Support the Arctic Sea Ice Forum and Blog

Author Topic: Slater's thread  (Read 9555 times)

Lord M Vader

  • Nilas ice
  • Posts: 1248
    • View Profile
  • Liked: 18
  • Likes Given: 15
Slater's thread
« on: September 20, 2016, 08:42:45 AM »
Neven has published a sad blog text due to Andrew Slaters unexpectedly leaving the world. That's why I think there should be a "Slater thread".

Personally, it was not until this spring I became aware of Slaters excellent page and started to follow it. I have for a while wondered why the page hadn't been updated but I would never have expected such a tragic reason for that! :(

Best, LMV

oren

  • Young ice
  • Posts: 4159
    • View Profile
  • Liked: 666
  • Likes Given: 1186
Re: Slater's thread
« Reply #1 on: September 20, 2016, 09:28:47 AM »
Oh dear. A sad loss.

Neven

  • Administrator
  • First-year ice
  • *****
  • Posts: 7013
    • View Profile
    • Arctic Sea Ice Blog
  • Liked: 598
  • Likes Given: 396
Re: Slater's thread
« Reply #2 on: September 20, 2016, 09:47:55 AM »
Slater posted here on the ASIF under the moniker of Blizzard_of_Oz. Only know do I appreciate how exceptional that was.
Il faut comparer, comparer, comparer, et cultiver notre jardin

slow wing

  • Grease ice
  • Posts: 766
    • View Profile
  • Liked: 129
  • Likes Given: 355
Re: Slater's thread
« Reply #3 on: September 20, 2016, 11:00:36 AM »
That is very sad.

Here is the link to Neven's blog post:
http://neven1.typepad.com/blog/2016/09/in-memoriam-andrew-slater.html.

I only knew of him from his forecast page:
http://cires1.colorado.edu/~aslater/SEAICE/

I'll record now that his methodology there really impressed me. Dr Slater looked at each bit of Arctic sea ice in turn over the Arctic and predicted its survival probability. That approach makes a lot of sense to me.

He had already developed quite a good model to do that and presumably had progressive improvements in mind to make it even better.

Sad news. RIP Andrew Slater.

ktonine

  • Frazil ice
  • Posts: 363
    • View Profile
  • Liked: 0
  • Likes Given: 0
Re: Slater's thread
« Reply #4 on: September 20, 2016, 12:12:11 PM »
We were privileged to have Dr Slater stop by occasionally.  For those not aware, Walt Meier was his predecessor at NSIDC. It's always gratifying to have someone at that level of expertise take the time to interact with the likes of us :)

Condolences to his friends and family - he will be missed.

Archimid

  • Nilas ice
  • Posts: 2171
    • View Profile
  • Liked: 196
  • Likes Given: 161
Re: Slater's thread
« Reply #5 on: September 20, 2016, 12:45:38 PM »
This is very sad and bad news indeed. My deepest condolences to all his family and friends. My deepest condolences to all the people that will miss out on his outstanding work. I hope someone continues it. As far as I know his model was the most accurate forecast for arctic see ice available. This is a great loss for all mankind.
I am an energy reservoir seemingly intent on lowering entropy for self preservation.

crandles

  • Nilas ice
  • Posts: 2389
    • View Profile
  • Liked: 52
  • Likes Given: 42
Re: Slater's thread
« Reply #6 on: September 20, 2016, 03:52:48 PM »
Yes a very sad loss.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #7 on: July 27, 2019, 10:40:09 PM »
Re-booting this thread as a home for Slater model-related discussion that doesn't directly relate to the current melt season. Hope this is the appropriate place.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #8 on: July 27, 2019, 10:46:23 PM »
Can we stop posting this nonsense about the Slater model, its been debunked several times already in this thread. It makes no prediction about the minimum at all, merely what the ice will be like in 50 days time based on current conditions.

What do you mean debunked? It's just a model. It uses previous years to project 50-day survival probabilities based on current concentration. Yes, it does make a prediction about the minimum. If not, then why did Slater himself do assessments of its performance?

http://cires1.colorado.edu/~aslater/SEAICE/

You're welcome to, and should, use your own judgement about how much to weight different evidence or models in your own assessment of reality, but please don't attempt to impose your own preferences on others.

gerontocrat

  • First-year ice
  • Posts: 5794
    • View Profile
  • Liked: 952
  • Likes Given: 18
Re: Slater's thread
« Reply #9 on: July 27, 2019, 10:51:40 PM »
Having a thread called Slater's Thread means it can be ignored if so desired.

