Digital surface models (DSMs) for the intertidal zone at Post Point, Bellingham Bay, WA, 2019-06-06

Vertical accuracy was estimated using two methods to compare the DSM vertical elevations to concurrently collected post-processed kinematic (PPK) GPS measurements. The first method used a comparison of SfM-derived ground control point (GCP) positions to PPK GPS measurements at those points. Due to the time-intensive process of placing GCPs in the field, all available GCPs were used for registration and camera optimization in the SfM processing workflow during the creation of the final DSM. To evaluate the vertical positional accuracy of the DSM after processing was completed, a subset of GCPs was disabled one-at-a-time using a python script to create 'temporary check points'. With a single GCP temporarily disabled, camera optimization was performed with all lens parameters fixed, and all other GCPs enabled. The residual errors of the check point relative to its GPS-measured position were recorded. After all temporary check point iterations were complete, the root-mean-square error (RMSE) and mean-absolute error (MAE) were calculated. For the Post Point 4-centimeter resolution DSM (PostPoint_2019-06-06_DSM_4cm.tif), the vertical RMSE was 0.040 meters (MAE 0.030 meters), and for the Post Point 2-centimeter resolution DSM (PostPoint_2019-06-06_DSM_2cm.tif), the vertical RMSE was 0.042 meters (MAE 0.034 meters) relative to the GPS measurements. The addition of the estimated vertical GPS uncertainty (0.030 meters) in quadrature results in a total vertical accuracy estimate of 0.050 meters for the 4-centimeter resolution DSM and 0.052 meters for the 2-centimeter resolution DSM. It should be noted that this estimate is for areas of bare ground or low vegetation where GCPs were placed. Additional sources of error such as poor image-to-image point matching due to vegetation or uniform substrate texture (such as sand) resulting in poor surface reconstruction may cause localized errors in some portions of the DSM to exceed this estimate. A second method was used to attempt to quantify the vertical errors in areas away from the GCPs. During field data collection, topographic measurements on unvegetated areas were collected with backpack-mounted PPK GPS. These measurements were compared to the DSM elevations using bilinear interpolation at each GPS point to derive additional accuracy estimates for the DSM. For the Post Point 4-centimeter resolution DSM (PostPoint_2019-06-06_DSM_4cm.tif) the vertical RMSE of 7,465 backpack-mounted PPK GPS measurements compared to the DSM elevations was 0.061 meters (MAE 0.043 meters). The mean-error (vertical bias) of the GPS measurements relative to DSM elevations was -0.020 meters, meaning the 4-centimeter resolution DSM was, on average, higher than the GPS measurements. For the Post Point 2-centimeter resolution DSM (PostPoint_2019-06-06_DSM_2cm.tif) the vertical RMSE of 3,589 backpack-mounted PPK GPS measurements compared to the DSM elevations was 0.065 meters (MAE 0.048 meters). The mean-error (vertical bias) of the GPS measurements relative to DSM elevations was -0.016 meters, meaning the 2-centimeter resolution DSM was, on average, higher than the GPS measurements. The addition of the estimated vertical uncertainty of the backpack-mounted GPS (0.035 meters) in quadrature results in a total vertical accuracy estimate of 0.070 meters for the 4-centimeter resolution DSM and 0.074 meters for the 2-centimeter resolution DSM using this method. These slightly higher errors are due in part to the less precise nature of the backpack-mounted GPS measurements. However, these backpack-mounted GPS comparisons do provide a more conservative, and more spatially distributed estimate of the true accuracy of the DSM than the GCP check point method. We present both estimates here to provide the end-user with a more complete understanding of the accuracy of the final data product.