Should climate velocities be dynamic?

[rBatt]

It might be possible to take trends over a 5 year rolling window, or to do some sort of spline to get local slopes.

I think that this would have the potential to produce very different results.

It would also minimize the impact of arbitrary start and end dates (because they have so much leverage when the whole time series has 1 slope/ climate velocity).

Then again, doing this could end up reading too much into the noise.

[rBatt]

Identifying periods of change in time series with GAMs (or more specifically, Additive modelling and the HadCRUT3v global mean temperature series, which is referred to in the post I linked).

Seems like a reasonable way to get a dynamic temporal slope. Although this could produce really different results for climate velocity trajectories depending on the number of "knots" (i'm not too familiar with GAMs right now, but I think this is the right word).

Not an urgent matter, but @mpinsky and @JWMorley might have thoughts.

[mpinsky]

Some quick thoughts:

• Yes, GAMs seem like a reasonable way to approach this. GAMs typically use a model-selection approach to choose the optimal number of knots given the data.
• That said, this strikes me as unlikely to make a real difference in the results. Unless you see a clear sign that it should be important?

On Sun, Sep 20, 2015 at 3:04 PM, Ryan Batt notifications@github.com wrote:

Identifying periods of change in time series with GAMs http://www.fromthebottomoftheheap.net/2014/05/15/identifying-periods-of-change-with-gams/ (or more specifically, Additive modelling and the HadCRUT3v global mean temperature series http://www.fromthebottomoftheheap.net/2011/06/12/additive-modelling-and-the-hadcrut3v-global-mean-temperature-series/, which is referred to in the post I linked).

Seems like a reasonable way to get a dynamic temporal slope. Although this could produce really different results for climate velocity trajectories depending on the number of "knots" (i'm not too familiar with GAMs right now, but I think this is the right word).

Not an urgent matter, but @mpinsky https://github.com/mpinsky and @JWMorley https://github.com/JWMorley might have thoughts.

— Reply to this email directly or view it on GitHub https://github.com/rBatt/trawl/issues/24#issuecomment-141821440.

[rBatt]

OK. Any place where the temperature see-saws it could matter. I don't have a figure handy, but imagine a temperature time series that looks like a sine wave, or triangle, etc.

On Tue, Sep 22, 2015 at 9:08 PM, mpinsky notifications@github.com wrote:

Some quick thoughts:

• Yes, GAMs seem like a reasonable way to approach this. GAMs typically use a model-selection approach to choose the optimal number of knots given the data.
• That said, this strikes me as unlikely to make a real difference in the results. Unless you see a clear sign that it should be important?

On Sun, Sep 20, 2015 at 3:04 PM, Ryan Batt notifications@github.com wrote:

Identifying periods of change in time series with GAMs < http://www.fromthebottomoftheheap.net/2014/05/15/identifying-periods-of-change-with-gams/

(or more specifically, Additive modelling and the HadCRUT3v global mean temperature series < http://www.fromthebottomoftheheap.net/2011/06/12/additive-modelling-and-the-hadcrut3v-global-mean-temperature-series/ , which is referred to in the post I linked).

Seems like a reasonable way to get a dynamic temporal slope. Although this could produce really different results for climate velocity trajectories depending on the number of "knots" (i'm not too familiar with GAMs right now, but I think this is the right word).

Not an urgent matter, but @mpinsky https://github.com/mpinsky and @JWMorley https://github.com/JWMorley might have thoughts.

— Reply to this email directly or view it on GitHub https://github.com/rBatt/trawl/issues/24#issuecomment-141821440.

— Reply to this email directly or view it on GitHub https://github.com/rBatt/trawl/issues/24#issuecomment-142466535.

[rBatt]

Possibly, but this isn't a top-priority anymore. Also, could be hard. I don't think our perspective on this topic will change much soon. Worth keeping in mind for future studies.