Posting a criticism seems - surplus to requirements, self-indulgence ?

And what if his model is spot on the mark this year?
"Para a Causa do Povo a Luta Continua!"
"And that's all I'm going to say about that". Forrest Gump
"Damn, I wanted to see what happened next" (Epitaph)

be cause

  • Grease ice
  • Posts: 806
  • Citizenship .. a Lurker gets asylum
    • View Profile
  • Liked: 182
  • Likes Given: 169
Re: Slater's thread
« Reply #10 on: July 27, 2019, 11:18:25 PM »
well found petm .. nothing could be more appropriate . As a source of misunderstanding and controversy every time it arises in the main threads , it will sit much better in a home of it's own  .. b.c.
2007 + 5 = 2012 + 4 = 2016 + 3 = 2019 ...

Archimid

  • Nilas ice
  • Posts: 2171
    • View Profile
  • Liked: 196
  • Likes Given: 161
Re: Slater's thread
« Reply #11 on: July 27, 2019, 11:50:17 PM »
Probabilities is just one of the many ways the Arctic must be measured. Slater model provides that. A probability. A close enough probability that is consistent with itself. I love to see it. It is one more data point and a useful one.
I am an energy reservoir seemingly intent on lowering entropy for self preservation.

oren

  • Young ice
  • Posts: 4159
    • View Profile
  • Liked: 666
  • Likes Given: 1186
Re: Slater's thread
« Reply #12 on: July 28, 2019, 12:00:41 AM »
The model does not project the minimum but a point in time 50 days ahead.
It also does not project the path to that point.
Both these assumptions were made on the melting season thread today. I think this is what ticked RR off.
You can guess the path from now to t+50. If point t+55 suddenly goes up 5 days from now, you can assume the whole imaginary path from t+5 to t+55 has moved upwards.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #13 on: July 28, 2019, 12:28:57 AM »
Either I don't fully understand the objection or I don't agree with it.

It's a very simple model. It does make a prediction about every day, 50-days in advance -- as you said. Which means that it does make a prediction about both the minimum and the path to get there.

If you look at his poster describing the model, Slater even points out how it predicted a melt slow-down in 2009. He assesses its skill across each day of the year. And he compares the accuracy of the model's predicted September extent average vs. other models' predictions.

http://cires1.colorado.edu/~aslater/SEAICE/Slater_AGU_2013_poster.pdf

You can guess the path from now to t+50. If point t+55 suddenly goes up 5 days from now, you can assume the whole imaginary path from t+5 to t+55 has moved upwards.

Not sure what you mean by this.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #14 on: July 28, 2019, 08:11:02 AM »
Got me confused there for a bit, but wasn't difficult to find out how the chart is really made:

Quote from: http://cires1.colorado.edu/~aslater/SEAICE/
i.e. every point in blue is/was made 50-Days in advance

So every day a new point is added at the end. This is also easy to see by taking two Slater graphs published 24 days apart as shown below, the shape of the blue line does not change one bit between 4th July and 28th July from one image to the other.

It seems to me that this is what Oren is trying to say, while petm seems to be claiming that every day all the future 49 days are recalculated and a new 50th day calculated.

So Oren is right in saying that the Slater graph does not predict the minimum, or in other words, that the minimum that is shown on the Slater graph towards the end of August is a false mininum.

Oren is also right in saying that the path from today to the end of the blue line will not be with the bumps and vallyes that Slater's grap has, therefore the path between the two can only be guessed at.

slow wing

  • Grease ice
  • Posts: 766
    • View Profile
  • Liked: 129
  • Likes Given: 355
Re: Slater's thread
« Reply #15 on: July 28, 2019, 08:21:59 AM »
OK, I'm somewhere in the middle here.

I do agree with petm that the model does make a prediction for each day.

On the other hand, consider the prediction for 1 September (showing as about the current lowest) vs. the 15 September (which is where we are up to now). The 1 Sept prediction was made with the data for 2 weeks ago.

If the Slater model instead used today's data for the 1 September prediction then it would have more skill than the point shown on the plot because it would be a 36-day prediction rather than a 50-day prediction.

