#### Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Options

# Paper - Yamasaki et al - Climate networks around the globe are significantly effected by El Niño

edited February 2 in General

Here we are analyzing:

EDIT: I removed some stuff, and renamed this thread.

• Options
1.

This is great so far. I have lots of criticisms of the work done so far, and I'm struggling to figure out how to do things better. But it's good to spend time understanding what's been done!

Comment Source:This is great so far. I have lots of criticisms of the work done so far, and I'm struggling to figure out how to do things better. But it's good to spend time understanding what's been done!
• Options
2.

In particular, K. Yamasaki, A. Gozolchiani and S. Havlin seem to think El Niños break "climate links", but I'm afraid this is due to a misleading definition of those climate links.

Comment Source:In particular, K. Yamasaki, A. Gozolchiani and S. Havlin seem to think El Ni&ntilde;os break "climate links", but I'm afraid this is due to a misleading definition of those climate links.
• Options
3.

Namely, it seems that climate links in the Pacific Ocean are getting stronger and, by virtue of a definition that says only the strongest links count as links, they claim links elsewhere in the world are getting broken.

Comment Source:Namely, it seems that climate links in the Pacific Ocean are getting stronger and, _by virtue of a definition that says only the strongest links count as links_, they claim links elsewhere in the world are getting broken.
• Options
4.
edited June 2014

John wrote:

In particular, K. Yamasaki, A. Gozolchiani and S. Havlin seem to think El Niños break “climate links”, but I’m afraid this is due to a misleading definition of those climate links.

I guess you'll expand on that shortly.

I'm having difficulties with quite a few of their formulations but they are non-native english speakers (/affect/effect/)so maybe it's that rather than my ignorance.

Comment Source:John wrote: > In particular, K. Yamasaki, A. Gozolchiani and S. Havlin seem to think El Niños break “climate links”, but I’m afraid this is due to a misleading definition of those climate links. I guess you'll expand on that shortly. I'm having difficulties with quite a few of their formulations but they are non-native english speakers (/affect/effect/)so maybe it's that rather than my ignorance.
• Options
5.
edited June 2014

There are various things to say, Jim, but mainly:

1) It probably makes more sense to focus on this paper:

It seems to have a simpler and more reasonable methodology than this earlier one:

2) I have the strong feeling that people in this subject - "climate network theory" - are just messing around trying to see what works. The definitions of various quantities are a bit ad hoc, not justified by deep reasoning. This is good because it means we can mess around a bit ourselves, and try to do something that works better. But it means we can't just learn the stuff by reading papers and trusting them.

Comment Source:There are various things to say, Jim, but mainly: 1) It probably makes more sense to focus on this paper: * Josef Ludescher, Avi Gozolchiani, Mikhail I. Bogachev, Armin Bunde, Shlomo Havlin, and Hans Joachim Schellnhuber, [Improved El Niño forecasting by cooperativity detection](http://www.pnas.org/content/early/2013/06/26/1309353110.full.pdf+html), _Proceedings of the National Academy of Sciences_, 30 May 2013. It seems to have a simpler and more reasonable methodology than this earlier one: * K. Yamasaki, A. Gozolchiani, and S. Havlin, [Climate networks around the globe are significantly effected by El Niño](http://arxiv.org/abs/0804.1374), April 2008. 2) I have the strong feeling that people in this subject - "climate network theory" - are just messing around trying to see what works. The definitions of various quantities are a bit _ad hoc_, not justified by deep reasoning. This is good because it means we can mess around a bit ourselves, and try to do something that works better. But it means we can't just learn the stuff by reading papers and trusting them.
• Options
6.
edited June 2014

John wrote:

In particular, K. Yamasaki, A. Gozolchiani and S. Havlin seem to think El Niños break “climate links”, but I’m afraid this is due to a misleading definition of those climate links.

Do you think that it is an interesting question to look at -- the effect of El Niños on the connectivity structure of the graph -- using a more appropriate definition of the climate links?

