Monitoring of snow cover on Italian Alps using AMSR-E and Artificial Neural Networks Osservazione di aree innevate sulle Alpi italiane con AMSR-E

glassesbeepingAI and Robotics

Oct 20, 2013 (4 years and 2 months ago)

81 views

Monitoring of snow cover on Italian Alps using AMSR
-
E and
Artificial Neural Networks


Osservazione di aree innevate sulle Alpi italiane
con AMSR
-
E


E.Santi
1
, G. Fontanelli
1
,

S.

P
aloscia,
S. Pettinato
1
, A. Crepaz
2

1
Institute of Applied Physics
-
National
Research Council, Florence, Italy

2
CVA Centro Valanghe Arabba, Italy

Ph: 0039 522 6431; FAX: 0039 522 6434; E
-
mail: e.santi@ifac.cnr.it


T
he Advanced Microwave Scanning Radiometer (AMSR
-
E)
,

installed onboard AQUA satellite
, r
ecently
stopped working and has been switched off, after ten years of
activity
. The large number of studies carried
out
worldwide

in these years pointed out that the AMSR
-
E data can be successfully used for global and
regional investigations of surface par
ameters
(i.e.

soil moisture, vegetation biomass and snow cover
)

and
their temporal changes. These studies
combined with

field experiments

have revealed
a
very high sensitivity
of

microwave emission to physical parameters of soil, snow
-
pack and vegetation,
which are of great interest
in hydrology, meteorology, climatology and agriculture. The
coarse
spatial resolution of
AMSR
-
E

is,
undoubtedly,

a strong limitation, because
it
hampers a detailed analysis of the surface, especially in
variegated landscape terr
itories.
However
, this characteristic, associated with the frequent revisiting of the
Earth’s surface, allows
an
extended
spatial and temporal
monitoring
of

surface features.
The
incoming
GCOM mission will carry onboard the AMSR2 sensor,
heir
of the AMSRE
, and therefore we expect that the
retrieval algorithms developed in the past decade for AMSRE
could

be successfully applied to this new
radiometer.

Among the other surface parameters, AMSR
-
E revealed a good sensitivity to snow depth (SD) and snow
water eq
uivalent (SWE). Several algorithms for the retrieval of
SD/
SWE from multi
-
frequency radiometric
systems have been developed basing on empirical relationships or statistical approaches [e.g. 1
-
3
].

A
n algorithm
that merges empirical approaches and
Artificial Neural Network
(ANN)
techniques
for
estimating the

snow
cover extent (SCE) and the snow depth

(SD)
,

from AMSR
-
E data
,

is proposed

i
n this
paper.
This algorithm uses the Frequency Index at Ku
-

and Ka
-

bands to detect the SCE and applies an ANN
in
version method using the
AMSR
-
E measurements from X
-

to Ka
-

band, in V pol

to estimate the SD.

The algorithm includes a

simple
filtering procedure
, derived from the smoothing filter
-
based intensity
modulation technique (SFIM) [
4
], and aimed at increasing
the resolution of X
-

and Ku
-

bands up to the
higher resolution Ka
-
band
, in order to prevent possible errors related to the different field of view of the
sensor at different frequencies. The

field of view
of the AMSRE sensor increases noticeably when the
f
requency decreases, and therefore the data collected at different frequencies are in fact related to slightly
different targets: this fact could hamper the retrieval accuracy when multi
-
frequency data are merged in the
ANN retrieval algorithm
.

Th
e

algorit
hm has been developed and validated by using
two

large dataset
s

of snow parameters collected
over
Siberia (provided by JAXA within the framework of the GCOM/AMSR2 mission) and
Scandinavia
(derived from the archive

http://meteo.infospace.ru/
).


After validation, the algorithm has been applied to the time series of data collected in ten years on Italian
Alps, in order to evaluate the yearly evolution of snow cover and snow depth. A more in
-
depth
analysis has
been ca
rried out for a limited period
by comparing the SD outputs of the algorithm to an area of about
100x100 km
2

located
in

the Italian Eastern Alps, in Veneto region
, with

snow depth measurements derived
from a network of meteorological stations.