Would a 36-day prediction for 1 September likely be higher, or lower, than the value shown? We might suspect higher, given that later predictions have tended to be higher (and also that the mimimum in prior years has usually (always?) been later than that). That is oren's point.

They don't do predictions for days 1-49 into the future (see .pdf referenced by petm), only for 50 days. So the model hasn't really been optimized for finding the extent minimum - that's not its purpose.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #16 on: July 28, 2019, 08:39:25 AM »
OK, I'm somewhere in the middle here.

I do agree with petm that the model does make a prediction for each day.
Well, we all agree that the model makes a prediction for each day. The model makes one prediction per day, 50 days in advance, which is what their own site says. And not what petm has been saying.
Quote
On the other hand, ...

They don't do predictions for days 1-49 into the future (see .pdf referenced by petm), only for 50 days. So the model hasn't really been optimized for finding the extent minimum - that's not its purpose.
Ok, reminds me of the Larry David episode I watched yesterday: "What's up with that when people say 'on the other hand' - are they disagreeing with themselves?" (quoted from memory).

El Cid

  • Frazil ice
  • Posts: 472
    • View Profile
  • Liked: 106
  • Likes Given: 37
Re: Slater's thread
« Reply #17 on: July 28, 2019, 12:40:57 PM »
Either I don't fully understand the objection or I don't agree with it.

It's a very simple model. It does make a prediction about every day, 50-days in advance -- as you said. Which means that it does make a prediction about both the minimum and the path to get there.

petm is perfectly right about this

Nonetheless, it is just a model, although the best publicly available known to me. Also, it does overestimate extent by about 0,5 M km2 in strong melt years.

Given that the model had a minimum slightly under 4 M sq km, I believe that (with 2019 having strong momentum) the actual, final NSIDC low will be 3,4-3,7 m sq km, giving 2019 a second place.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #18 on: July 28, 2019, 12:58:10 PM »
Either I don't fully understand the objection or I don't agree with it.

It's a very simple model. It does make a prediction about every day, 50-days in advance -- as you said. Which means that it does make a prediction about both the minimum and the path to get there.

petm is perfectly right about this
<snip>
I'm probably very dense, but I don't see how that can be.

The Slater model cannot predict the minimum in any meaningful way. It makes no attempt at predicting the timing of the mininum and it never makes a claim about what that mininum will be.

Looking at the graph, I am unable to see anything that indicates what the mininum will be or when. Absolutely nothing.

And the path to get there ... well, downwards for sure. But does anybody actually believe that the model is currently predicting a few days of positive extent numbers at the beginning of August (1), or a sharp drop in the last week of August (2) followed by a flat week and then a week of extent increases (3)?

If you think that the Slater model predicts the path to the mininum then you will have to answer yes to these questions. But you would be absolutely wrong!

Take for example the short rise in extent at the beginning of August - this is an artifact resulting from a shortlived rise in extent 50 days previously (four days in July saw extent going from 10.19 to 10.21 to 10.19 again) which has absolutely no predictive ability as to what will happen in the beginning of August.

DrTskoul

  • Nilas ice
  • Posts: 1402
    • View Profile
  • Liked: 174
  • Likes Given: 59
Re: Slater's thread
« Reply #19 on: July 28, 2019, 01:34:03 PM »
The projections of the Slater model are similar to the shooting method of solving partial differential equations or the hurricane path projections in NOAA. They are more correct at the start and loose accuracy toward the end. However if you follow those projection curves over time they are pretty close in defining the envelope of the actual trajectory of the melting. Nothing more nothing less.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #20 on: July 28, 2019, 02:15:23 PM »
So every day a new point is added at the end. ... It seems to me that this is what Oren is trying to say, while petm seems to be claiming that every day all the future 49 days are recalculated and a new 50th day calculated.

I'm not trying to say that. I understand the model (and the graph and the map). I just disagree with the interpretation.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #21 on: July 28, 2019, 02:23:25 PM »
If the Slater model instead used today's data for the 1 September prediction then it would have more skill than the point shown on the plot because it would be a 36-day prediction rather than a 50-day prediction.

Almost certainly true. That would be a model refinement. Slater even insinuates that he was thinking about doing that at some point:

Quote
Each day is forecast independently of all other days i.e. I currently do not forecast for days 1 to 49.
[Emphasis added.]
http://cires1.colorado.edu/~aslater/SEAICE/

But this has to do with accuracy, not interpretation.

