OR/14/051 DEM assessment

From Earthwise
Jump to: navigation, search
Whitbread, K. 2014. The geomorphic impact of road construction: a case study of the A9 in Scotland. British Geological Survey Internal Report, OR/14/051.

Accurate reconstruction of ground surface change using a borehole — DEM comparison is dependent upon the accuracy of both the measured ground surface elevation recorded in the borehole log, and the DEM representation of the post-construction ground surface.

Neither the accuracy of the borehole ground surface levels (given in metres), nor the methods used to derive them are recorded in the logs of the boreholes utilised in this study. Consequently it is difficult to assess the accuracy of the recorded borehole surface levels. Indirect methods are used in Borehole records of the ground surface and Quantifying ground surface change to assess the relative accuracy of the recorded borehole start heights.

The accuracy of the available DEMs was assessed for the southern test area through comparison with OS spot heights. Fifty nine OS spot heights were digitised in ARCGIS from 1:10 000 scale raster maps (last revised 2000) within a range of 500 m either side of the A9 corridor. The 1:10 000 scale raster maps have been superseded by OS VectorMap data products since March 2014, however comparison of the distributions of spot heights from OS VectorMap with the 1:10 000 scale raster maps demonstrated that significantly more spot height values were recorded in the vicinity of the A9 on the older dataset and therefore the 1:10 000 scale raster maps were used.

DEM elevations were extracted for each spot height from the NEXTMap 5 m resolution DTM, the PGA 2 m DSM (not corrected for trees/buildings), and the 5 m resolution PGA DTM. The specified vertical accuracy of these dataset is given in Table 1.

DEM versus OS spot heights

Comparison of the OS spot heights with the DEMs indicates that elevations derived from the PGA DTM (HPT) show the closest correspondence with the OS spot height elevations (HOS), i.e. the closest mean elevation and the lowest mean difference (Tables 2 and 3). A two sample t-test shows that there is no significant difference in the means of HPT and HOS, and a one sample t-test indicates that the mean of the difference between HOS and HPT is not significantly different from zero. These results indicate that there is good correspondence between the PGA DTM elevation and the actual ground surface level measured at the OS spot heights.

At the spot height locations, both the PGA DSM (HPS) and the NEXTMap DTM (HNM) yield mean elevations higher than the mean of HOS, i.e. the models overestimate the real ground surface elevation.

Table 2    Statistics for the OS spot heights and the extracted DEM surface elevation
(Calculated in Minitab)
Mean 77.71 83.9 78.3 85.8
SE 2.57 2.8 2.57 2.96
St Dev 19.78 21.5 19.7 22.74
95% CI 5.2 5.6 5.1 5.9
n 59 59 59 59
Table 3    Statistics for the difference between OS spot heights and the DEM surface elevation (Calculated in Minitab)
HOS minus HPS HOS minus HPT HOS minus HNM
Mean -6.19 -0.32 -8.09
SE 1.02 0.27 0.88
St Dev 7.84 2.09 6.76
95% CI 2.0 0.54 1.76
Variance 61.52 4.36 45.69
Skewness -0.89 0.87 -0.54
Kurtosis 0.02 7.41 -0.80
n 59 59 59

In the case of the NEXTMap DTM, HNM is significantly higher than the mean HOS (p<0.05) and the mean of the difference between HOS and HNM is different to zero (p<0.001). For the PGA DSM, the mean of HPS is not significantly higher than the mean of HOS (p>0.05), but the mean of the difference between HOS and HPS is significantly different to zero (p<0.05), indicating that the PGA DSM does generally over-estimate the surface elevation at the spot height locations.

DEM selection

The overestimation of surface elevations by the PGA DSM is likely to be due to the effects of trees or buildings as many of the spot height locations are in forested or urbanised areas. The overestimation of the ground surface level by the NEXTMap DTM also suggests that the algorithms used to remove the trees and buildings from the model were not effective. By contrast the procedures used to remove surface features from the PGA DTM have yielded a terrain model that is consistent with the OS terrain data.

Within both of the test areas, the A9 route is flanked by extensive areas of woodland, and cuttings and embankments associated with the road construction are also generally forested. For this reason, the PGA DTM is used in the following analysis as it is found to be the most accurate ‘bald earth’ model in the site areas.