2012 
)Ximate, 
and the 
OL are 
ochastic 
ns are 
) that is 
like, for 
rties are 
lied to 
classical 
reported 
ily and 
ause the 
number 
)L, the 
| of the 
vations: 
red and 
iation of 
ons and 
methods 
Spline) 
lytically 
needed 
ple the 
by the 
rvations 
e cross- 
tions. € 
' surface 
is to be 
quire as 
; provide 
as both 
d spatial 
nate the 
> heights 
el. If the 
ion, the 
vant grid 
bilinear 
odes of a 
eed each 
le, while 
nalytical 
ta based 
rpolation 
odel and 
pect to a 
DACH 
liminary 
roximate 
based on 
width in 
lensity in 
ormation 
" splines, 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
corresponding to different halving steps, are considered. A new 
level halves the width of the support of the previous level. 
We suppose that a field A(t) = h(x, y) has been sampled at m 
locations t,,t,,...,t 
stm» 
t; 2 (x,, yj) . The interpolation domain is 
[Emin>Enax ]- The field observations y, are modelled by means 
of a suitable combination of bilinear spline functions and noise 
V;: yy =h(t,)+v, . 
Let define the following 
1-4 0sg«l 
9(q) - 10 q>1 
p--q) q<0 
INO UA 
The height field is given by the 
M-A | N41 N,-1 
h(t)= S > > Ai iy Pax, {ox Yor (ay) (1) 
h=0 | iy=0 iy=0 
  
where qx = X= Axi, i Xmin , qy = Vi = Ay,l, = min > 
AX, Tmax ET mn Ay, = J max = s. Mis the number of 
2 or 
levels, Ais is the coefficient of the spline at the grid node 
(..), N,is the number of nodes at the A level, 
N za" rp, 
In the estimation, all the field observations are tiled in a vector 
y, -y*Yv-Ak*v, E[v| 20 C, 2 C,, - o7l 
where AA are respectively the vector containing all the 
À 
hjix siy 
coefficients to be estimated and the design matrix 
obtained by applying (1) to the observations. The estimation of 
A=(ATA)'Ay, is based on the well known LS principle. 
Two innovative aspects characterize our interpolation 
approach. 
Given a level, each local spline is individually activated if no 
spline of some lower level has the same application point. 
Moreover the spline is activated if at least f,f 21, 
observations exist in at leastk (k=1,2,3,4) quarters of the 
spline support: f,k are input by the user. They must be 
choosen according to two criteria. Clearly f=1k =1 
correspond to no redundancy in the estimation, while bigger 
values smooth the interpolating ficld. Moreover, particular 
spatial configurations of the observations can produce a LS 
system that, although redundant, is either rank deficient or ill 
conditioned: in these cases, f and k should be increased 
independently of redundancy considerations. The individual 
activation of the splines guarantees a real multi-resolution 
interpolation. 
The levels are activated iteratively from level 0 to level M. A 
new level is activated if and only if its splines significantly 
improve the accuracy of the interpolation. 
Let suppose that M (A20,L..,M —-1) levels have been 
already activated, for a total number of N,, splines estimated 
coefficients. The criterion to activate or not the M + 1 -th level 
is based on a significance analysis. Let suppose that N,,., is 
the number of splines activated with the new level and use 
A, to indicate the vector containing all the splines 
coefficients of the new level. We want to evaluate the 
following hypothesis 
H, s 2EÍS ls 0} Q) 
If HO is true, the coefficients of the new level are not 
significant, the relevant estimates can be discarded and the 
iterative process can be stopped. Otherwise, the coefficients 
should be kept and a new higher level should be tested. Let 
define the following quantities 
2 2 
Yu = Yo Yu Tu 7 Vy /(m- Ny) 
ut ^ A2 Ed a = 
Vua P Yo Yu Gun = Yun fm —Ny 7 Nu) 
where y, -[A...,^,] is the vector containing all the field 
observations, $,,,y,,., are the a posterior estimates provided 
by LS. From a geometrical point of view the situation is 
depicted in Fig. 1. 
  
Yo 
9S(w)/ | 0 me) 
i 
y CRT 
rra. 
Yim) 
Vim) 
VtMa) 
  
Figure 1. Geometric interpretation of the significance analysis 
ofa new level. 
If (2) holds, yz E {yo} €Vu » and the usual significance 
analysis on the a posterior variances can be applied 
22 A2 
(m em Ny JO TR (m = Ny = Ny, )O 4 A 
a2 
Nya Ow 
2 
Tr Nd 
s 
Vus =F 
7 M+1 N Nr Nr 
A ne Nu Nu /(m-N,-Ny,4) 
  
(3) 
  
where 7;,F,, indicate respectively a chi square variable with 
i degrees of freedom and a Fisher variable with (i, j) degrees 
of freedom. A threshold value F, with significance a can be 
set and the zero hypothesis (2) can be tested by (3). 
2.1 Storage requirements 
To evaluate the DTM storage requirement of the different 
models, at first a numerical comparison between grids, TINs 
and our multi-resolution approach is here presented. 
Particularly, an occupation of 64 bits (8 bytes) is hypothesized 
for the horizontal coordinates and the height of a point. In the