So the model hasn't really been optimized for finding the extent minimum - that's not its purpose.

Good point. Eyeballing at the results over different years in the poster (linked above), the model is often not that great in predicting the timing nor exact value of the single-day minimum. Presumably that why he presents a comparison of Sept. averages instead.
« Last Edit: July 28, 2019, 02:28:35 PM by petm »

Richard Rathbone

  • Frazil ice
  • Posts: 496
    • View Profile
  • Liked: 19
  • Likes Given: 7
Re: Slater's thread
« Reply #22 on: July 28, 2019, 02:31:22 PM »
Can we stop posting this nonsense about the Slater model, its been debunked several times already in this thread. It makes no prediction about the minimum at all, merely what the ice will be like in 50 days time based on current conditions.

What do you mean debunked? It's just a model. It uses previous years to project 50-day survival probabilities based on current concentration. Yes, it does make a prediction about the minimum. If not, then why did Slater himself do assessments of its performance?

http://cires1.colorado.edu/~aslater/SEAICE/

You're welcome to, and should, use your own judgement about how much to weight different evidence or models in your own assessment of reality, but please don't attempt to impose your own preferences on others.

There's no prediction of a minimum without at least 3 days projected from the same starting conditions. Simple maths, you can't get a minimum from a single point, you need a minimum of 3.

The Slater model just projects a single point. 50 days out, not 49, not 51. Therefore no minimum.

You are confusing yesterday's projection of 50 days out with today's of 49.

If you had the model you could set it up to do predictions for 48, 49, etc. as well as 50 and thus predict the minimum, but the plot you posted isn't doing that.


petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #23 on: July 28, 2019, 02:33:14 PM »
Well, we all agree that the model makes a prediction for each day. The model makes one prediction per day, 50 days in advance, which is what their own site says. And not what petm has been saying.

Where did I say otherwise? I figured out how this model works a few weeks ago, when it was being discussed in the main thread. Despite all the confusion over it, it's a very simple model (which is one reason I like it). I doubt that I have made inaccurate claims about it since then. As I said, I am  simply disagreeing with what seems to be the accepted interpretation (on ASIF).

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #24 on: July 28, 2019, 02:38:56 PM »
So every day a new point is added at the end. ... It seems to me that this is what Oren is trying to say, while petm seems to be claiming that every day all the future 49 days are recalculated and a new 50th day calculated.

I'm not trying to say that. I understand the model (and the graph and the map). I just disagree with the interpretation.
Ok fair enough - I wasn't sure. But is the Slater graph open to interpretation? I don't think so, it does what it does, nothing more and nothing less.

It is not possible to interpret the result of a mathematical formula, and it is not possible to put another interpretation into the Slater graph than  that the last point on the blue line is always a product of a well-defined mathematical formula + real-world input.

However, one could argue that a side effect of the Slater model is that it does make predictions about the shape of the melting season and the timing and value of the September minimum.

I would maintain that it doesn't. The Slater model makes no real prediction about the shape of the melting season which can be seen quite well in the fact that the forecast blue line has bumps that will not eventuate - the forecast blue line has no predictive value for temporary stalls or falls in melt as they eventuate.

And the Slater model is unable to make any real prediction for the timing of the September minimum, nor of the magnitude of that minimum.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #25 on: July 28, 2019, 02:40:02 PM »
If you think that the Slater model predicts the path to the mininum then you will have to answer yes to these questions. But you would be absolutely wrong!

Then why does Slater (on his poster) point out how in 2009 the model successfully predicted a transient slow-down part way through the melt? I think Slater himself thought the model does have something to say about the path, and it also makes sense to me.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #26 on: July 28, 2019, 02:43:04 PM »
Well, we all agree that the model makes a prediction for each day. The model makes one prediction per day, 50 days in advance, which is what their own site says. And not what petm has been saying.

Where did I say otherwise? I figured out how this model works a few weeks ago, when it was being discussed in the main thread. Despite all the confusion over it, it's a very simple model (which is one reason I like it). I doubt that I have made inaccurate claims about it since then. As I said, I am  simply disagreeing with what seems to be the accepted interpretation (on ASIF).
Sorry to have put words into your mouth. I made this assumption based on other things you were saying.

