International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012
2012 XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
order to spatial resolutions have been tested in the interpolation, starting In the interpolation of TRI and TR2, POL function does not
différent from 10 m and decreasing it by steps of one meter. The provide consistent accuracies, even pushing the resolution of
residuals of each resulting DTMggrip have been computed with DTMggp to one meter. The other algorithms provide
the grid respect to the starting TR and to the set of check points (TE). satisfactory results, the best being CRS that provides one meter
Olerance For each interpolation function and for each starting TR, the accuracy already at the resolution of three meters.
vely the coarser DTMGrp Whose accuracy is consistent with the starting As expected, the TIN model requires less storage space than the
dad: dain data set has been selected. In Table 3, the statistics and the grid model. The saving ranges from 72% to 97% for TR1, from
equal to characteristics of the final DTMGrm are reported. 68% to 80% for TR2, from 59% to 67% for TRS.
e finally
> results. RO CILE
le 1 the esObs esChec
T 1K/ Rôy (NX | NY KA) Arm) | RMSE (m): | Vinh] MSE)
Im 1 1830 | 1784 25506 -0.0 0.1 -0.1 1.1
IDW 2m 5 366 357 1021 -0.3 2.0 -0.2 2.0
Sm 10 183 179 256 -0.9 5.0 -0.7 4.8
Im 1 1830 | 1784 25506 -0.2 1.9 -0.2 1^5
POL 2m 1 1830 | 1784 25506 -0.4 35 -0.4 2.6
5m 8 229 223 399 -0.8 5.5 -0.9 5.0
Im 3 610 595 2836 -0.2 1.3 -0.1 ;
CRS 2m 5 366 357 1021 -0.4 2.4 -0.1 1.8
5m 10 183 178 255 -0.9 5,2 -0.8 4.9
Im 2 915 892 6376 -0.2 1.1 -0.1 1.1
SWT 2m 5 366 357 1021 -0.4 2.3 -0.3 2.0
5m 9 203 198 314 -0.7 4.7 -0.7 4.8
Im 1 1830 | 1784 25506 | -0,10 0,50 | -0,03 1,19
TPS 2m 4 458 446 1596 | -0,44 2,06 | -0,10 2,05
5m 10 183 178 2551-139 5,71 -0,77 4,99
: Table 3. Characteristics of the interpolated grids. T: technique used to interpolate; TR: interpolated dataset; R: final resolution; NX
and grid and NY: number of nodes in X and Y; S: storage size; ResObs, ResCheck: statistics of the residuals on the used observations and the
ms check points; M: mean; RMSE: root mean square error.
| Table 2 -
TR NSplines Number of coefficients for each level L ResObs ResCheck S
polation LO L] L2 L3 L4 LS L6 L7 M(m) |RMSE (m)) M(m) |RMSE (m) (KB)
eighting Im | 5616 9 9 34 | 106 | 336 | 955 | 2080 | 2087 0.0 1.1 0.0 1.4 55
mpletely 2m 1767 9 9 34| 106| 292] 610, 707 / 0.0 1.8 0.0 2.2 17
WT) and 5m 349 9 9 34 sS5| 112 100 / / 0.0 3.4 -0.4 4.8 3
ve been Table 4. Statistics of DTMyg. TR: interpolated dataset. NSplines: total number of activated splines. ResObs, ResCheck: statistics of
natically the residuals on the used observations and the check points. M: mean. RMSE: root mean square error. S: storage size.
ResObs ResCheck
TE T TR[ Be | NX | NY | EB Pun] Ran DManl EMSE QS
31869 IDW Im 3 610 595 2836 -0.2 1.1 -0.1 1.2
97.47 CRS Im 4 458 446 1596 -0.3 1.9 -0.2 1.4
32.23 SWT im 4 458 446 1596 -0.3 1.9 -0.2 1.4
35.59 TPS Im 3 610 595 2836 -0.2 1.2 -0.0 1.4
27.83 Table 5. Characteristics of the grids interpolated from TR1 with accuracy similar to DTMyg. T: technique used to interpolate; R:
n m. spatial resolution; NX and NY: number of nodes in X and Y; S: storage size; ResObs, ResCheck: statistics of the residual on the
used observations and the check points; M: mean; RMSE: root mean square error.
S
(KB) 3.1 Interpolation and storage size of the multi-resolution checkpoints are reported. Moreover the storage size in Kbytes
795 model is given.
The interpolations of TR1, TR2 and TR5 activate eight, seven
325 By interpolating TRI, TR2 and TR5, the coefficients of the and six levels of splines respectively. However, the final
104 multi-resolution splines model (DTMyr) have been estimated, accuracies of DTMyr(TRI) and DTMyr(TR2) are not
N V: according to the criteria described in Sect. 2. Then, the relevant completely satisfactory: indeed, their RMSEs exceed one and
N F: accuracies and storage sizes have been analyzed. two meters respectively and are worse than those of the
storage The statistics of the DTMyrs obtained by interpolating T1, T2 respective DTMGrms. The criteria that we are adopting to
and T5 are summarized in Table 4. For each TR, the total automatically activate the splines (f=1 and k=4) is conservative
ve to be number of splines, the number of splines per level, the statistics and, in these cases, provide sub optimal results: more analyses
coarser of the residuals on the used observations and on the are needed and will be discussed in a final paper. Moreover, it
different should be taken into account that the decimation criterion
11