I had trouble parsing the definitions in this paper, which at times seemed ill-defined. But maybe they are basically well defined, and the trouble is that I don't have the context to supply the missing words. For instance, they didn't define their angle brackets, and why is the year a parameter of the cross-correlation function.

John, leaving aside the fact that you think the definitions were not appropriate, were you able to parse them, in detail, and did you find them to be fundamentally well-defined?

Footnote: This theoretical discussion is not idle, because it can help to guide our experimental designs.

EDIT: I was having trouble parsing the paper because my hard-copy was mangling some of the symbols!

Comment Source:John wrote: > In particular, K. Yamasaki, A. Gozolchiani and S. Havlin seem to think El Niños break “climate links”, but I’m afraid this is due to a misleading definition of those climate links. Do you think that it is an interesting question to look at -- the effect of El Niños on the connectivity structure of the graph -- using a more appropriate definition of the climate links? I had trouble parsing the definitions in this paper, which at times seemed ill-defined. But maybe they are basically well defined, and the trouble is that I don't have the context to supply the missing words. For instance, they didn't define their angle brackets, and why is the year a parameter of the cross-correlation function. John, leaving aside the fact that you think the definitions were not appropriate, were you able to parse them, in detail, and did you find them to be fundamentally well-defined? Footnote: This theoretical discussion is not idle, because it can help to guide our experimental designs. EDIT: I was having trouble parsing the paper because my hard-copy was mangling some of the symbols!
• Options
7.
edited June 2014

John, leaving aside the fact that you think the definitions were not appropriate, were you able to parse them, in detail, and did you find them to be fundamentally well-defined?

They seemed well-defined to me.

They are using abbreviations and shorthands the way physicists and many other scientists do. For example, when they have a function of several variables, they don't bother to write the variables that they are uninterested in at that moment. The idea is that you're supposed to figure out what the notation means by thinking about what would be the only reasonable thing for it to mean. The idea is that writing things out in a completely unambiguous way makes for big ugly expressions which they'd rather avoid.

Since this seems to drive computer programmers crazy, I will modify their notation a bit.

For instance, they didn’t define their angle brackets...

Angle brackets are a standard symbol for an arithmetic mean, or average. So is an overline or a capital letter $E$.

So, here is what they are doing. They start with a bunch of temperatures $\tilde{T}(\ell,y,d)$: one for each grid point $\ell$, year $y$ ($1 \le y \le N$) and day of the year $d$ ($1 \le d \le 365$). Then they consider

$$T(\ell,r,y,d) := \tilde{T}(\ell,y,d) - \frac{1}{N} \sum_{y = 1}^N \tilde{T}(\ell,y,d)$$ This tells you how much "hotter it is today than it usually is here at this time of year".

Then, for two grid points $\ell$ and $r$ and a year $y$ they work out a kind of correlation between $\tilde{T}(\ell,y,d)$ and $\tilde{T}(\ell,y,d+\tau)$ where $\tau = 1,2,3,\dots$) :

$$X(\ell,r,y,\tau) = \frac{1}{365} \sum_{d = 1}^{365} T(\ell,y,d) T(r,y,d+\tau)$$ I assume that when $d+\tau$ exceeds 365 you go over to the next year, since that's the reasonable thing to do.

The point is that $X(\ell,y, \tau)$ tells you how much the temperature at grid point $\ell$ is correlated to the temperature at grid point $r$, $\tau$ days later, during year $y$.

Does this make sense so far?

So far this is what anyone would be tempted to do when studying how a randomly fluctuating quantity here and now is correlated to the same quantity somewhere else and a bit later. The fun part starts later.

