[stevewepay]

How exactly do they do a fair comparison between 1880 and today? I'm assuming the temperature readings today are very accurate, but how accurate were the measuring devices going back in history, and their standards for taking temperature? Most of the thermometers I have in my house are +- 2F, which would eliminate the differences entirely.

[cryptoz]

There are significant temperature records that go back much farther than 1880 even. Many navies and armies kept records from all over the world, so we actually have lots more data than people think. There are also bodies of science that deal with 'aftcasting', which is making a weather forecast for a date in the past. These techniques have been refined to the point where we can make 1-day aftcasts for ~1850 that are comparable to modern 2-3 day weather forecasts in terms of accuracy, even with the limited data available (things like barometric pressure readings from ships).

I'm not sure about the technology used at the time, but the basic answer to your question that that they had a lot more thermometers than you expect, distributed globally too. And as we're discussing global averages, the noise about +-2F in your house isn't really the same problem.

Edit: Here's a neat 100-year reanalysis paper if you're interested. Abstract link here, full pdf available on the page: http://journals.ametsoc.org/doi/abs/10.1175/BAMS-87-2-175. "Feasibility of a 100-Year Reanalysis Using Only Surface Pressure Data"

[rnovak]

To be honest, and this is my opinion only, but 2-3 day forcasts, at least in my area, are horribly inaccurate. In my state, it's a common joke that being a meteorologist is the only job you can have where you can be wrong 95% of the time, and still keep your job.

Seriously, our forecasts are nearly always wrong. So.....saying we can make as accurate of forecasts for 200 years ago...that's not really saying much.

This is after moving to two cities in the same state, more than 100 miles apart, the forecasts still never get better.

[cryptoz]

Yes, weather forecasting is hard - that's why there's lots of research into 'aftcasts'. If we can learn to make better predictions while limiting the data we take in, we can make big improvements to the accuracy of modern forecasts.

You can make fun of meteorologists if you want, but they are not wrong 95% of the time. In fact, over the last 20 years the accuracy of predictions has gone WAY up on average. There are still many places where forecasting is difficult and it's going to be a long road to improve it to get people like you to trust the forecast. But we're improving and we're making the best forecasts we've ever made in the industry.

[rnovak]

I'm not trying to make fun of meteorologists at all, I totally get that your job is hard. I actually wanted to take the same approach (My company provides lunch for employees, I wanted to train a neural net on past lunch menus, and trading/derivative market data for food type resources, to see if it could predict tomorrows menu), unfortunately that was shot down, so I never got the chance to do something like that.

If it came off as me trying to be down on meteorologists, that wasn't my intent, and I can't edit my original post to reflect that, so hopefully you'll see my comment here.

[criley2]

>To be honest, and this is my opinion only, but 2-3 day forcasts, at least in my area, are horribly inaccurate. In my state, it's a common joke that being a meteorologist is the only job you can have where you can be wrong 95% of the time, and still keep your job.

Meteorologists lower the quality of the data they receive.

People don't want bad news. People don't want to hear that it's going to rain on their weekend. People won't tune in to hear constant bad news.

So meteorologists take the positive and optimistic side.

The reality of forecasting is that much of the data, models, the warnings, are all created by the NWS.

The further you get from real NWS data, the lower the quality tends to be. It's like a game of telephone, except with forecasting.

National forecasters like The Weather Channel utilize tons of NWS data and models, and present it in a more friendly way. Generally there is a loss of quality here (and, when you look at TWC's buzzfeed approach to weather, it's not hard to see why they might respect clicks above all else).

Then, further down the pipe, away from the NWS and the actual source, away from National broadcasters, we get to local broadcasters. Once again, these guys are largely relying on upstream info, just packaging for the local audience.

It's not surprising that local meteorologists provide the lowest quality forecasting, because that's not really their job to provide highly accurate forecasting. Their job is actually to keep you watching through a commercial break, so the show remains on-air. The NWS is the service whose job is it to accurately forecast.

[mikeash]

Another big problem is that people don't understand uncertainty, and therefore forecasters don't express uncertainty.

For example, just look at whatever near-term forecast you have handy. What are the forecast high and low temperatures for the next few days? What are the error bars on those highs and lows?

Oh, you couldn't find error bars? Yeah.

[leni536]

That's why I always check radar map forecasts. They forecast rain everywhere in my area but not at my exact spot? Yeah, I take my umbrella.

[jrs235]

> People don't want bad news ... People won't tune in to hear constant bad news.

Where do you live? Every time I turn on the news its 90+% FUD.

[rdtsc]

> So.....saying we can make as accurate of forecasts for 200 years ago...that's not really saying much.

It is saying quite a lot actually because the past has already happened. In the first case you are talking about the future, and yes, in some places weather is more unpredictable. You can of course move to the desert and enjoy clockwork predictable weather if it bothers you that much.

The OP was talking about "afcasting" though. In that case measurement have already been taken, but there are other records: ice cores, sea sediment, and many others. So it makes for a very interesting topic.

[rnovak]

wait a second. The OP was comparing "aftcasting" TO "forecasting" 2-3 days in advance, and that's what my statement was based on.

