# Difference between revisions of "1885: Ensemble Model"

 Ensemble Model Title text: I'm in talks with Netflix to produce an alternate-universe crime drama about the world where sliced bread was never re-legalized, but it's going slowly because they keep changing their phone numbers and the door lock codes at their headquarters.

## Explanation

 This explanation may be incomplete or incorrect: Requires descriptions of each entry. Do NOT delete this tag too soon.If you can address this issue, please edit the page! Thanks.

An ensemble model is a combination of multiple, similar models to show a wider range of possible outcomes. The graphs on the left are tracks of predictions from multiple models. In this comic, Randall starts out describing actual changes that ensemble models show, but sinks into absurdity, describing strange alternate universes and scenarios that likely would not be necessary in an actual model.

The upper graph looks like one plotting global temperatures with time using different scenarios, like this one: https://www.ipcc.ch/publications_and_data/ar4/wg1/en/figure-spm-5.html The bottom right graph is a typical hurricane path-prediction graphic.

Not all of the outcomes are serious. They are explained below:

…rain is 0.5% more likely in some areas

Historical rain data are used to estimate the probability of rainstorms of a certain size and duration occurring, e.g. the Flood Studies Report in the UK. Randall here is suggesting that an alternate universe exists where these estimates are higher (and presumably lower) in some areas, and that the estimates of rainfall in this alternate universe is accounted for within ensemble modelling in our own universe. This sort of change in prediction is frequently used when accounting for 'worst case scenarios' in the design processes of structures such as dams. However, the figures to the left appear to indicate time-dependent models, which are typically physics based, e.g. Large Eddy Simulation models or other atmospheric process based models. In those sorts of models, likelihood of rain is usually a prediction rather than a parameter, but might be used as a parameter in a second iteration.

…wind speeds are slightly lower

A usual parameter.

…pressure levels are randomly tweaked

A very vague but otherwise understandable parameter.

…dogs run slightly faster

This is where the comic diverges from reality; there is no reason to have the locomotion speed of dogs as a parameter.

…there is one extra cloud in the Bahamas

This situation is most likely too specific and subtle a difference to be useful to the model.

…Germany won WWII

"What if Germany won World War II" is a very popular subject for alternate history stories.

…snakes are wide instead of long

Snakes being as wide as they are long (and as long as they are wide) in present reality would have enormous consequences for zoology and other fields of biology, including evolutionary biology. It would also have an impact on art history, especially where it involves paintings depicting certain scenes from the book of Genesis. Compared to these effects, the expected upshot for meteorology seems to be limited.

…Will Smith took the lead in The Matrix instead of Wild Wild West

Actor Will Smith famously turned down the lead role of Neo in The Matrix, instead taking the role of Captain James T. West in the widely-panned action-comedy Wild Wild West. The role of Neo ultimately went to Keanu Reeves. For a more detailed discussion of how the cinematic world would have been different had Smith taken the role, see "How Will Smith Turned Down "The Matrix" - And Blew A Chance To Change Hollywood Forever."

…swimming pools are carbonated

A simple calculation reveals this as a serious greenhouse problem. In the United States there are not less than 5,000,000 private owned pools. Conservatively assumed a volume of 25,000 liters per pool gives 125 billion liters of carbonated soda. According to Wikipedia the U.S. sales reached around 30 billion bottles of water in 2008 (including non carbonated water) which is surely much less than all the pool water. While all those bottles are not considered to have an impact on the green house effect this scenario is getting even worse. Open a bottle of carbonated water and fill the content into glasses. More or less soon the sprinkling is over, meaning you have to open the next bottle and so on. In a pool at the bottom the pressure is high enough to hold the carbon dioxide but on the surface it behaves like the glass. So, while a glass needs new carbonated water every two hours, or ten times per day, let's say it's three times per day for the pool which leads to one thousand times per year. The total number in this scenario would be 125 trillion liters of carbonated soda, ejecting carbon dioxide, per year. But stop: The carbon dioxide used for artificial carbonated water is taken from the air and because of the pressure at the bottom of the pool it doesn't release all back this should have a positive effect. But as Randall has shown in Soda Sequestration this effect would be minimal.