Comment Source:> John, leaving aside the fact that you think the definitions were not appropriate, were you able to parse them, in detail, and did you find them to be fundamentally well-defined? They seemed well-defined to me. They are using abbreviations and shorthands the way physicists and many other scientists do. For example, when they have a function of several variables, they don't bother to write the variables that they are uninterested in at that moment. The idea is that you're supposed to figure out what the notation means by thinking about what would be the only reasonable thing for it to mean. The idea is that writing things out in a completely unambiguous way makes for big ugly expressions which they'd rather avoid. Since this seems to drive computer programmers crazy, I will modify their notation a bit. > For instance, they didn’t define their angle brackets... Angle brackets are a standard symbol for an arithmetic mean, or average. So is an overline or a capital letter $E$. So, here is what they are doing. They start with a bunch of temperatures $\tilde{T}(\ell,y,d)$: one for each grid point $\ell$, year $y$ ($1 \le y \le N$) and day of the year $d$ ($1 \le d \le 365$). Then they consider $$T(\ell,r,y,d) := \tilde{T}(\ell,y,d) - \frac{1}{N} \sum_{y = 1}^N \tilde{T}(\ell,y,d)$$ This tells you how much "hotter it is today than it usually is here at this time of year". Then, for two grid points $\ell$ and $r$ and a year $y$ they work out a kind of correlation between $\tilde{T}(\ell,y,d)$ and $\tilde{T}(\ell,y,d+\tau)$ where $\tau = 1,2,3,\dots$) : $$X(\ell,r,y,\tau) = \frac{1}{365} \sum_{d = 1}^{365} T(\ell,y,d) T(r,y,d+\tau)$$ I assume that when $d+\tau$ exceeds 365 you go over to the next year, since that's the reasonable thing to do. The point is that $X(\ell,y, \tau)$ tells you how much the temperature at grid point $\ell$ is correlated to the temperature at grid point $r$, $\tau$ days later, during year $y$. Does this make sense so far? So far this is what anyone would be tempted to do when studying how a randomly fluctuating quantity _here and now_ is correlated to the same quantity _somewhere else and a bit later_. The fun part starts later.
• Options
8.
edited June 2014

Yes, very clear. Can we keep proceeding with this discussion, at this level of clarity -- which is more explicit and less contextual. By the end of the discussion we will have a clear recipe for writing programs.

Comment Source:Yes, very clear. Can we keep proceeding with this discussion, at this level of clarity -- which is more explicit and less contextual. By the end of the discussion we will have a clear recipe for writing programs.
• Options
9.
edited June 2014

Then they give a formula for negative $\tau$:

$$X(\ell,r,y,-\tau) = X(r,\ell,y,\tau)$$

Comment Source:Then they give a formula for negative $\tau$: $$X(\ell,r,y,-\tau) = X(r,\ell,y,\tau)$$
• Options
10.
edited June 2014

Proceeding with their development...

Next, define:

$$Max(\ell,r,y) = Max { |X(r,\ell,y,\tau)| : : \tau_{min} \leq \tau \leq \tau_{max} }$$ $$Std(\ell,r,y) = Std { |X(r,\ell,y,\tau)| : : \tau_{min} \leq \tau \leq \tau_{max} }$$ $$W(\ell,r,y) = Max(\ell,r,y) \, / \, Std(\ell,r,y)$$ $W(\ell,r,y)$ is called the correlation strength of the link between $\ell$ and $r$ during year $y$.

The value of $\tau$ at which $Max(\ell,r,y)$ is achieved is the time delay between points $\ell$ and $r$ during year $y$.

Comment Source:Proceeding with their development... Next, define: $$Max(\ell,r,y) = Max \{ |X(r,\ell,y,\tau)| : \: \tau_{min} \leq \tau \leq \tau_{max} \}$$ $$Std(\ell,r,y) = Std \{ |X(r,\ell,y,\tau)| : \: \tau_{min} \leq \tau \leq \tau_{max} \}$$ $$W(\ell,r,y) = Max(\ell,r,y) \, / \, Std(\ell,r,y)$$ $W(\ell,r,y)$ is called the _correlation strength_ of the link between $\ell$ and $r$ during year $y$. The value of $\tau$ at which $Max(\ell,r,y)$ is achieved is the _time delay_ between points $\ell$ and $r$ during year $y$.
• Options
11.
edited June 2014

EDIT: moved the analysis of Ludescher et. al to another thread.

