Climate Modeling

And Numerical Weather Prediction (NWP)

Neil Conway

CISC 497

Modelling

• One way to understand the behaviour of a complex process: construct a model
  • State represented as a mathematical object
  • System of laws that describe the future state of the process as a function of its current state
  • Typically, computational and scientific limitations require the model to be a simplified, idealized version of the actual process
• We can predict the future state of the system by observing its current state and iteratively applying the rules

Numerical Weather Prediction (NWP)

• NWP has been a goal of meteorology since the beginning of the 20th century
• Infeasible until the 1950s, when computers and better observational techniques were introduced
• Modern meteorology largely relies on computer modeling
  • The role of the analyst is to construct the model, input the observations, and make manual adjustments to account for imperfections in the model

NWP Models

• The atmosphere is modelled as a 3-D grid
  • The smaller the cells of the grid, the more accurate the model and the more detailed its predictions, but the more computational resources are required
  • Variable grid sizes allow for better accuracy in specific areas
• The evolution of the system over time is governed by the laws of fluid dynamics
  • Nonlinear differential equations (Navier-Stokes)
  • Advance the system in time steps, record predicted states

Types of Models

• Regional and global models exist
  • Regional models have smaller grid sizes and time steps, and make more detailed predictions over a shorter time frame
  • Typical grid sizes: 2km for regional models, 100km or more for global models
• Many different models of varying sophistication and computation time
• Specialized models for particular applications

Complexity of Models

• All models are to some degree a simplification of the real world:
  • Lack of computational resources: must choose a relatively large grid size
    • Therefore, processes within a grid cannot be simulated directly
  • Incomplete understanding of the physical processes themselves
  • Absent or poor quality observational data

Parameterization

• A partial solution is to parameterize the model: replace processes that cannot be simulated in the model directly with simplified representations
• For example, in a global model with a 100km grid size, individual clouds typically not representable
  • Particularly at the fine level of detail needed to convection within clouds via fluid dynamics
• Therefore, parameterize the model with a simplified process that approximates cloud behaviour
  • In sophisticated models, parameters are often complex models in their own right

Applications of NWP

• Classical weather forecasting (National Weather Service, Meteorological Service of Canada)
• Disaster prevention and prediction
  • Firefighting
  • Hurricane and cyclone modelling
• 1996 Atlantic Olympic Games
• Military & business applications

Climate Modelling

• Same fundamental technique as NWP, with some notable differences:
  • Much longer time scale
    • Many processes that can be safely ignored in NWP must be considered in climate modelling
  • Relatively little historical data, much of which is imprecise or incomplete
    • Satellite temperature measurements of the troposphere only possible since 1979
  • Dependent on human behaviour
    • Such as rate of greenhouse gas emissions
  • Interested in answering different questions, such as climate sensitivity:
    • "If the amount of carbon-dioxide in the atmosphere were to double, how would this affect global climate?"

Hard Problem

• Climate modelling is a fundamentally harder problem than NWP
• Also, the interesting questions in climate modelling are more difficult to answer
• Therefore, the predictions it makes are more uncertain, less precise, and subject to more controversy

Important Questions

1. Is the globe warming, and if so, at what rate? (Detection)
2. What factors are responsible for the warming? (Attribution)
   • Climate is known to vary over time, so the question is really whether observed change is consistent with historical change
3. How will the climate change in the future? (Prediction)

Current Consensus

• Detection is essentially a solved problem
  • Global warming over the 20th century estimated at 0.6° ± 0.2° C
• Increasingly, a consensus has been reached on the attribution problem as well
  • "Most" of this warming can be attributed to man-made actions, such as the increase in greenhouse gas omissions
• This consensus is relatively recent
• Prediction is naturally the most difficult and contentious problem
  • Current estimate: 1.4-5.8° C increase in global mean temperature
• Climate sensitivity estimated at 2-4.5° C

Supercomputing

• Climate modelling is amongst the most computationally-difficult tasks in modern science
  • The smaller the grid size, the more computational resources required
• Rapid pace of improvement: faster hardware, more accurate models
• Example: NEC Earth Simulator, a supercomputer intended mostly for climate modelling and weather simulation
• Why does more supercomputing capacity matter?
  • Doesn't it just mean the same results, faster?

Cartesian Reductionism

• Reductionism: the assumption that the behaviour of a complex system can be reduced to the behaviour of its components
• This allows us to model complex systems in pieces:
  • Divide the system into components
  • Model the behaviour of each component separately
  • Combine the models to describe the behaviour of the entire system
• Reductionism has been responsible for much scientific progress
  • However, increasingly seen as a barrier to the understanding of certain complex phenomena

Holistic Simulation

Thank You