…sliced bread, after being banned in January 1943, was never re-legalized.

Sliced bread was in fact banned in the US for about two months in early 1943, as a supposed wartime conservation measure. The issue was not the bread itself, but that the pre-sliced loaves required a heavier wax paper wrapping to prevent them from drying out too quickly.

The title text suggests that Randall has been pitching an absurd "alternate-universe crime drama" to Netflix, apparently based on the premise that a permanent sliced-bread ban would spawn a criminal underground (similar to those created by alcohol and drug prohibitions in actual history). He indicates that a breakdown in communication has occurred between them, though he does not directly assume responsibility for this situation. It is nonetheless clear that Netflix has zero interest in the pitch, and so Randall has become overzealous in pushing his idea, to the point that Netflix employees are changing their numbers (presumably they can't block his number because he has resorted to calling from many different phones). He has even taken to infiltrating Netflix's corporate headquarters using ill-gotten security codes, which is definitely illegal[citation needed], much like Elaine's "meetings" with Steve Jobs in 1337: Part 3.

However, it is clear that Netflix is uninterested and is attempting to prevent Randall from contacting them (or trespassing into the building).

## Transcript

In an ensemble model, forecasters run many different versions of a weather model with slightly different initial conditions. This helps account for uncertainty and shows forecasters a spread of possible outcomes.
[To the left side a picture shows several gray overlapping swirling lines emitted from a point, then gradually diverging rightwards. Below are two smaller pictures; the first shows the lines connected to several loops and in the second it's still a similar figure to the above but moving into the opposite direction with the point emerged to a spiral.]
[The text right to the pictures reads:]
Members in a typical ensemble:
A universe where…
…rain is 0.5% more likely in some areas
…wind speeds are slightly lower
…pressure levels are randomly tweaked
…dogs run slightly faster
…there's one extra cloud in the Bahamas
…Germany won WWII
…snakes are wide instead of long
…Will Smith took the lead in The Matrix instead of Wild Wild West
…swimming pools are carbonated
…sliced bread, after being banned in January 1943, was never re-legalized.

# Discussion

Where's the guy who knows how to make tables? A table would be good for this article, so we could explain each joke scenario. 172.68.26.5 15:41, 4 September 2017 (UTC)

I don't like tables when the text in the data cells is more than only a few words. That's bad layout. I have entered all the text from the list into separate headers for the appropriate floating text layout.--Dgbrt (talk) 18:39, 4 September 2017 (UTC)

Might be worth mentioning the context for this comic; viz. the approach of hurricane Irma, with a wide range of predictions as to where it might end up (and which areas it would hit), making weather modeling (and hurricane modeling in particular) – and the uncertainties involved – topical. It's clear to us now, but won't be clear to readers a few years from now. Pelosujamo (talk) 01:37, 5 September 2017 (UTC)

Wait - you mean it's not related to Harvey? (In other words, I'm not part of the "us" you speak about.) -- Hkmaly (talk) 02:17, 5 September 2017 (UTC)global warming https://www.explainxkcd.com/wiki/index.php/1885:_Ensemble_Model
I'm pretty sure this was inspired by Irma, not Harvey, because it's about uncertainty in weather modeling; which has received more attention with Irma than it did with Harvey. By the time America started paying real attention to Harvey the National Hurricane Center already had a very good (and accurate) idea about its future path. By contrast, the uncertainties in the Irma models made CNN's front page long before Irma was anywhere near populated areas. Also, it would be a bit late for Randall to do a Harvey comic; Harvey was last week's news. (Of course, Harvey did make hurricanes cool again.) Pelosujamo (talk) 13:24, 5 September 2017 (UTC)