If your 2-3 day forecasting is off by 5-6 degrees (celcius), why would your "aftcasting" forecast for 200 years ago be any more accurate if it is "just as good" as a 2-3 day forecast? That means I would expect an error of 5-6 degrees, same as a 2-3 day forecast

edit: formatting

[cryptoz]

If everyone is nitpicking that part of my comment, it would be worth noting that I was paraphrasing from the paper I linked.

> For the beginning of the twentieth century, the errors of such upper-air circulation maps over the Northern Hemisphere in winter would be comparable to the 2–3-day errors of modern weather forecasts.

Might be worth reading the abstract / full paper if you're interested in the topic.

[rnovak]

what's the actual error bound on both forecasts and aftcasts?

The article this discussion is linked to is claiming "1.46°F (0.81°C)" variance over recorded history. Recorded meaning they know what actually happened.

But I cannot believe that 2-3 day forecasts are accurate to within 2 degrees. Even the services linked by other commenters criticizing my original point claimed only 77% accuracy in the best case,

[zajd]

I'm guessing you have some data that backs up this point?

There are some pretty accurate forecasting services

http://www.wunderground.com/most-accurate-weather

http://forecastadvisor.com/

[Frondo]

In fairness, he did say that the whole thing was his opinion, so I'm guessing he's just hoping we'll notice the sounds-like-facts nature of his opinions and go along with it.

[ori_b]

> To be honest, and this is my opinion only, but 2-3 day forcasts, at least in my area, are horribly inaccurate.

It gets easier over longer periods of time. I may not be able to predict very well whether it's going to be 27 degrees C on July 17th 2016, but I can make a pretty good guess at the average temperature throughout the entire month, and I would be willing to put money down that it will lie within a relatively narrow band.

[BillTheCat]

By "horribly inaccurate" do you mean 3 degrees Celsius off or 20 degrees Celsius off? These aftcasts are accurate enough for us to plot the overall course of global climate change.

[rnovak]

I mean turning on the "weather channel" and having it say it's sunny out, only to look outside to see pouring rain across the entire city. That's pretty inaccurate in my book.

[shanusmagnus]

Where in god's name do you live where the distinction between sun and rain is beyond the skill of your weather people?

[BillTheCat]

That has nothing to do with the temperature.

[rnovak]

It's all linked together, and while that statement had nothing to do with temperature, the same can be said in my area with regard to that.

Also, if a meteorologist can't get the current state of the weather correct, why on earth would I trust his statement of what the temperature is going to be like? At that point I'm just going to consider it a guess.

[malchow]

Do out of band "aftcasts" correctly show the decade-long global warming hiatus from 1998 to 2008? [1]

[1] https://en.wikipedia.org/wiki/Global_warming_hiatus

[abritinthebay]

Anyone who thinks this is an issue doesn't understand trends and variance. It's a high-school level failure of mathematics.

[malchow]

I suspect you fall prey to a question of scope. I imagine you spend more time with statisticians than, say, paleoceanographers and geologists. They might say, and reasonably so, that my reference to the ten year "break" in alleged global warming is irrelevant (as you say it is) and that your reference to a purely centurial warming trend is just as irrelevant. They might then accuse you of failing to understand variance and of having failed high school mathematics.

[melling]

Yeah, your comment about weather is an FAQ. I started a little blog FAQ a few weeks ago so I wouldn't have to repeatedly answer the same questions.

https://2cco2.wordpress.com/2015/08/04/the-difference-betwee...

[jnbiche]

Well, you didn't answer it in your blog, either. You simply posted link to a video that many HNers will not even click on, given our well-known preference for text presentations and the fact that many of us browse from work locations where video playing is not appreciated (even if the presenter in the video happens to be a pretty cool guy in this case).

I'd recommend at least writing a brief summary to the video if you really want to maintain an effective FAQ.

[melling]

Just watch Neil deGrasse Tyson walk the dog while explaining it. I can't really do it any better than him.

[marcstreeter]

While I see your point, the forecasts for past weather are based on data very different from forecasts for tomorrow's weather. The former is for past occurred events and the latter is for future probabilities.

[alelefant]

I think that joke is made in every state.

[guscost]

Prior to the satellite record historical temperature data is much less reliable, since each source is in a local area and may have different heat island effects, calibration, or quality of instruments. Additionally not as many stations were operating in 1880, which means more uncertainty about the data we do have.

Folks have tried to get around this by finding proxies for temperature (measurements that are theorized to track temperature, like the width of tree rings for example), and then calibrating that data by comparing it with the intervals for which we have reliable data (since 1970 for satellite data, or earlier for whatever thermometer data might be suitable). The long-term trends are identified and argued based on these proxies.

So to answer your question, they can calibrate a proxy using the satellite record, see what it suggests about 1880, and combine that with the available temperature data, but that's about it.

[pgrote]

I do not know. I am ignorant of the means in which NASA gathers temperature from the middle of the ocean and presume it is satellite. If that is the case, how was the temperature in the middle of the ocean ascertained in 1880? I would presume they base it on datapoints collected using thermometers by those at sea.

[bjwbell]

In 1880 via ship measurements. They'd take out a bucket of sea water and measure the temperature. They still use buoys as part of the sea surface temperature measurement.

[bjwbell]

It's not actually possible without making inferences via models. The NOAA recently switched their global temperature model [0].

[0] http://www.ncdc.noaa.gov/sotc/global/2015/5/supplemental/pag...