uses the equation

Orthometric = ellipsoidal height + geoidal height.

This means that you must first massage your data in excel and change the signs on your geoidal heights.

For more information on how to do this, check out this tutorial.


Accuracy Assessment

Now that the LiDAR points have been transformed to orthometric heights they need to be compared to measured check points.

The output control report tool was used to compare known points to the LiDAR project points and produce an accuracy report using Max variable slope of 6 degrees and an expected variation of elevations of 15cm.

This tool creates a TIN (Triangulated Irregular Network) from LiDAR points per block. Each control point elevation is compared to the elevation of the TIN at that point. The reports of the 4 files used (CVA, EVA, FVA, SVA) were saved for future analysis.

Individual check points with >25cm error were checked (maximum of 2 per check point file). Error explanations are discussed below. The best controls turned out to be points on very flat terrain without much surrounding vegetation (roads, grassy plains, parking lots etc). The worst errors were locations with slopes like a grassy hill, near guard rails and heavily vegetated areas.

The 95% confidence level of the FVA control points are calculated and compared to the USGS LiDAR base specifications for quality level.

The SVA and CVA are expected to be more variable are analyzed using 95th percentile accuracy. EVA is experimental and is not used in this accuracy assessment.

In the image below one block is 1km x 1km.


FVA Check Points

 Point 9 and point 4 for were assessed for errors. Point 9 was located on a bridge. The positioning of the point suggested that the slope error reported (very large error) was caused by a railing or other such safety structure on the edge of the bridge. The maximum slope was set to 6 degrees so it is assumed that this point did not meet those standards. Point 4 does not have major errors (24cm error) but it was examined to determine why it had a larger error than the average acceptable 15cm error. A specific cause was not determinable suggesting that it was due to extrapolation or surveying error.


Profile 1, Point 9, Slope Error – Caused by guard rails on a bridge. Looking North

Profile 2, Point 4, Parking Lot, <25cm error, uncertain cause. Looking Northwest

EVA Check Points

The second file checked was the EVA Check points file. Point 42 and point 45 were assessed for errors. Both had slope errors. The profiles below show that these points are clearly on slopes. Both are also beside a road on a grassy hill. It is best practice to avoid hills during check point acquisition as they increase errors. The error associated with point 45 is also related to ground slope.


Profile 3, Point 42, Slope Error – Road side grassy slope. Looking Northeast

 Profile 4, Point 45, Slope Error – Road side grassy slope. Looking North

SVA Check Points

Finally the SVA file was checked. There were no major errors with one of the largest errors being approximately 20cm. This is possibly due to data gathering error caused by the encroaching vegetation on either side of the road.

 Profile 5, Point 37, 0.203 error – road error due to vegetative cover. Looking West


ASPRS Vertical Accuracy

 The next step was to compare the 95% confidence level and 95th percentile of the different check points. For the FVA the 95% Confidence Level must be used which expects a normal distribution of the data. For CVA and SVA the 95th Percentile is used.

FVA is being switched to NVA (non-vegetated vertical accuracy) and SVA is being switched to VVA (vegetated vertical accuracy) in 2014 guidelines. The USGS has a document that gives LiDAR specification standards which the resulting data will be compared to.

With an FVA RMS of 12.1cm this data falls into the QL1 and QL2 level for RMS error, but QL3 for its confidence level. I presume that this means that it is Quality Level 3 data even though the RMS is better than Quality Level 3.

The SVA RMS of 15.44cm means that it fits into the VVA quality level 1.

Whether or not this was acceptable would depend on your clients quality level requirements.

The above Dz values are measured in metres.

Table copied from USGS LiDAR Base Specifications 2014 document