Wikipedia:Reference desk/Archives/Science/2017 January 26

Welcome to the Wikipedia Science Reference Desk Archives
The page you are currently viewing is an archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages.

January 26

Weather forecasting

If we lost all historical weather data, how long would we have to collect more data before our weather forecasting tools are as good as they are now? 24.255.17.182 (talk) 03:16, 26 January 2017 (UTC)

Weather forecasting talks about how forecasting is done. ←Baseball Bugs ^{What's up, Doc?} carrots→ 03:57, 26 January 2017 (UTC)

What counts as "all weather data"? Are you, for example, counting the data that constitutes the bits- and bytes- of the computer software that models weather? Without a more precisely-defined query, this is totally unanswerable.

In lieu of a poor-quality answer, have a read through the National Weather Service models and forecasts information page. My favorite forecast model is the HRRR, (rhymes with "brrr"); because it informs the Prog Charts (my favorite quick-glance-all-around weather forecast product). The HRRR uses current-conditions data (I don't think there's a sliding window of weather history data - it's current-conditions only)... but the software that implements HRRR gets tuned up every few years based on historical trends. If a bias is found - say, the cloud-cover predictions were too high when compared with actual data over time - they change the algorithms or the physics or the coefficients or any other tunable parameters - to make it work better. Would those kinds of model-changes count as "historical weather data"?

...but prog charts are essentially hand-drawn by a meteorologist or a briefer... so when we elide all historical weather data, do we also need to make those humans, including all the weather knowledge that they encapsulate, disappear too? It looks like we're starting down on that path, "to reduce the size of the Federal Government's workforce through attrition" ... (presumably including National Weather Service employees)! We'll know soon enough as the subtle effects of amok-policy trickle down in unforeseen ways!

Nimur (talk) 04:07, 26 January 2017 (UTC)

Assuming that we retained all the current technology, weather forecasting should be back up and running fairly quickly (one year's data should be plenty for most short term forecasts, while two satellite images are enough to tell you that the rain will arrive this afternoon). What would take much longer without the historic date is climate forecasting. Wymspen (talk) 10:30, 26 January 2017 (UTC)

Fascinating! How do you determine, or forecast, precipitation from satellite imagery?

If you look at infrared, you can estimate cloud top height (by combining infrared returns with other weather information).

If you look at visible light, you can estimate cloud coverage.

If you look at water vapor (a satellite camera that is sensitive to specific wavelengths corresponding to water vapor), you can see moisture.

But none of those things are rainfall!

In themselves, these things don't even predict rainfall!

Satellites only see the tops of clouds - and if you plan to stand on Earth, the cloud tops are the least-important-side-of-the-cloud to know about!

I depend on RADAR to determine precipitation; and barometers and temperatures (among other weather data) to predict short-term future precipitation; and in my estimation, satellites are the least-useful tools for figuring out where it will be raining this afternoon!

As a perfect example, have a look at current conditions in California (2017-01-26, 1600Z): the visible satellite is useless - the sun has not risen yet for the latest GOES imagery - all I see is darkness on this half of the USA! The infrared shows strong returns over the Sacramento delta - but are those cloud-tops morning fog or high altitude? I can guess from the returns that they're high altitude ... which doesn't tell me if it's raining in Vacaville! And if I look at the water vapor, all I see is that massive atmospheric river that has been developing all week, bringing moisture off the Pacific and into California and the central USA. There are clouds and moisture everywhere - this much I glean from the satellite - but when and where is the rain?

As it turns out, there is no rain forecast this afternoon in this region; but the weather service discussion predicts some in the morning with low confidence - based on some RADAR returns; but the forecast for the afternoon is driven by the air pressure that is building up. "Otherwise, dry weather conditions are forecast through the upcoming weekend and into early next week as high pressure builds inland."

As I recall, when you're training to interpret weather, there is a lot of emphasis on correct interpretation of data: clouds aren't rain; rain isn't clouds; these are (obviously) different nouns. Regarding satellite images, you can interpret that "a lot of high-level moisture and may also indicate cloudiness. This cloudiness could simply be high-level cirrus types or thunderstorms. That determination cannot be ascertained from this image by itself." You have to combine the satellite imagery with other weather information and do some mental weather-modeling!

But if you have a method to forecast the rain, using satellites, I'd love to hear it - I know at least a few people who would care to have great confidence in extraordinarily precise and accurate precipitation forecasts!

(Our regular readers probably know: Weather Prediction is one of my favorite branches of Science (with a capital S): we have the opportunity to do Real Actual Hypothesis-Testing to validate our ideas about basic physics and thermodynamics, and the answer matters! People throw all kinds of advanced computational mathematics at the problem; people build spaceships just to see the weather, and people still try wildcat ideas; and in the case of short-term forecasts, there is a straightforward way to validate the efficacy of any of these methods: it either rains or it does not rain, and there's no way to tool around the conclusive test of a prediction. This helps "shake out" the pseudoscientists, like so many grains of rice in the proverbial rain stick.

