244
April 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
does not endorse the association of imagery scale, ground
sampling distance (GSD) or contour interval (CI) with the
final accuracy of the geospatial products. The new standard
defines product’s accuracy in terms of the expected root
mean square error (RMSE) of that product. For Victor’s
project, the vertical accuracy needs to be stated as “15-
cm class,” “20-cm class,” etc., where 15-cm accuracy class
means the vertical accuracy of the product should be within
RMSE
z
= 15 cm. To mitigate this issue, we would have to
offer a hybrid approach during this transition period. By
“hybrid approach,” I mean condoning the 2-foot contour
vertical accuracy requirement as used by the legacy ASPRS
standard to derive a figure that is suitable for reporting the
accuracy according to the new standards. According to the
legacy ASPRS standard, the vertical accuracy for 2-foot
contours products is RMSE
z
= 1/3
rd
of the CI, or about 20
cm. Therefore, we can state that the digital surface model
for this project needs to meet a vertical accuracy class of
20-cm according to the new ASPRS Positional Accuracy
Standards for Digital Geospatial Data.
2.
TheTestedLandCategories
:Theprevious tables specify
five land cover categories. Those are “Bare Earth,” “Low
Vegetation,” “Medium Vegetation,” “High Vegetation,”
and “Urban.” The new standard classifies the terrain into
only two categories: “Vegetated,” which represents open
terrain and “Non-vegetated,” which represents the part of
the terrain where the ground is obscured by vegetation.
Therefore, according to the new standard, we need to
consolidate similar categories to form the two categories.
All the checkpoints located within the categories “Low
Vegetation,” “Medium Vegetation,” and “High Vegetation”
need to be combined to form the “Vegetated” category,
and points located within the categories “Bare Earth” and
“Urban” need to be combined together to form the “Non-
vegetated” category. The results of checkpoints within the
“Non-vegetated” category represent the “Non-vegetated
Vertical Accuracy (NVA)” according to the new standards,
while checkpoints fromthe “Vegetated” categorydetermine
the “Vegetated Vertical Accuracy (VVA)”. Accordingly,
the NVA needs to meet a vertical accuracy class of 20
cm. However, there is no VVA accuracy figure that can
be derived from the project specifications, as the old
standard only specifies accuracy for bare earth. According
to the old practices and ASPRS legacy standard, vegetated
areas used to be represented with dashed contour lines to
indicate non-guaranteed or lower accuracy areas. Luckily,
the new standard in Table 7.2 specifies the VVA to be
equal to ≤3.00 * vertical accuracy class. Accordingly, the
VVA for this project is ≤ 60-cm as 95
th
percentile.
3.
The Measurement Units
: The new ASPRS Positional
Accuracy Standards for Digital Geospatial Data is based
on the metric system. Therefore, converting all the values
in Victor’s tables to meters and centimeters as the new
standard suggests is advisable.
Now that all the observations on the data are presented,
here is what Victor needs to do to state the product accuracy
according to the ASPRS Positional Accuracy Standards for
Digital Geospatial Data:
Computing NVA
:
• Consolidate the 21 checkpoints located in “Bare Earth”
areas and the 21 checkpoints located in the “Urban”
areas into one table to represent the NVA.
• Recalculate the statistics for the differences in the 42
checkpoints as illustrated in the following table:
NVA
(ft)
(centimeter)
Count
42
42
Mean
0.084
2.57
Median
0.093
2.82
Minimum
-0.342
-10.42
Maximum
0.525
16.00
STD DEV
0.191
5.82
RMSEz
0.207
6.30
NVA (1.96 x RMSEz)
0.405
12.35
Or simply,
NVA
(centimeter)
Count
42
Mean
2.57
STD DEV
5.82
RMSEz
6.30
NVA (1.96 x RMSEz)
12.35
As you may have noticed, I eliminated any reference in the
last table to the foot/inch units and kept only the metric units.
I also removed some statistical terms such as MIN and MAX,
as they are not relevant to the final reporting. However, they
may come handy during the results analysis stage. I kept the
mean and standard deviation values in the table to compare it
to the computed RMSE
z
value. Evaluating the mean and the
standard deviation and comparing it to the calculated RMSE
z
value may help discover biases in the results. To calculate
the RMSE
z,
and the NVA follow the following instructions,
Annex D of the new ASPRS standard provides step-by-step
instructions and numerical examples on computing all the
statistical terms that I previously mentioned:
1. Compute the differences between the surveyed elevation
and the lidar-derived elevation for all the check points
located in open ground:
Difference =
(Z
i(map)
-Z
i
(
surveyed
)
)
“Specifying that product accuracy should meet 2-foot
contours does not align with the spirit of the new
standard”
