4 ms·

It’s complex and chaotic, sure. But as you increase time horizons, and area of simulation larger and larger, that complexity and chaos starts to average out. So

It’s complex and chaotic, sure. But as you increase time horizons, and area of simulation larger and larger, that complexity and chaos starts to average out. So you can’t take a singular datapoint from a climate model, and make any concrete statements, but you can take thousands to millions of them and start making probabilistic statements. Additionally it’s not like these models are untested. Their predictions are continually compared to actual reality, including by taking know data, and asking the model to predict more known data. Pretty basic stuff for testing the capability of any model, and guess what, that data shows these models are pretty darn good. Everything in this world is complex and chaotic, how cars behave in collisions is complex and chaotic. But no one doubts our ability to use computer models to design safer cars. Equally most people generally trust the one week weather forecast, despite the fact that it’s substantially more difficult to predict weather than it is to predict climate.

fasteddie31003 5y ago
Can you give me another example of a chaotic system that can be correctly forecasted years out?
- dTal 5y ago
  Sure. Pick a spot offshore and measure the water depth, to the centimeter. Moment to moment, the precise value varies chaotically due to countless transient phenomena, from the direction and strength of the wind to the heading of the last boat to come anywhere near it. However the tides can be forecast years out. Just because weather is chaotic and unpredictable doesn't mean that climate is.
  - hammock 5y ago
    >Sure. Pick a spot offshore and measure the water depth, to the centimeter. Moment to moment, the precise value varies chaotically due to countless transient phenomena, from the direction and strength of the wind to the heading of the last boat to come anywhere near it. That's a good analogy of my skepticism for AGW. We are saying temps went up 1c over 100 years, taking daily measurements, and comparing them to ice cores which are our measurements of historic temps... Only those ice cores aren't daily- they're thousands of years apart. So the ice cores data is irreparably smoothed in a way the recent data is not.
    - avianlyric 5y ago
      Ice core measurements give you a value for each year. You’re looking for changes in the ice that are driven by yearly seasonal changes, notably the periods of summer where the ice received 24hour of sun for weeks. It’s perfectly reasonable to take daily measurements, average them into yearly measurements. Then compare them to yearly averages taken from ice cores. The only reason we can even use ice cores for temperature measurements is because they’re have high enough resolution that we’ve been able to observe the impact of global temperature on the most recent layers of ice. It’s not like someone drilled an ice core one day and discovered a bunch of temperatures sharpied down the side. They drilled many ice cores, then spent months correlating observed patterns to temperature records so they could build a model that allows them to understand the link between temperature and observed patterns in the ice. Don’t mistake a record measuring thousands of years of history for a record that only records every thousand years.
      - hammock 5y ago
        That makes sense and forgive me if I'm ignorant on the subject, willing to learn. What I am talking about is charts like this from the NOAA: http://www.climate.gov/media/5147 http://www.climate.gov/media/5147 Thes resolution is yearly going back only about 2000 years. Everything beyond that, the resolution is much lower. Where is that data from and don't you see how much "smoother" it is?
        avianlyric 5y ago
        Your link has a citation leading to the source article. I would recommend you read it if you’re interested (use Scihub if you need to). The thrust of the underlying article is looking at the exact question you’re asking, can ancient ice records that have a more limited resolution of 20-500 years be compared to centennial or millennial scale data, and can you draw useful conclusions about time periods of that range.
        hammock 5y ago
        Thank you for pointing that out, it was a fascinating read. I did observe that they said "essen-tially no variability preserved at periods shorter than 300 years" which seems to imply that drawing a conclusion between their older data and the recent data could be perilous when looking at periods shorter than that (e.g. last 100 years)
- hpcjoe 5y ago
