These points were observed along 146 parallel sections, with 20 meters
interval, distributed across the slide body in the direction of the maxi-
mum slope. In this way two nearly regular DEM, before and after the event,
have been created; all the measures were carried out by a Zeiss Planicomp
C-100 analytical plotter.
Besides, a particular program was developped to drive automatically the
second time the operator on the same grid point previously observed. In
such a way it has been possible to achieve directly the height variations
occurred in the shape of the terrain.
The treatment of these observed Az values (about 9,000 points) in order
to evaluate the ground deformations, was firstly worked out by the clas-
sical method of finite elements, i.e. by computing the elementary informa-
tion of each almost regular triangular mesh. Afterwards, a more refined
analysis, supplying also statistical parameters such as accuracy and re-
liability, was attempted by applying the least squares collocation method.
1. DATA MANAGEMENT
With the measured az values, showing the surface of the landslide defor-
mations, an interpolation was performed according to the least squares
collocation method, filtering out the elevation data in order to predict
a height digital model on a 100 meters square grid (500 points instead
of 9,000). The choice of such a large interval is only due to the compu-
tational capacity required by the interpolation method. Figure 1 shows
the original vertical shifts which were to be elaborated.
One can notice that the landslide has produced a central zone character-
ized by negative variations (hill side) and another zone with positive
variations along the coast-line (North-East).
Table 1 summarizes the results obtained by the collocation method. Fi-
gures 2 and 3 show the noise and the distribution on the predicted grid
of the signal pointed out by the collocation. Figure 4 reports the dia-
gram of the covariance function and noise of the vertical shifts, the
radial distance being chosen as ordering parameter (radius). The covari-
ance function of the vertical displacements reveals a signal larger than
the noise and resulting in a four-hundred meters correlation length
(four time larger than the grid interval).
It means that the least squares collocation method can explain a very
high percentage of the phenomenon. The behaviour of the signal is very
regular inside the landslide and almost zero outside. The residual noise
has a r.m.s. of 0.51 m; this value becomes 0.30 m after rejection of 29
points (over 376) out of tolerance. Moreover, these noises result uncor-
related, as shown by the diagram of the covariance function in figure 4.
- 135 -
