The original technique, described in [2] foresaw linear
prediction for all of the parameters, while, as described
above, we tried and introduced some RBF forecasting into
the forecast phase. We also introduced a pyramidal
organisation of the synthesis phase, able to speed up
synthesis of the first frame by 58% on average.
Forecast efficiency was tested on a radar rain map sequence
sensed by an S-Band weather radar located in Spino
d’Adda, Northern Italy. The sensed area covers a large
(80 km radius) portion of the Po valley, a flat area
surrounded by mountain ranges. During an event, a new
map is acquired every 90 seconds, composed of 160×160
pixels, 500×500 m2
each. Rain intensity is derived from
reflectivity using the well-known Marshall-Palmer relation.
We considered a sequence composed of 190 time-adjacent
rain maps sensed during a convective storm event on 15th
May 1989. One of such maps is depicted in figure 1 as an
example. Darker tones are associated with heavier rain.