It was particularly this that got me thinking:
Which means that it does make a prediction about both the minimum and the path to get there.
But it doesn't. If the model makes, every day, a one-off prediction for extension in 50 days time, then it simply can't make any prediction about the minimum nor the path to get there. So logically I assumed that you mistakenly thought that the Slater model did something else than what it does.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #27 on: July 28, 2019, 02:47:12 PM »
There's no prediction of a minimum without at least 3 days projected from the same starting conditions. Simple maths, you can't get a minimum from a single point, you need a minimum of 3.

Perhaps you could elaborate? In my understanding, minimum simply means the lowest value of any set.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #28 on: July 28, 2019, 02:48:31 PM »
But it doesn't. If the model makes, every day, a one-off prediction for extension in 50 days time, then it simply can't make any prediction about the minimum nor the path to get there.

Why not? This is exactly the interpretation that I disagree with.

Assume that in some universe the survival of ice was completely determined by its concentration 50 days prior. In that universe, the model would predict the exact trajectory of the ice, including the minimum. Obviously, we don't live in that universe; nevertheless, the whole point of the model is that this is a surprisingly good way of predicting ice melt.

I do agree that it's not particularly accurate about the exact path, As Dr. Tskoul and others pointed out, nor would I expect it to be. But that's qualitatively different from saying that it makes no prediction.
« Last Edit: July 28, 2019, 02:55:54 PM by petm »

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #29 on: July 28, 2019, 02:48:57 PM »
If you think that the Slater model predicts the path to the mininum then you will have to answer yes to these questions. But you would be absolutely wrong!

Then why does Slater (on his poster) point out how in 2009 the model successfully predicted a transient slow-down part way through the melt? I think Slater himself thought the model does have something to say about the path, and it also makes sense to me.
So was Slater getting metaphysical or was he making a small joke?

But let's put that aside, and put the model to the test:

The bumps and slides of the red line and the blue line have not been following each other very closely so far this year, look at the beginning of June where they were heading in opposite directions.

As for the rest of the year, the bump at beginning of August and the (false?) minimum a month later are some pretty definite predictions. If the model is at all able to make predictions about the shape of the melt curve, then we should see both eventuate.

So let's see how well they pan out? And when you think that the Slater model is actually predicting the minimum let us know and we can see how reliable that prediction was.

DrTskoul

  • Nilas ice
  • Posts: 1402
    • View Profile
  • Liked: 174
  • Likes Given: 59
Re: Slater's thread
« Reply #30 on: July 28, 2019, 02:54:09 PM »
You folk have no idea what a model is and what a model should or cannot do.. the back and forth is silly. If you think a model should be right or perfect to guide the analysis you are fooling yourselves. A 70-80% accurate model with the right observations is quite valuable.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #31 on: July 28, 2019, 02:59:30 PM »
But it doesn't. If the model makes, every day, a one-off prediction for extension in 50 days time, then it simply can't make any prediction about the minimum nor the path to get there.
Why not? This is exactly the interpretation that I disagree with.
Well, why not indeed! Bu how would that work?

The ability of the Slater model to predict the shape of the melting season is obviously not very good - if existent at all. just compare the red and the blue, extremely bad match.

And gow could Slater's statistical map-derived model have predicted last year's minimum (September 19) by making statistical calculations based on the pixels on a map from the 1st of August. What mechanism do you think is underlying any such predictive ability?

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #32 on: July 28, 2019, 03:06:13 PM »
You folk have no idea what a model is and what a model should or cannot do.. the back and forth is silly. If you think a model should be right or perfect to guide the analysis you are fooling yourselves. A 70-80% accurate model with the right observations is quite valuable.
Well this  fool would like to know if a particular model is 70-80% accurate or 0% accurate.

The Slater model is obviously close to 0% accuarate in predicting the shape of the melting curve ("the path") and I've seen nothing to indicate that it should be any more accurate in predicting the minimum. So when other people are making such claims, and using the model to predict things that it can't predict and wasn't made to predict, isn't it reasonable to try and find out why?