  Planetary orbits. Stability is important there. But 3 and greater body problem in Newtonian mechanics is inherently chaotic. Add in spin-orbit coupling, relativity and other effects, and its a real mess, with no exact closed form solutions. That said, we can predict, with some level of accuracy the stability of of orbits in a number of cases. Approximate solutions can be managed with various levels of theory, perturbations, and numerical simulation.
- brutusborn 5y ago
  Calculation of the future locations of astronomical bodies? The difference is, Astronomical models can be simplified significantly, whereas climate models must be high fidelity to produce accurate results. To me the critical question is at what fidelity are the models useful for our (humanities) purposes? If our purpose is to predict climate and its impact we will need at least +/- 0.1C at a resolution of 100km at a timestep of 1 month. It must be possible to do mesh refinements to determine how accuracy changes with changing resolution, and then combine that with the prior resolution conditions to choose an appropriate mesh. As far as I know most modern climate models are accurate to within a degree for at least a 30 year range. I haven't looked at the mesh sizes for a while but I'm sure someone has done something similar.
- avianlyric 5y ago
  Just about any basic physical system. If you zoom into our universe enough, down to the quantum level, you’ll find completely unpredictable chaos and complexity. But just because the quantum world is pretty much unpredictable doesn’t mean the macro world is unpredictable. A climate model can’t you whether or not is going to rain in exactly 300 years time, but it can tell you how much rain will fall during an entire year, 300 years from now. Climate modelling is macro scale modelling to weathers quantum scale modelling. You’re operating at entirely different scales, and further you zoom out, the more you can ignore the minutia and still produce useful predictions. Additionally taking about absolute time periods is the wrong approach. The real question is now many time steps can your model take before it starts to significantly diverge from reality. A car collision simulation will be operating with time steps measured in milliseconds to nanoseconds, a climate model in time steps measured in days to weeks. Both of these models will probably be predicting a similar number of time steps into the future, and both will have a similar number of quantised simulation areas. For the car you’re simplifying cubes measured in millimetres, for climate, zones measured in km. Again, climate models have been measured, tested and repeatedly proven their predictive power. If empirical results aren’t enough to convince you, then what is?
8note 5y ago
Generally the defining feature of non linear mechanisms is that the long term behaviour is non-intuitive. I'd expect that to include not having a predictable mean
- godelski 5y ago
  Non intuitive doesn't mean not predictable. Nor does chaos. Chaos just means that a small perturbation in input parameters can result in a large change in the output. So to have good predictive power you need to pick good inputs with good accuracy. But chaos is also a spectrum so not all chaotic systems are as sensitive as others.
  - avianlyric 5y ago
    Chaos is also different at different scales. At certain point chaos can be simulated as noise with a known distribution. You can’t make predictions within the noise, but you can describe the shape of the noise and it’s impact on macro scale systems. Perfect example of this is quantum effects. Zoom in far enough and quantum effects make the world entirely unpredictable, and yet I can still accurately predict where I throw is going to land with no knowledge of quantum effects. That doesn’t mean those effects don’t influence the ball, it just that noisy chaos of those influences is tightly bound enough that it doesn’t matter. Newtonian physics is enough to make a prediction, despite being an incomplete description of the physical system at play.
    - godelski 5y ago
      But different noises do different things and have different predictive powers. I think your statements about these systems being impossible to calculate are too strong. We use gaussian noise all the time. In fact there's plenty of machine learning models that use noise and change of variables (we learn the transforms) to represent different distributions. Ones that do actually have good clustering. We can be predictive in these domains. Chaos and randomness don't make inference impossible. This is the 21st century, we're not beholden to Newton nor to deterministic processes.