Nimur (talk) 16:48, 26 January 2017 (UTC)

The answer to "how long would we have to collect more data before our weather forecasting tools are as good as they are now" has two parts.

In terms of numerical weather forecasting models, the answer is "about 12 to 24 hours to recover most of the skill, and a few days to get the rest." It's a common misconception that weather (and climate) forecasts work by extrapolation from a historical record but they don't. For all practical purposes numerical weather forecasting is an initial value problem. This isn't quite right because of the role of data assimilation for the land surface and oceans, which have long time scales, but it's roughly correct for day-to-day forecasts.

The human element is more complicated. If you're not going to blank the memories of current forecasters then loss of data would not be crippling. There would be enough time to collect data as a new generation of forecasters grows into the job. If you're mind-erasing forecast personnel things could get dicey, especially for extreme events. Shock Brigade Harvester Boris (talk) 00:49, 27 January 2017 (UTC)

Double slit experiment - observer variation

What are the lab instruments used to conduct Young's double slit experiment with the 'observer' variation? Specifically, what is the instrument used to observe which slit the particle went through? Any references would be helpful. Thanks, Gil_mo (talk) 11:43, 26 January 2017 (UTC)

Most variations of the Young's experiment were originally gedanken experiments, designed as points of discussion regarding implications of quantum theory. Young actually did do the original experiment, in 1801, and we have an article about it called Young's interference experiment. The reason for two articles is that Young didn't do his original experiment to prove quantum theory. He did his original experiment to test the wave nature of light. Later variations were proposed to analyze the implications of wave-particle duality in modern quantum theory. As noted in the article at Wikipedia titled Double-slit experiment, and I quote, "Naive implementations of the textbook gedanken experiment are not possible because photons cannot be detected without absorbing the photon." In other words, with the photon variation of Young's experiment, you cannot passively detect a photon (since photons do not interact at a distance with matter), so you can't observe which slit they pass through. You can do variations involving particles like electrons (because electrons DO interact at a distance with matter), atoms, and molecules. As noted in the section titled "Other variations", and I quote "In 2012, researchers at the University of Nebraska–Lincoln performed the double-slit experiment with electrons as described by Richard Feynman, using new instruments that allowed control of the transmission of the two slits and the monitoring of single-electron detection events. Electrons were fired by an electron gun and passed through one or two slits of 62 nm wide × 4 μm tall." There's a reference to the original publication describing the experiment. You can follow it there. --Jayron32 11:59, 26 January 2017 (UTC)

Thank you! Gil_mo (talk) 22:37, 26 January 2017 (UTC)

Feynman Lectures. Lecture 45. Ch.45-3 [1]

...

However, they will turn out to be exactly the same if we assume UG−UL=const, instead of L=const. If we assume UG−UL=const, independent of temperature, then the argument leading to Eq. (45.15) will produce Eq. (45.16). Since the pressure is constant while the volume is changing, the change in internal energy UG−UL is equal to the heat L put in minus the work done P(VG−VL), so L=(UG+PVG)−(UL+PVL).

— Feynman • Leighton • Sands, The Feynman Lectures on Physics, Volume I

Can you show that eq. [45.15 http://www.feynmanlectures.caltech.edu/I_45.html#mjx-eqn-EqI4515] will transform to eq. [45.16 http://www.feynmanlectures.caltech.edu/I_45.html#mjx-eqn-EqI4516] if we assume UG−UL=const? We have :
${\tfrac {\partial P}{\partial T}}={\tfrac {L}{RT^{2}}}P$
${\tfrac {\partial P}{\partial T}}={\tfrac {(U_{G}+PV_{G})-(U_{L}+PV_{L})}{RT^{2}}}P$
${\tfrac {\partial P}{\partial T}}={\tfrac {U_{G}-U_{L}+P(V_{G}-V_{L})}{RT^{2}}}P$
${\tfrac {\partial P}{\partial T}}={\tfrac {const+P(V_{G}-V_{L})}{RT^{2}}}P$
What should I do next? Username160611000000 (talk) 14:58, 26 January 2017 (UTC)

For eq. like ${\tfrac {dy}{dx}}={\tfrac {ay^{2}+by}{x^{2}}}$ we have the next solution:
$(ay^{2}+by)^{-1}dy=x^{-2}dx$