Comment Source:EDIT: moved the analysis of Ludescher et. al to another thread.
• Options
12.
edited June 2014

EDIT: moved the analysis of Ludescher et. al to another thread.

Comment Source:EDIT: moved the analysis of Ludescher et. al to another thread.
• Options
13.
edited June 2014

Thanks for catching those typos, David. Almost everything you say looks fine to me!

But this formula

$$X(\ell,r,y,\tau) = \frac{1}{365} \sum_{d = 1}^{365} \tilde{T}(\ell,y,d) \tilde{T}(r,y,d+\tau)$$

still has a serious typo in it, due to me - there should not be tildes on those $T$s. Here's the right formula:

$$X(\ell,r,y,\tau) = \frac{1}{365} \sum_{d = 1}^{365} T(\ell,y,d) T(r,y,d+\tau)$$ The point is that it would be silly to define a correlation between temperatures $\tilde{T}$ using the top formula; we need to subtract off their means and use $T$s instead: we don't want the correlation to be big just because it's hot all the time!

I have taken the liberty of rewriting history and fixing my original comment, taking into account your typo fixes and also this.

In general people write $\langle X \rangle$ for the mean of any random variable $X$, so the standard deviation of $X$, say $Std(X)$ in your notation, is

$$Std(X)^2 = \langle (X - \langle X \rangle)(X - \langle X \rangle) \rangle$$ Given two random variables $X$ and $Y$ we compute their correlation by first working out

$$\langle (X - \langle X \rangle)(Y - \langle Y \rangle) \rangle$$ and then normalizing it by dividing by $Std(X) Std(Y)$: this makes the result lie between $-1$ and $1$.

I hope you see that

$$X(\ell,r,y,\tau) = \frac{1}{365} \sum_{d = 1}^{365} T(\ell,y,d) T(r,y,d+\tau)$$ is almost an example of this

$$\langle (X - \langle X \rangle)(Y - \langle Y \rangle) \rangle$$ concept: we've taken $\tilde{T}(\ell,y,d)$ and $\tilde{T}(r,y,d+\tau)$, subtracted off their mean (average over all years), multiplied them, and then taken the mean of that (but now averaging over days of a given year). In the paper it looks like they use $\langle \rangle_d$ and $\langle \rangle_y$ to mean averaging over days of a given year and averaging over years.

More to say, but it's my bedtime!

Comment Source:Thanks for catching those typos, David. Almost everything you say looks fine to me! But this formula > $$X(\ell,r,y,\tau) = \frac{1}{365} \sum_{d = 1}^{365} \tilde{T}(\ell,y,d) \tilde{T}(r,y,d+\tau)$$ still has a serious typo in it, due to me - there should not be tildes on those $T$s. Here's the right formula: $$X(\ell,r,y,\tau) = \frac{1}{365} \sum_{d = 1}^{365} T(\ell,y,d) T(r,y,d+\tau)$$ The point is that it would be silly to define a correlation between temperatures $\tilde{T}$ using the top formula; we need to subtract off their means and use $T$s instead: we don't want the correlation to be big just because it's hot all the time! I have taken the liberty of rewriting history and fixing my [original comment](http://forum.azimuthproject.org/discussion/1357/climate-networks/?Focus=10714#Comment_10714), taking into account your typo fixes and also this. In general people write $\langle X \rangle$ for the mean of any random variable $X$, so the standard deviation of $X$, say $Std(X)$ in your notation, is $$Std(X)^2 = \langle (X - \langle X \rangle)(X - \langle X \rangle) \rangle$$ Given two random variables $X$ and $Y$ we compute their correlation by first working out $$\langle (X - \langle X \rangle)(Y - \langle Y \rangle) \rangle$$ and then normalizing it by dividing by $Std(X) Std(Y)$: this makes the result lie between $-1$ and $1$. I hope you see that $$X(\ell,r,y,\tau) = \frac{1}{365} \sum_{d = 1}^{365} T(\ell,y,d) T(r,y,d+\tau)$$ is almost an example of this $$\langle (X - \langle X \rangle)(Y - \langle Y \rangle) \rangle$$ concept: we've taken $\tilde{T}(\ell,y,d)$ and $\tilde{T}(r,y,d+\tau)$, subtracted off their mean (average over all years), multiplied them, and then taken the mean of that (but now averaging over days of a given year). In the paper it looks like they use $\langle \rangle_d$ and $\langle \rangle_y$ to mean averaging over days of a given year and averaging over years. More to say, but it's my bedtime!
• Options
14.
edited June 2014

