Weather ForeCast

Using computer science to solve atmospheric science problems.

Index

DifficultY

What makes weather forecasting such a hard task?

  • Deterministic Systems

    • Sensitive to initial conditions

    • Topologically transitive

    • Dense periodic orbits

Chaos 

Chaos 

  • Deterministic Systems

    • Sensitive to initial conditions

    • Topologically transitive

    • Dense periodic orbits

  • Weather is a chaotic dynamic system

When the present determines the future, but the approximate present does not approximately determine the future.

– Edward Lorenz

  • We may predict the weather if we have 100% accuracy

  • We simply can't

    • The world is too big

    • People move

    • Uncertainty principle

data measuring

  • Lyapunov time

  • Difference grows exponentially

  • Make forecasts that have acceptable, tiny errors

What can we do?

our method

How do we predict the weather?

History of weather prediction

How did we get here?

1904

Vilhelm Bjerknes formulated the primitive equations

1922

Lewis Fry Richardson tried to produce a forecast by hand

1950

The ENIAC was used to approximate the weather with simplified equations

1960s

Operational forecasts based on primitive equation models were made

1956

Norman Phillips developed the first successful climate model

1970s

 Tracks of tropical cyclones and air quality

can be better predicted

1980s

Soil and vegetation were accounted for

1990s

 Model ensemble forecasts have been used

present

The accuracy of forecasts can be acceptable for up to 14 days

Primitive equations

  • A set of non linear partial differential equations

    • A continuity equation

    • Conservation of momentum

    • A thermal energy equation

  • Fundamentals of many atmospheric models

  • First general-purpose computer 

    • Turing complete

    • Programmable

  • $487,000 USD($5,900,000 USD now)

  • 60 cubic metres

eniac

Ensemble forecasting

  • The more, the merrier
  • Monte Carlo method
    • Randomness
    • Used in many physical simulations
    • High hardware requirements
  • Uncertainties
    • Observational error
    • Imperfect model
  • Any dynamical system forecast

supercomputers

They compute, perhaps faster than you can imagine

performance measuring

  • FLOPS(floating point operations per second)
  • IPS(instructions per second)
10^{17} \footnotesize FLOPS
10^{12} \footnotesize FLOPS

What else?

Limitations

speed

Data can't be transmissed faster than the speed of light.

heat

The generated heat could damage the machine.

power

It's very energy intensive.

solutions

speed

Circular design to minimise route.

heat

Water cooling and other liquids.

power

Lower the temperature to increase performance.

Weather forecasts in taiwan

Central Weather Bureau

supercomputer(Gen 5)

  • Made by Fujitsu
  • around 1.46 PFLOPS
  • Replaced with a new one with 10 PFLOPS
  • Resolution: 3 km -> 1 km

  • Time: 7 days -> 10-14 days

accuracy

  • Run many simulation

  • Different initial values

ensemble system

End of presentation

Thank you for listening

questions

Are you ready for the suffering in the finals?

p1: Which one isn't turing complete?

  1. Lambda calculus
  2. ENIAC
  3. C++
  4. Primitive equations

p2-1

Give an example of Monte Carlo method.

p2-2

What is the spirit of Monte Carlo method? (1 word)

p3: Which one isn't in the primitive equations?

(1): \oint_{\partial \Sigma} \mathbf{E} \cdot \mathrm{d}\boldsymbol{\ell} = - \frac{\mathrm{d}}{\mathrm{d}t} \iint_{\Sigma} \mathbf{B} \cdot \mathrm{d}\mathbf{S}
(2): 0 = -\frac{\partial \Phi}{\partial p} - \frac{RT}{p}
(3):\frac{\partial u}{\partial x} + \frac{\partial v}{\partial y} + \frac{\partial w}{\partial p} = 0
(4):\frac{\partial T}{\partial t} = \frac{\partial T}{\partial t} + u \frac{\partial T}{\partial x} + v \frac{\partial T}{\partial y} + w \frac{\partial T}{\partial z}

DW

Q4: spell the name of the father of computer science 

DH

BC

CSxAS

By BrineTaiwan

CSxAS

It'd be great to turn in two homeworks with one presentation

  • 176