And this is not to say that I do not think that the Slater model is extremely valuable. But the only predictive ability it has (or claims to have) is making a statistical guess at the extent after 50 days. And the accuracy of the prediction falls drastically the closer to the turning point we get, as DrTskoul points out.

oren

  • Young ice
  • Posts: 4159
    • View Profile
  • Liked: 666
  • Likes Given: 1186
Re: Slater's thread
« Reply #33 on: July 28, 2019, 03:11:37 PM »
petm, I fail to see the possibility for a different interpretation. The model predicts point 50. When 5 days later new data comes in, the model makes another prediction for 50 days ahead, let's call that point 55.
After this new data on day 5, the model does not predict the value for day 50 anymore, only the value for day 55. Day 50 value is shown on the graph not as a prediction of the path to day 55, but as a recording of the day 50 prediction made 5 days ago.
Let's say the new data between day 0.and day 5 hints at a massive slowdown of the melting season. The model would generate a day 55 prediction that is much higher than it would have been otherwise. This by necessity requires the whole path to day 55 to be higher. The day 50 prediction is not valid anymore, and if somehow the model would be tweaked to produce predictions for 45 days ahead using the same methodology, the day 50 prediction made on day 5 would have been much different, much higher, than the day 50 prediction made on day 0. Because of the new data.
In other words, when day 55 prediction is made, all previous predictions are no longer valid, as they are based on old data. Therefore the curve of all past predictions is not the predicted curve the ice will take to get to the last prediction 50 days ahead of today. No matter how you slice it and dice it, the model does not predict the path. Not because it's a good or bad model, but because it simply does not predict the path, regardless of model skill.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #34 on: July 28, 2019, 03:13:58 PM »
So was Slater getting metaphysical or was he making a small joke?

But let's put that aside, and put the model to the test:

The bumps and slides of the red line and the blue line have not been following each other very closely so far this year, look at the beginning of June where they were heading in opposite directions.

As for the rest of the year, the bump at beginning of August and the (false?) minimum a month later are some pretty definite predictions. If the model is at all able to make predictions about the shape of the melt curve, then we should see both eventuate.

So let's see how well they pan out? And when you think that the Slater model is actually predicting the minimum let us know and we can see how reliable that prediction was.

As I have said, I think the model actually does have something to say about the path, it's just not particularly accurate that aspect of its predictions.

But getting back to the original point about minimums, I think it is certainly noteworthy when the model appears to reach its minimum prediction and increases for several days. Since the model has been quite skillful at September averages, this point on the curve may give some early indication of what the 2019 minimum may be.

Not the exact minimum, nor the exact timing, that I concede. And perhaps should have been more clear about it in my original post, although I actually did caveat what I said. But in my opinion it is a very useful data point to consider.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #35 on: July 28, 2019, 03:20:05 PM »
What mechanism do you think is underlying any such predictive ability?

I would guess that concentration is (on average) a proxy for both melt rate (via mobility, reduced albedo, etc) and ice characteristics (e.g. thickness) that are key in determining how long it will take ice to melt.

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #36 on: July 28, 2019, 03:35:23 PM »
petm, I fail to see the possibility for a different interpretation. The model predicts point 50. When 5 days later new data comes in, the model makes another prediction for 50 days ahead, let's call that point 55.
After this new data on day 5, the model does not predict the value for day 50 anymore, only the value for day 55. Day 50 value is shown on the graph not as a prediction of the path to day 55, but as a recording of the day 50 prediction made 5 days ago.

Let's say the new data between day 0.and day 5 hints at a massive slowdown of the melting season. The model would generate a day 55 prediction that is much higher than it would have been otherwise.
True (if such a slowdown resulted in either concentration increases or stalls).

Quote
This by necessity requires the whole path to day 55 to be higher. The day 50 prediction is not valid anymore, and if somehow the model would be tweaked to produce predictions for 45 days ahead using the same methodology, the day 50 prediction made on day 5 would have been much different, much higher, than the day 50 prediction made on day 0. Because of the new data.

Yes and no. Yes if you want the model to be sensitive to changes in the melt rate from day 50 to day 0. Which it does not, so the model is not sensitive to such and therefore the predicted path is not very accurate. Nonetheless that doesn't mean that the model makes no prediction of the curve and, depending on the concentration distribution, it can in fact predict large general features of the curve such as the slowdown in 2009.