John, there is one consistent set of typos in your formulas for $X(\ell,r,y,\tau)$ -- you've left out the second parameter $r$. Can you edit them, so we can make this thread look nice.

Comment Source:John, there is one consistent set of typos in your formulas for $X(\ell,r,y,\tau)$ -- you've left out the second parameter $r$. Can you edit them, so we can make this thread look nice.
• Options
15.
edited June 2014

Next, they state the idea of a physical threshold $Q$ so that only pairs of nodes with a link strength greater than $Q$ will be regarded as significantly linked.

To this end, they define a new matrix, that takes on the value 1 whenever two nodes are significantly linked, otherwise it is 0:

$\rho(\ell,r,y) = \Theta(W(\ell,r,y) - Q)$

where $\Theta$ is the Heaviside function, mapping negative values to 0, positives values to 1 (and 0 to 0.5).

Comment Source:Next, they state the idea of a physical threshold $Q$ so that only pairs of nodes with a link strength greater than $Q$ will be regarded as significantly linked. To this end, they define a new matrix, that takes on the value 1 whenever two nodes are significantly linked, otherwise it is 0: $\rho(\ell,r,y) = \Theta(W(\ell,r,y) - Q)$ where $\Theta$ is the Heaviside function, mapping negative values to 0, positives values to 1 (and 0 to 0.5).
• Options
16.
edited June 2014

Then they observe that some of the elements of $\rho$ may "blink" as a function of $y$, i.e., and disappear.

They state that:

and they state the goal of distinguishing them from the more robust links.

Comment Source:Then they observe that some of the elements of $\rho$ may "blink" as a function of $y$, i.e., and disappear. They state that: > Blinking links seem to be a signature of structural changes and they state the goal of distinguishing them from the more robust links.
• Options
17.
edited June 2014

Then they introduce a new matrix $M$ which counts the number of times a link appeared before continuously (without a blink):

$$M(\ell,r,y) = \sum_{n = 0}^{y-1} \prod_{m = y-n}^{y} \rho(\ell,r,m)$$ I.e. it counts the length of the maximal run of 1's, going backwards from year $y$.

Comment Source:Then they introduce a new matrix $M$ which counts the number of times a link appeared before continuously (without a blink): $$M(\ell,r,y) = \sum_{n = 0}^{y-1} \prod_{m = y-n}^{y} \rho(\ell,r,m)$$ I.e. it counts the length of the maximal run of 1's, going backwards from year $y$.
• Options
18.
edited June 2014

Let $k$ be a parameter.

Define a link to be k-robust if $M(\ell,r,y) \ge k$.

Comment Source:Let $k$ be a parameter. Define a link to be _k-robust_ if $M(\ell,r,y) \ge k$.
• Options
19.
edited June 2014

Let $n_k(y)$ be the count of the number of k-robust links in year $y$:

$$n_k(y) = \sum_{\ell = 0}^{N} \sum_{r=l+1}^{N} \Theta(M(\ell,r,y) - k + 1)$$

Comment Source:Let $n_k(y)$ be the count of the number of k-robust links in year $y$: $$n_k(y) = \sum_{\ell = 0}^{N} \sum_{r=l+1}^{N} \Theta(M(\ell,r,y) - k + 1)$$
• Options
20.

Here are the parameters chosen by the authors:

• $k = 5$

• $Q = 2$

Comment Source:Here are the parameters chosen by the authors: * $k = 5$ * $Q = 2$
• Options
21.