I would say that one the idea of randall is related to point the change climate denier invalid reasoning that despite all scenario of global warning show increase of temperature, the fact that none of each is very likely to be wrong then all are wrong. (The fallacy is in the last then: the reunion of little probability can lead to high confidence or a the reunion of sum of various probable things can lead to absolutely certain ) Xavier Combelle (talk) 02:35, 5 September 2017 (UTC)

I have to disagree with the original explanation (now fixed) that "there is no reason to have the locomotion speed of dogs as a parameter". Dogs are known to chase cats, cats kill a large number of birds, birds eat insects including butterflies. If dogs would run slightly faster there could be a significant variation in the amplitude of the Butterfly effect. --141.101.69.147 12:13, 5 September 2017 (UTC)

Besides, the running speed of dogs would presumably impact how often, and where, one would experience raining cats and dogs.162.158.155.32 15:30, 6 September 2017 (UTC)

"[one extra cloud in the Bahamas] is most likely too specific and subtle a difference to be useful to the model." - Doesn't that depend on the size and disposition of said cloud? I'd say the problem here is vagueness, rather than insignificance.162.158.155.32 15:35, 6 September 2017 (UTC)

The upper graph looks like one plotting global temperatures with time using different scenarios, like this one: https://www.ipcc.ch/publications_and_data/ar4/wg1/en/figure-spm-5.html". I do not think this is an appropriate example of an ensemble model. The several trajectories for global temperature are for different policy decisions. In an ensemble model various trajectories reflect uncertainty about are a result of uncertainty about initial conditions or the physical rules that control the evolution of the system. TLDR: A map is not an ensemble model. The uncertainty (shaded area) for each track may or may not be the result of an ensemble, but if it is an ensemble for one of the scenarios would be a better example. Also ensembles are typically used for non-linear, chaotic systems and this should probably be somewhere in the explanation. 162.158.62.159 17:06, 6 September 2017 (UTC)

The global temperature doesn't decrease in any model. So I have changed this in the explanation and added the possibility of a depicted tornado, makes more sense for the big point at the beginning. Nevertheless I'm not sure what Randall means in this particular graph.--Dgbrt (talk) 14:35, 7 September 2017 (UTC)

Ensemble models are a form of a Monte Carlo Analysis. They are used in many engineering analyses, usually to determine an upper limit for some particular limiting quantity. The idea is that you do not necessarily believe any of the individual analyses, but that the ensemble forms an envelope of outcomes, so that if you design for the most extreme case, you can be confident that your design will not fail. They are used to make sure that the design is robust and has margin to failure. Of course, you cannot consider all of the uncertainties, which is why it is important to carefully identify sources of uncertainty before you do the analyses. If you do generate an ensemble envelope, and the data for the particular event falls outside the envelope, it is time to seriously reconsider the models, or the sources of uncertainty.13:20, 7 September 2017 (UTC)~~

…rain is 0.5% more likely in some areas

I have removed this because it's not accurate. This comic refers to the Universe (mathematics) and this outcome is high realistic.

Historical rain data are used to estimate the probability of rainstorms of a certain size and duration occurring, e.g. the Flood Studies Report in the UK. Randall here is suggesting that an alternate universe exists where these estimates are higher (and presumably lower) in some areas, and that the estimates of rainfall in this alternate universe is accounted for within ensemble modelling in our own universe. This sort of change in prediction is frequently used when accounting for 'worst case scenarios' in the design processes of structures such as dams. However, the figures to the left appear to indicate time-dependent models, which are typically physics based, e.g. Large Eddy Simulation models or other atmospheric process based models. In those sorts of models, likelihood of rain is usually a prediction rather than a parameter, but might be used as a parameter in a second iteration.

Check my more realistic explanations on the first three outcomes, they are no jokes.--Dgbrt (talk) 15:01, 7 September 2017 (UTC)

Pretty sure this has nothing to do with the mathematician's notion of universe - the math notion is used to dodge set-theoretic problems, but crucially everything one does is supposed to not depend on the specific choice of a universe (it may depend on the existence of one...). This is exactly not how the word is used here. 162.158.90.102 09:46, 10 September 2017 (UTC)