Transcript Nonlinear analysis/forecasting and chemical monitoring

Nonlinear analysis/forecasting
and chemical monitoring
Formation and training of researchers
Antonello Pasini
CNR - Institute of Atmospheric Pollution
Rome - ITALY
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
1
The Institute of Atmospheric Pollution of the
National Research Council of Italy (CNR)
proposes an activity of formation and knowhow transfer to researchers of Small Island
States in the field of climate/global change:
• Topic 1: nonlinear analysis/forecasting by
means of neural network modeling.
• Topic 2: chemical monitoring techniques
with emphasis to remote pollution.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
2
Neural network modeling
Preliminary considerations
• The atmosphere-ocean-biosphere system is
complex and nonlinear.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
3
Neural network modeling
Preliminary considerations
• At present, GCMs show good results, but
not fully reliable regional forecasts.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
4
Neural network modeling
Preliminary considerations
• These models require
specialized man-power
and high-cost
supercomputers 
GCMs are not easily
managed by
developing countries.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
5
Neural network modeling
• In this situation, a non-dynamical statistical
method can be useful either for influence
analysis and for localized forecasting.
• Neural networks learn from past data and
are able to generalize to unknown
situations.
• Recently, they have been applied to
environmental problems.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
6
Neural network modeling
The biological inspiration
co nnectio n
neuro n
Inp ut
layer
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
H id d en
layer
O utp ut
layer
7
Neural network modeling
• A weighted sum converges at each neuron of
hidden and output layers…
• … where a nonlinear activation function is
calculated.
1
sigmoid output
0.8
n=3
0.6
n=8
n=20
0.4
n=50
0.2
0
-10
-8
-6
-4
-2
0
2
4
6
8
10
weighted sum
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
8
Neural network modeling
co nnectio n
neuro n
Inp ut
layer
2nd AOSIS Workshop,
Apia (Samoa)
H id d en
layer
O utp ut
layer
• After a training on a
statistical significant
data set, the network is
able to fix its
connection weights...
• … thus giving us a
diagnostic or
prognostic relationship
between inputs and
output.
26 July - 4 August 2000
9
Neural network modeling
How to use a neural network model in
the field of climate change?
• Three options:
- Influence analysis;
- forecasting from observations;
- post-processing of a GCM’s output.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
10
Neural network modeling
Influence analysis
• Estimation of forcing relevance on climate
change and global heating related to
temperature scenarios:
- at global level;
- at regional level.
• Reconstruction and forecast of heating.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
11
Neural network modeling
Forecasting from observations
• Departing from relevant meteo-climatic
observations, a neural model is able to
forecast SST or other physical variables.
• Attempts at forecasting future scenarios
(CO2 doubling problem) are now in
progress.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
12
Neural network modeling
Forecasting from observations
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
13
Neural network modeling
Post-processing of a GCM’s output
• Neural networks may be used in order to
improve forecasts of GCMs on a local or
regional scale.
• Typical variables to be predicted are
temperature and precipitation.
• Model errors become lower.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
14
Neural network modeling
Conclusive considerations
• Dynamical treatment (via GCMs) of climate
change topics leads to good results even
with difficult parameterizations.
• Neural modeling offers a way for analyzing
these topics with a fully nonlinear nondynamical tool.
• Neural models require simple workstations
or PCs and can be managed by few experts.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
15
Neural network modeling
Conclusive considerations
• We think that the formation of one or more
neural modelers is a good opportunity for
the Small Island States in order to meet the
advanced research in the topic of climate
change and to analyze and forecast
phenomena that strongly affect life and
development.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
16
Neural network modeling
Suggested outline of training
• Basic training (20-30 days) in neural
modeling at an AOSIS University (e.g.
USP) for a wide audience.
• Advanced training (3-12 months) and
application to local data to be, eventually,
held at the CNR - Institute of Atmospheric
Pollution (Italy) for two AOSIS researchers.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
17
Neural network modeling
Essential bibliography
• Hsieh and Tang - Bull. Am. Met. Soc. 79,
1855-1870, 1998.
• Gardner and Dorling - Atmos. Environm.
32, 2627-2636, 1998.
• Walter et al. - Met. Zeitschr. 7, 171-180,
1998.
• Tangang et al. - J. Clim. 11, 29-41, 1998.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
18
Neural network modeling
Essential bibliography
• Tangang et al. - J. Geophys. Res. 103 (C4),
7511-7522, 1998.
• Marzban and Stumpf - Wea. Forecast. 13,
151- 163, 1998.
• Pasini and Potestà - in Neural Nets: WIRN
Vietri-96 (Springer), 263-269, 1997.