fasteddie31003 5y ago

Can you give me another example of a chaotic system that can be correctly forecasted years out?

dTal 5y ago
Sure. Pick a spot offshore and measure the water depth, to the centimeter. Moment to moment, the precise value varies chaotically due to countless transient phenomena, from the direction and strength of the wind to the heading of the last boat to come anywhere near it. However the tides can be forecast years out. Just because weather is chaotic and unpredictable doesn't mean that climate is.
- hammock 5y ago
  >Sure. Pick a spot offshore and measure the water depth, to the centimeter. Moment to moment, the precise value varies chaotically due to countless transient phenomena, from the direction and strength of the wind to the heading of the last boat to come anywhere near it. That's a good analogy of my skepticism for AGW. We are saying temps went up 1c over 100 years, taking daily measurements, and comparing them to ice cores which are our measurements of historic temps... Only those ice cores aren't daily- they're thousands of years apart. So the ice cores data is irreparably smoothed in a way the recent data is not.
  - avianlyric 5y ago
    Ice core measurements give you a value for each year. You’re looking for changes in the ice that are driven by yearly seasonal changes, notably the periods of summer where the ice received 24hour of sun for weeks. It’s perfectly reasonable to take daily measurements, average them into yearly measurements. Then compare them to yearly averages taken from ice cores. The only reason we can even use ice cores for temperature measurements is because they’re have high enough resolution that we’ve been able to observe the impact of global temperature on the most recent layers of ice. It’s not like someone drilled an ice core one day and discovered a bunch of temperatures sharpied down the side. They drilled many ice cores, then spent months correlating observed patterns to temperature records so they could build a model that allows them to understand the link between temperature and observed patterns in the ice. Don’t mistake a record measuring thousands of years of history for a record that only records every thousand years.
    - hammock 5y ago
      That makes sense and forgive me if I'm ignorant on the subject, willing to learn. What I am talking about is charts like this from the NOAA: http://www.climate.gov/media/5147 http://www.climate.gov/media/5147 Thes resolution is yearly going back only about 2000 years. Everything beyond that, the resolution is much lower. Where is that data from and don't you see how much "smoother" it is?
      - avianlyric 5y ago
        Your link has a citation leading to the source article. I would recommend you read it if you’re interested (use Scihub if you need to). The thrust of the underlying article is looking at the exact question you’re asking, can ancient ice records that have a more limited resolution of 20-500 years be compared to centennial or millennial scale data, and can you draw useful conclusions about time periods of that range.
        hammock 5y ago
        Thank you for pointing that out, it was a fascinating read. I did observe that they said "essen-tially no variability preserved at periods shorter than 300 years" which seems to imply that drawing a conclusion between their older data and the recent data could be perilous when looking at periods shorter than that (e.g. last 100 years)
hpcjoe 5y ago
Planetary orbits. Stability is important there. But 3 and greater body problem in Newtonian mechanics is inherently chaotic. Add in spin-orbit coupling, relativity and other effects, and its a real mess, with no exact closed form solutions. That said, we can predict, with some level of accuracy the stability of of orbits in a number of cases. Approximate solutions can be managed with various levels of theory, perturbations, and numerical simulation.
brutusborn 5y ago
Calculation of the future locations of astronomical bodies? The difference is, Astronomical models can be simplified significantly, whereas climate models must be high fidelity to produce accurate results. To me the critical question is at what fidelity are the models useful for our (humanities) purposes? If our purpose is to predict climate and its impact we will need at least +/- 0.1C at a resolution of 100km at a timestep of 1 month. It must be possible to do mesh refinements to determine how accuracy changes with changing resolution, and then combine that with the prior resolution conditions to choose an appropriate mesh. As far as I know most modern climate models are accurate to within a degree for at least a 30 year range. I haven't looked at the mesh sizes for a while but I'm sure someone has done something similar.
avianlyric 5y ago
Just about any basic physical system. If you zoom into our universe enough, down to the quantum level, you’ll find completely unpredictable chaos and complexity. But just because the quantum world is pretty much unpredictable doesn’t mean the macro world is unpredictable. A climate model can’t you whether or not is going to rain in exactly 300 years time, but it can tell you how much rain will fall during an entire year, 300 years from now. Climate modelling is macro scale modelling to weathers quantum scale modelling. You’re operating at entirely different scales, and further you zoom out, the more you can ignore the minutia and still produce useful predictions. Additionally taking about absolute time periods is the wrong approach. The real question is now many time steps can your model take before it starts to significantly diverge from reality. A car collision simulation will be operating with time steps measured in milliseconds to nanoseconds, a climate model in time steps measured in days to weeks. Both of these models will probably be predicting a similar number of time steps into the future, and both will have a similar number of quantised simulation areas. For the car you’re simplifying cubes measured in millimetres, for climate, zones measured in km. Again, climate models have been measured, tested and repeatedly proven their predictive power. If empirical results aren’t enough to convince you, then what is?

8note 5y ago

Generally the defining feature of non linear mechanisms is that the long term behaviour is non-intuitive. I'd expect that to include not having a predictable mean

godelski 5y ago
Non intuitive doesn't mean not predictable. Nor does chaos. Chaos just means that a small perturbation in input parameters can result in a large change in the output. So to have good predictive power you need to pick good inputs with good accuracy. But chaos is also a spectrum so not all chaotic systems are as sensitive as others.
- avianlyric 5y ago
  Chaos is also different at different scales. At certain point chaos can be simulated as noise with a known distribution. You can’t make predictions within the noise, but you can describe the shape of the noise and it’s impact on macro scale systems. Perfect example of this is quantum effects. Zoom in far enough and quantum effects make the world entirely unpredictable, and yet I can still accurately predict where I throw is going to land with no knowledge of quantum effects. That doesn’t mean those effects don’t influence the ball, it just that noisy chaos of those influences is tightly bound enough that it doesn’t matter. Newtonian physics is enough to make a prediction, despite being an incomplete description of the physical system at play.
  - godelski 5y ago
    But different noises do different things and have different predictive powers. I think your statements about these systems being impossible to calculate are too strong. We use gaussian noise all the time. In fact there's plenty of machine learning models that use noise and change of variables (we learn the transforms) to represent different distributions. Ones that do actually have good clustering. We can be predictive in these domains. Chaos and randomness don't make inference impossible. This is the 21st century, we're not beholden to Newton nor to deterministic processes.