That's what I mean by a difference in interpretation.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #37 on: July 28, 2019, 03:51:15 PM »
Nonetheless that doesn't mean that the model makes no prediction of the curve and, depending on the concentration distribution, it can in fact predict large general features of the curve such as the slowdown in 2009.
Well this year (and all the other years I've been seeing the Slater predictions) the red and the blue lines have never been particularly likely to follow each other. So the slowdown in 2009 being predicted by the Slater model - I would think that very unlikely. Single anectdotal evidence like this looks more like a fluke than a prediction.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #38 on: July 28, 2019, 03:55:43 PM »
What mechanism do you think is underlying any such predictive ability?

I would guess that concentration is (on average) a proxy for both melt rate (via mobility, reduced albedo, etc) and ice characteristics (e.g. thickness) that are key in determining how long it will take ice to melt.
Well, yes, I think we all agree on the underlying physics and the general state of the system. But what is it that could, on August 1st 2018, have enabled the model to make any informative guess for a minimum on the 19th of September that year?

I am unable to see any method of making such predictions. I do not think that there is any  method or process that could take the state of the ice today, July 28th, and make any valuable non-zero prediction as to what date the minimum will be. And other than going on pure averages (which is what Slater and Gerantocrat both do) to make a prediction as to how low it will go.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #39 on: July 28, 2019, 03:58:44 PM »
But getting back to the original point about minimums, I think it is certainly noteworthy when the model appears to reach its minimum prediction and increases for several days. Since the model has been quite skillful at September averages, this point on the curve may give some early indication of what the 2019 minimum may be.

Not the exact minimum, nor the exact timing, that I concede. And perhaps should have been more clear about it in my original post, although I actually did caveat what I said. But in my opinion it is a very useful data point to consider.
I think this is were it all started - that apparent minimum at the beginning of September. I would maintain that there is no predictive value here, and that it is incorrect to say that the model "appears to reach its minimum prediction" because the model makes no such prediction.

But if I understand you correctly, you think that Slater's model is predicting a very early minimum, around the end of August / beginning of September (2). Well, we will see, won't we?

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #40 on: July 28, 2019, 04:10:20 PM »
But if I understand you correctly, you think that Slater's model is predicting a very early minimum, around the end of August / beginning of September (2). Well, we will see, won't we?

Not exactly. I'm saying it looks like the model's prediction has reached its minimum. If so, the model's September average will be some amount greater than this minimum of ~ 4M km2.

Given that, IF this year is an average year (big IF), we should not expect a new minimum extent record.

El Cid

  • Frazil ice
  • Posts: 472
    • View Profile
  • Liked: 106
  • Likes Given: 37
Re: Slater's thread
« Reply #41 on: July 28, 2019, 04:14:16 PM »
Suppose a world where 50 days out any ice with less than 80% concentration totally melts out and everything above 80% remains. In this world if you know the current concentration, you know exactly the ice extent 50 days ahead. 5 days from now you will know exactly the extent for day 55. Therefore you have exact forecasts 50 days out, ie. you also have a path. And the forecast on day 5 won't change the previous forecasts.

The Slater model does exactly this,  but in our world we have uncertainity therefore it gives (based on past years' survival rates) you a probability of the given ice"pixel" surviving.

Therefore the Slater model gives you estimates for each day that don't change retroactively just because we know more, ot because time passed. Therefore it gives you both a path and a minimum forecast.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #42 on: July 28, 2019, 04:15:00 PM »
Not exactly. I'm saying it looks like the model's prediction has reached its minimum.

Ok, here we can agree totally!
[qute]If so, the model's September average will be some amount greater than this minimum of ~ 4M km2.
[/quote]
Yes, it will be interesting to see if the Slater model can be used to make such a prediction.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #43 on: July 28, 2019, 04:16:19 PM »
Therefore the Slater model gives you estimates for each day that don't change retroactively just because we know more, ot because time passed. Therefore it gives you both a path and a minimum forecast.
This does not follow - there is no logical connection between a series of 50 days ahead point predictions and the path or the minimum.

Besides, the path is obviously totally out, just look at the blue and red lines!

El Cid

  • Frazil ice
  • Posts: 472
    • View Profile
  • Liked: 106
  • Likes Given: 37
Re: Slater's thread
« Reply #44 on: July 28, 2019, 04:20:22 PM »
Therefore the Slater model gives you estimates for each day that don't change retroactively just because we know more, ot because time passed. Therefore it gives you both a path and a minimum forecast.
This does not follow - there is no logical connection between a series of 50 days ahead point predictions and the path or the minimum.

