Climate short-term self-learning climate model for Northern Europe

1
Climate short-term self-learning climate model
for Northern Europe

• O.M.Pokrovsky
• Main Geophysical Observatory
• St.Petersburg

2
Outlines

• Theoretical background
• Observing data filtering by low oscillation
  approach
• Nonlinear forward and feedback link description
  by neural network tools
• Encoding and decoding of the atmospheric
  circulation patterns by fuzzy analysis
• Implementation
• Climate short-term self-learning model
  model for Northern Europe
• Result verification

3
Air Temperature at Northern Europe
4
Global Irradiance at Northern Europe
5
Atlantic Ocean Surface Temperature
6
INSTABILITY IN WEATHER ANALYSIS AND FORECAST

• Ill-condition feature of objective analysis of
  meteorological fields
• Lorenz forecasting equations
• instability generated by
• linear
• nonlinear terms

7
Why slow oscillationswere selected?

• Slow oscillations exhibit the highest amplitudes
  among other frequencies
• Slow oscillations explain the most part of total
  variability
• Slow oscillations are most stable to
  perturbation in initial conditions for numerical
  weather models based on differential equation
  solution

8
Spatial Differentiation
9
Time Differentiation
10
Multi-Layer Neuron Model
11
Neural network methodology advantages

• Flexibility in description of the arbitrary
  non-linear dependencies
• Easy incorporation of the additional and renewal
  feedbacks
• Fast self-learning feature
• High approximation fit to measurement data
  achieved by NN model

12
(No Transcript)
13
March monthly SAT for St. Petersburg observed
and explained by model (NAO, AO)
14
March monthly SAT for St. Petersburg observed
and explained by model (AO, NAO , EA, EAJ, S,
EAWR )
15
March, 1996
16
May, 1996
17
May, 1998
18
Conclusions

• First attempt to develop an intellectual climate
  model has been carried out
• Its principal distinctive features are as
  following
• Self-learning ability
• Accumulation of all past observing information
  with changing inter-parameter links
• Adaptation of model feedbacks to changes in
  observing samples and its trends