$d({\tfrac {\ln {y}}{b}}-{\tfrac {\ln {ay+b}}{b}})=d(-x^{-1}+{\text{C}})$

${\tfrac {\ln {y}}{b}}-{\tfrac {\ln {ay+b}}{b}}=-x^{-1}+{\text{C}}$

$\ln {y}-\ln {(ay+b)}=-{\tfrac {b}{x}}+{\text{C}}_{2}$

$\ln {\tfrac {y}{ay+b}}=-{\tfrac {b}{x}}+{\text{C}}_{2}$

$e^{-{\tfrac {b}{x}}+{\text{C}}_{2}}={\tfrac {y}{ay+b}}$

$y={\tfrac {be^{{\text{C}}_{2}}}{e^{b/x}-ae^{{\text{C}}_{2}}}}$

Here $a=(V_{G}-V_{L})/R;b=(U_{G}-U_{L})/R.$ So we have :

$P={\tfrac {{\tfrac {U_{G}-U_{L}}{R}}e^{{\text{C}}_{2}}}{e^{(U_{G}-U_{L})/(RT)}-{\tfrac {V_{G}-V_{L}}{R}}e^{{\text{C}}_{2}}}}$
But it differs from eq. 45.16 $n={\biggl (}{\frac {1}{V_{a}}}{\biggr )}e^{-(U_{G}-U_{L})/RT}$.

Username160611000000 (talk) 06:19, 27 January 2017 (UTC)

I think you need to back up a step at the very beginning. Rather than:

${\tfrac {\partial P}{\partial T}}={\tfrac {L}{RT^{2}}}P$

Start with:

${\tfrac {\partial P}{\partial T}}={\tfrac {L}{T(V_{G}-V_{L})}}$

Which leads to:

${\tfrac {\partial P}{\partial T}}={\tfrac {U_{G}-U_{L}}{T(V_{G}-V_{L})}}+{\tfrac {P}{T}}\approx \left({\tfrac {U_{G}-U_{L}}{RT^{2}}}+{\tfrac {1}{T}}\right)P\approx \left({\tfrac {U_{G}-U_{L}}{RT^{2}}}\right)P$

08:30, 27 January 2017 (UTC)

Thank you very much! Username160611000000 (talk) 20:34, 27 January 2017 (UTC)

Triplicate experiments

Hi, I need a reference on the fact that it is common practice to triplicate experiments in biology. Can someone help me?

Thanks!

Megalexandros (talk) 20:49, 26 January 2017 (UTC)

doi:10.1038/492180a? DMacks (talk) 21:02, 26 January 2017 (UTC)

Thanks DMacks, but you know, it was required by the office as a part of the paperwork. Therefore it should be something more straightforward, or easy to understand for them, if you want. Something authoritative that contain a sentence loud and clear.

Megalexandros (talk) 21:07, 26 January 2017 (UTC)

You want a scientific analysis, you get a scientific paper. There are some fairly direct sentences in it. Some journals probably have specific requirements. If you are trying to support a position by lit-research, pull out the key part and cite it. DMacks (talk) 21:38, 26 January 2017 (UTC)

I have not heard of this triplication practice before before. If a single study has the proper controls and sufficient numbers of animals/organisms for an appropriate statistical power, it will not need duplication, let alone triplication. DrChrissy ^(talk) 21:15, 26 January 2017 (UTC)

It's triplication within the experiment, not triplication of the experiment. Fgf10 (talk) 08:29, 27 January 2017 (UTC)

Agreed, but I am not saying that every experiment needs triplication. Just that triplication is very common.

Megalexandros (talk) 21:18, 26 January 2017 (UTC)

But where have you got this impression from? DrChrissy ^(talk) 21:20, 26 January 2017 (UTC)

15 years of research in immunology.

Megalexandros (talk) 21:22, 26 January 2017 (UTC)

That's a considerable amount of experience. So why have you posted here asking people to do your research for you? DrChrissy ^(talk) 21:26, 26 January 2017 (UTC)

My job should be producing scientific results, not paperwork. Triplicate (or quadruplicate, or more) experiments is such a common thing that everybody just writes "averaged results from n experiments show that..." etc., but no one apparently cares stating that it is common in a research paper, that's why I asked here.

Megalexandros (talk) 21:31, 26 January 2017 (UTC)

One reason (I'm not saying it's good or the only), but only to make statements of clear and direct reasons for this practice. First, you need at least 3 in order to do standard deviation, and every scientist knows the importance of experimental reproducibility and appropriate statistical data analysis thereof (hold your laughter please). Second, it obviously costs more to repeat it. DMacks (talk) 21:38, 26 January 2017 (UTC)

Yeah, everything clear and logical. Except that I need a reference, and a good one. You know how stubbornly stubborn burocrats can be...

Megalexandros (talk) 21:42, 26 January 2017 (UTC)

Depends on specific field of study and ultimate use of the results. For example, doi:10.1373/clinchem.2008.112797 is a published set of guidelines for qPCR and Stephen Bustin is a pretty authoritative voice in the field. DMacks (talk) 21:50, 26 January 2017 (UTC)

On a closer inspection, your first reference maybe can go. I will use it if I don't find anything better. Thanks!

Megalexandros (talk) 21:56, 26 January 2017 (UTC)