Besides, the path is obviously totally out, just look at the blue and red lines!

But it does!

In my example with no uncertainity, the 50 day forecasts are the path. Same stands for the world with uncertainity

Besides, if in your view the model does NOT forecast the path and does NOT forecast the minimum, then what the hell does it forecast????

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #45 on: July 28, 2019, 04:25:22 PM »
Therefore the Slater model gives you estimates for each day that don't change retroactively just because we know more, ot because time passed. Therefore it gives you both a path and a minimum forecast.
This does not follow - there is no logical connection between a series of 50 days ahead point predictions and the path or the minimum.

Besides, the path is obviously totally out, just look at the blue and red lines!

But it does!

In my example with no uncertainity, the 50 day forecasts are the path. Same stands for the world with uncertainity
Ok, let's define the predicted path as the blue curve and the real path as the red curve. Then look at the two of them so far. The blue curve is not predicting the path of the red curve at all. Look at the beginning of June for a very good example.
Quote
Besides, if in your view the model does NOT forecast the path and does NOT forecast the minimum, then what the hell does it forecast????
It forecasts what it says it does: The most likely ice extension after 50 days, based on the current state of the ice and the historical record of ice melt.

That it does not forecast the minimum is easy to check: What minimum is it forecasting now? What? It isn't forecasting a minimum now? Ok, then when will it forecast the miminum? And how can we tell when it is forecasting the minimum?

Richard Rathbone

  • Frazil ice
  • Posts: 496
    • View Profile
  • Liked: 19
  • Likes Given: 7
Re: Slater's thread
« Reply #46 on: July 28, 2019, 04:28:15 PM »

Yes, it will be interesting to see if the Slater model can be used to make such a prediction.

It has been used to make these predictions, and does pretty well at them, but it doesn't do them 50 days in advance. It does them on the same day, different amounts in advance. e.g. on Aug 1st 31 days ahead for Sept 1st, 32 days ahead for Sept 2nd etc.

You can find them for past years at SIPN.
https://www.arcus.org/sipn

petm

  • Grease ice
  • Posts: 572
    • View Profile
  • Liked: 248
  • Likes Given: 21
Re: Slater's thread
« Reply #47 on: July 28, 2019, 04:31:17 PM »
And how can we tell when it is forecasting the minimum?

The forecasted minimum is the lowest point on the curve, which looks like it may have passed a few days ago.

(I'm not saying that it's accurate, or that Slater claimed it to be accurate, only that the lowest point on the curve -- once the season is over -- is by definition the model's forecasted minimum.)

DrTskoul

  • Nilas ice
  • Posts: 1402
    • View Profile
  • Liked: 174
  • Likes Given: 59
Re: Slater's thread
« Reply #48 on: July 28, 2019, 04:32:17 PM »
You folk have no idea what a model is and what a model should or cannot do.. the back and forth is silly. If you think a model should be right or perfect to guide the analysis you are fooling yourselves. A 70-80% accurate model with the right observations is quite valuable.
Well this  fool would like to know if a particular model is 70-80% accurate or 0% accurate.

The Slater model is obviously close to 0% accuarate in predicting the shape of the melting curve ("the path") and I've seen nothing to indicate that it should be any more accurate in predicting the minimum. So when other people are making such claims, and using the model to predict things that it can't predict and wasn't made to predict, isn't it reasonable to try and find out why?

And this is not to say that I do not think that the Slater model is extremely valuable. But the only predictive ability it has (or claims to have) is making a statistical guess at the extent after 50 days. And the accuracy of the prediction falls drastically the closer to the turning point we get, as DrTskoul points out.

Compare observed vs predictions. I think overall it does a pretty good job. It is useful to give you a sense of the direction. That is why previously it had been useful predicting possible slow downs. If you are looking for certainty, it will not give you any.

binntho

  • Grease ice
  • Posts: 741
    • View Profile
  • Liked: 179
  • Likes Given: 52
Re: Slater's thread
« Reply #49 on: July 28, 2019, 04:33:50 PM »
And how can we tell when it is forecasting the minimum?
is by definition the model's forecasted minimum
No, not if it is an artifact of the method used, in which case there is by definition no forecast for a minum.