Then they say that:

...we choose the $y$ resolution (the jumps between two subsequent dates represented by $y$) to be 50 days

Does this mean that they're using an arbitrary notion of what a year means, and that their "years" are not synched up with the calendar? I find this confusing, because their graphs refer to real, empirical years.

Comment Source:Then they say that: > ...we choose the $y$ resolution (the jumps between two subsequent dates represented by $y$) to be 50 days Does this mean that they're using an arbitrary notion of what a year means, and that their "years" are not synched up with the calendar? I find this confusing, because their graphs refer to real, empirical years.
• Options
22.
edited June 2014

Then they choose four representative zones around the globe.

For each zone they examine the evolution of $n_k(y)$ over time, and compare it with standard indices for El Niño -- the difference in sea level pressure between Tahiti and Darwin, and mean sea surface temperature in the standard basin.

Comment Source:Then they choose four representative zones around the globe. For each zone they examine the evolution of $n_k(y)$ over time, and compare it with standard indices for El Niño -- the difference in sea level pressure between Tahiti and Darwin, and mean sea surface temperature in the standard basin.
• Options
23.
edited June 2014

In general, their graphs show that $n_k(y)$ drops during times of El Niño, which leads them to their main conclusion that:

...the number of surviving links comprises a specific and sensitive measure for El Niño events.

Comment Source:In general, their graphs show that $n_k(y)$ drops during times of El Niño, which leads them to their main conclusion that: > ...the number of surviving links comprises a specific and sensitive measure for El Niño events.
• Options
24.
edited June 2014

They also claim that these results:

...are not sensitive to the choice of $k$. However, choosing too large $k$ values reduces the number of surviving links significantly, and therefore eliminates much of the effet. Choosing too small values of $k$, on the other hand, does not enable the elimination of blinking links, and therefore causes $n_k(y)$ to be more noisy, but the significant effect of breaking links is still evident.

Comment Source:They also claim that these results: > ...are not sensitive to the choice of $k$. However, choosing too large $k$ values reduces the number of surviving links significantly, and therefore eliminates much of the effet. Choosing too small values of $k$, on the other hand, does not enable the elimination of blinking links, and therefore causes $n_k(y)$ to be more noisy, but the significant effect of breaking links is still evident.
• Options
25.

Here is their justification for choosing the threshold $Q = 2$:

When observing the probability density function of $W$ it is clear that for non-El Niño time regimes, W = 2 is actually the minimal value that exists. It therefore appears that choosing this threshold makes the network very sensitive to El Niño events while remaining insensitive to other changes in climate. The reason is that the distribution of $W(\ell,r,y)$ tends to typical lower values of $W(\ell,r,y)$ during El Niño, as can clearly be seen in Fig. 3 (A-E). A remarkable property of this softening is that the lower limit of the distribution drops from being close to 2 to some significantly lower value. Changes in climate around the world due to El Niño events thus share a unified property of the correlation pattern, which can be tracked in a reliable way by the number of surviving links $n_k(y)$ in the climate network.

Comment Source:Here is their justification for choosing the threshold $Q = 2$: > When observing the probability density function of $W$ it is clear that for non-El Niño time regimes, W = 2 is actually the minimal value that exists. It therefore appears that choosing this threshold makes the network very sensitive to El Niño events while remaining insensitive to other changes in climate. The reason is that the distribution of $W(\ell,r,y)$ tends to typical lower values of $W(\ell,r,y)$ during El Niño, as can clearly be seen in Fig. 3 (A-E). A remarkable property of this softening is that the lower limit of the distribution drops from being close to 2 to some significantly lower value. Changes in climate around the world due to El Niño events thus share a unified property of the correlation pattern, which can be tracked in a reliable way by the number of surviving links $n_k(y)$ in the climate network.
• Options
26.
edited June 2014

Go, ye R progammers, go.

Comment Source:This article is rife with software-experimentally testable statements! Go, ye R progammers, go.
• Options
27.
edited June 2014

Now let's explore John's criticism:

Namely, it seems that climate links in the Pacific Ocean are getting stronger and, by virtue of a definition that says only the strongest links count as links, they claim links elsewhere in the world are getting broken.