• Pasini et al. - submitted to J. Geophys. Res.
D, 2000.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
19
Chemical monitoring
Preliminary considerations
• Small changes in the releases to the
atmosphere of some gases can have
unexpected and long-lasting global effects.
• Atmospheric chemistry determines the
extent to which source gases may be
important.
• In general, the less reactive a gas, the more
global the scale of the problem.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
20
Chemical monitoring
Preliminary considerations
• In the atmosphere complex interactions of
physical processes and chains of chemical
reactions are involved.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
21
Chemical monitoring
Preliminary considerations
• In general, in climate change studies, we
consider the effects of changes in gas and
aerosol concentrations on physical
parameters (temperature, sea level, etc.).
• But changes in the chemical composition of
the atmosphere may produce changes (and
problems) also in the natural environment
(soil, forests, lagoons, etc.).
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
22
Chemical monitoring
Preliminary considerations
HO NO
[ p m o l/ m o l]
• Even in remote areas we can observe transport
phenomena or, at least, influences of surface
interactions with the atmosphere.
10.0
1.0
0.1
80
2nd AOSIS Workshop,
Apia (Samoa)
100
120
140
Dayof- Year
26 July
4 August 2000
160
180
23
Chemical monitoring
Preliminary considerations
• Another element of
concern is the increase
in oxidizing capacity
of the atmosphere…
• … recognized even in
marine and remote
environments with
forests.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
24
Chemical monitoring
Preliminary considerations
• The tropical forests
are emitters of VOCs,
precursors of oxidized
compounds.
• The background of
ozone has shown a
doubling in the last 20
years.
• Phytotoxic effects?
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
25
Chemical monitoring
• The Institute of Atmospheric Pollution is
widely involved in researches concerning
remote or semi-remote areas (Himalaya,
North and South Polar regions, semi-remote
sites in Europe).
• Many innovative sampling and analytical
procedures were developed for monitoring
low concentrations of chemical species
(either gas and particulate matter).
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
26
Chemical monitoring
• Several automatic instruments (some of
them developed in our Institute) are used for
monitoring low concentrations of primary
and secondary pollutants.
• Particular attention is paid to secondary
pollutants and to particulate matter PM10
and PM2.5. Ad hoc instruments were
developed for their detection.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
27
Chemical monitoring
• These instruments include an automatic
monitor for particulate matter and annular
denuders (+ filter packs) for detecting
secondary pollutants either in gas and
particulate phase.
• DOAS (Differential Optical Absorption
System) and passive samplers (very useful
in remote areas) are also used.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
28
Chemical monitoring
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
29
Chemical monitoring
• These instruments are intended for
monitoring species which are dangerous for
surface ecosystems, including:
- tropospheric ozone,
- particulate matter of secondary origin,
- VOCs and photo-oxidants,
- acidity containing compounds,
- organic compounds,
- heavy metals (mercury).
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
30
Chemical monitoring
• The analysis of data from non-automatic
instruments involves techniques such as gaschromatography and mass spectrometry (great
expertise at our Institute).
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
31
Chemical monitoring
Conclusive considerations
• The study of time and space variability of
these compounds helps to clarify either
long-range transport phenomena and the
increase in background concentrations.
• These pieces of information are important
for assessing changes in radiative balance of
earth, but also for early warning on global
changes in chemical cycles.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
32
Chemical monitoring
Conclusive considerations
• These latter changes occurring in the
atmosphere may influence ecosystems in a
dangerous way.
• This is particularly serious in areas which
are universally considered as natural
heritage of the entire World.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
33
Chemical monitoring
Suggested outline of training
• Training on instruments and techniques for
chemical monitoring (6-12 months) to be
held at the Institute of Atmospheric
Pollution (Italy) for two/three AOSIS
researchers.
• Final in situ campaign (15-20 days) at an
AOSIS island, in order to apply learned
techniques in the local environment.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
34
Chemical monitoring
Essential bibliography
• Allegrini et al. - Int. J. Environm. Anal.
Chem. 55, 267-283, 1994.
• Allegrini et al. - J. Chromatography A846,
265-268, 1999.
• Hillamo et al. - Int. J. Environm. Atmosph.
Chem. 71, 353-372, 1998.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
35
Chemical monitoring
Essential bibliography
• Beine et al. - Atmosph. Environm. 30, 10671079, 1996.
• Herring et al. - J. Atmosph. Chem. 27, 155178, 1997.
• Beine - Chemosphere: Global Science
Change 1, 145-151, 1999.
2nd AOSIS Workshop,
Apia (Samoa)
26 July - 4 August 2000
36