They are claiming that links are getting broken during El Niños, so do you mean to say that the links in the Pacific Ocean get stronger during El Niños?

Comment Source:Now let's explore John's criticism: > Namely, it seems that climate links in the Pacific Ocean are getting stronger and, _by virtue of a definition that says only the strongest links count as links_, they claim links elsewhere in the world are getting broken. They are claiming that links are getting broken during El Niños, so do you mean to say that the links in the Pacific Ocean get stronger during El Niños?
• Options
28.

Ye R programmers, can you verify that climate links in the Pacific Ocean are getting stronger?

Comment Source:Ye R programmers, can you verify that climate links in the Pacific Ocean are getting stronger?
• Options
29.
edited June 2014

John, now that we have a terminology, can you explain your statement that with their definitions only the strongest links count as links.

I don't see any normalization of the link strengths taking place.

Comment Source:John, now that we have a terminology, can you explain your statement that with their definitions only the strongest links count as links. I don't see any normalization of the link strengths taking place.
• Options
30.
edited June 2014

The correlation strengths are always compared to a fixed number $Q$. These strengths are defined as a ratio of a Max value divided by a Std value, where the Max and Std is taken over a set of numbers that is per $\ell,r$ pair.

Comment Source:The correlation strengths are always compared to a fixed number $Q$. These strengths are defined as a ratio of a Max value divided by a Std value, where the Max and Std is taken over a set of numbers that is per $\ell,r$ pair.
• Options
31.

In any case, all of these claims provide great exercises for us to start working with the data and developing software to analyze it.

Comment Source:In any case, all of these claims provide great exercises for us to start working with the data and developing software to analyze it.
• Options
32.
edited June 2014

Ye R programmers, can you verify that climate links in the Pacific Ocean are getting stronger?

I can't see any indication of this in Fig 2 of

If anything, since about 2000, they've been weaker than typical.

Comment Source:> Ye R programmers, can you verify that climate links in the Pacific Ocean are getting stronger? I can't see any indication of this in Fig 2 of * Josef Ludescher, Avi Gozolchiani, Mikhail I. Bogachev, Armin Bunde, Shlomo Havlin, and Hans Joachim Schellnhuber, [Improved El Niño forecasting by cooperativity detection](http://www.pnas.org/content/early/2013/06/26/1309353110.full.pdf+html), _Proceedings of the National Academy of Sciences_, 30 May 2013. If anything, since about 2000, they've been weaker than typical. <img src = "http://math.ucr.edu/home/baez/ecological/ludescher_el_nino_cooperativity_2.jpg" alt = ""/>
• Options
33.

In my revised version of message 28, I asked John:

They are claiming that links are getting broken during El Niños, so do you mean to say that the links in the Pacific Ocean get stronger during El Niños?

Graham, do you see any any indication of this in the data?

Comment Source:In my revised version of message 28, I asked John: > They are claiming that links are getting broken during El Niños, so do you mean to say that the links in the Pacific Ocean get stronger during El Niños? Graham, do you see any any indication of this in the data?
• Options
34.
edited June 2014

Yes, again from Fig 2. (It will be some time before I could answer using my own analysis!) The red line (= signal strength ~=count of of links) do seem to go down during most El Niños.

Comment Source:Yes, again from Fig 2. (It will be some time before I could answer using my own analysis!) The red line (= signal strength ~=count of of links) do seem to go down during most El Niños. <img src = "http://math.ucr.edu/home/baez/ecological/ludescher_el_nino_cooperativity_2.jpg" alt = ""/>
• Options
35.

Hi Graham, yes I see that, which is the evidence that they are showing for their claim.

Sorry my question didn't make clear what I was really asking about, which is the second part of the sentence: is there evidence that the links in the Pacific Ocean get stronger (1) over time in general, and (2) during El Niños? I believe this would take our own analysis to answer, so we may have to build towards addressing such questions.

Comment Source:Hi Graham, yes I see that, which is the evidence that they are showing for their claim. Sorry my question didn't make clear what I was really asking about, which is the second part of the sentence: is there evidence that the links in the Pacific Ocean get stronger (1) over time in general, and (2) during El Niños? I believe this would take our own analysis to answer, so we may have to build towards addressing such questions.
• Options
36.

One more point to note about their methodology.

It is true that they talk about discarding the weakest links, but this is just a matter of discarding the "blinking links," i.e., only retaining links that have been over the threshold $Q$ for at least 5 consecutive years going back from the present. But I don't see how this set of surviving, robust links would get reduced by increasing of other link strengths in the Pacific. If anything, I would predict that an increasing of link strengths in the Pacific would just increase the number of robust links.

Comment Source:One more point to note about their methodology. It is true that they talk about discarding the weakest links, but this is just a matter of discarding the "blinking links," i.e., only retaining links that have been over the threshold $Q$ for at least 5 consecutive years going back from the present. But I don't see how this set of surviving, robust links would get reduced by increasing of other link strengths in the Pacific. If anything, I would predict that an increasing of link strengths in the Pacific would just increase the number of robust links.
• Options
37.
edited June 2014

David wrote:

John, there is one consistent set of typos in your formulas for $X(\ell,r,y,\tau)$ -- you've left out the second parameter $r$. Can you edit them, so we can make this thread look nice?

I've attempted to fix this everywhere; I've deleted your comment about this, and I will delete this one if seems all my $X(\ell,y,\tau)$s have been successfully corrected to $X(\ell,r,y,\tau)$'s. Thanks!

Comment Source:David wrote: > John, there is one consistent set of typos in your formulas for $X(\ell,r,y,\tau)$ -- you've left out the second parameter $r$. Can you edit them, so we can make this thread look nice? I've attempted to fix this everywhere; I've deleted your comment about this, and I will delete this one if seems all my $X(\ell,y,\tau)$s have been successfully corrected to $X(\ell,r,y,\tau)$'s. Thanks!
• Options
38.
edited June 2014

David wrote:

John, now that we have a terminology, can you explain your statement that with their definitions only the strongest links count as links? I don’t see any normalization of the link strengths taking place.

Hmm, I had thought their cutoff $Q$ varied with time, so that the presence of strong links increased $Q$ and made weaker links no longer count as links! This seems to have been a hallucination on my part. So, I withdraw this criticism. Thanks!

Part of my confusion is this: it seems that Yamasaki, Gozolchiani and Havlin seem to think El Niños "break climate links worldwide”, while Ludescher et al predict El Niños by detecting "increased cooperativity in the El Niño basin" before the El Niños.

These aren't contradictory, and of course one or both might be false.

But still, I want to get some better intuitive picture of what the authors of both papers think is going on, and see if there's a consistent reasonable-sounding story behind both accounts. Right now I don't have that.

Comment Source:David wrote: > John, now that we have a terminology, can you explain your statement that with their definitions only the strongest links count as links? I don’t see any normalization of the link strengths taking place. Hmm, I had thought their cutoff $Q$ varied with time, so that the presence of strong links increased $Q$ and made weaker links no longer count as links! This seems to have been a hallucination on my part. So, I withdraw this criticism. Thanks! Part of my confusion is this: it seems that Yamasaki, Gozolchiani and Havlin seem to think El Niños "break climate links worldwide”, while [Ludescher et al](http://forum.azimuthproject.org/discussion/1360/paper-ludescher-et-al-improved-el-nino-forecasting-by-cooperativity-detection/) predict El Niños by detecting "increased cooperativity in the El Niño basin" _before_ the El Niños. These aren't contradictory, and of course one or both might be false. But still, I want to get some better intuitive picture of what the authors of both papers think is going on, and see if there's a consistent reasonable-sounding story behind both accounts. Right now I don't have that.
• Options
39.

By the way, I've added an image of Figure 2 from Ludescher's paper to Graham's comments where he referred to that.

Comment Source:By the way, I've added an image of Figure 2 from Ludescher's paper to Graham's comments where he referred to that.