2

EDIT 6/13/2020: Actual Jan average from same website was 50.3 °F, which is even above my one-sigma bound, so I failed badly. What did I miss now that we have hindsight?

On Nov 1, I looked at the Dallas monthly weather averages (pasted below) and tried to predict the average temperature for Jan 2020. It's hard to see any trend, so I would simply predict 46.2 +/- 3.5 °F.

What else should I have measured to predict this significantly better? I was thinking about looking at solar irradiance data (with sunspot solar cycles), sea currents, or sea temperatures...what data would improve my prediction by more than 1°F?

I'm hoping for an answer like, "Well, if you go to website X, their measurement of Y indicates that temperatures will likely deviate by Z" with perhaps some science showing that this is a good way to predict long-term weather.

Year  Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec   Total
-----------------------------------------------------------------------------------
2019  45.8  50.2  55.0  66.0  73.4  79.9  84.6  87.4  85.5  65.5       
2018  45.8  51.1  63.3  61.6  79.0  85.7  88.8  85.2  78.1  66.2  52.5  48.4  67.1
2017  51.2  60.6  65.7  69.3  75.4  82.5  86.6  84.4  80.6  69.6  62.4  49.7  69.8
2016  47.0  55.2  61.2  68.1  72.5  84.0  87.4  85.8  81.5  74.1  63.5  49.7  69.2
2015  44.5  45.7  56.1  65.8  70.9  82.1  87.1  87.3  82.7  71.2  58.7  53.7  67.2
2014  45.3  47.0  55.1  66.3  74.4  82.4  83.8  86.2  80.3  71.6  51.5  50.1  66.2
2013  49.1  52.0  56.4  63.0  72.3  82.6  84.5  87.1  82.4  68.2  53.5  43.1  66.2
2012  50.4  52.5  64.3  70.3  77.9  84.3  87.7  86.5  80.0  67.0  59.7  51.2  69.3
2011  42.8  49.5  61.3  70.8  72.8  86.8  91.4  93.4  80.0  68.2  57.9  47.6  68.5
2010  44.3  41.7  55.8  66.7  76.9  86.5  85.9  89.8  79.8  68.8  58.5  49.3  67.0
2009  48.1  55.8  58.9  64.8  73.3  83.8  86.6  85.3  76.2  62.7  59.5  42.7  66.5
2008  46.9  54.0  61.0  66.5  77.0  86.5  89.0  86.7  78.5  69.6  59.7  49.0  68.7
2007  42.3  49.4  63.9  62.4  74.6  81.4  83.8  87.8  81.6  72.1  61.3  49.8  67.5
2006  55.1  49.4  62.2  72.0  78.0  83.7  87.7  89.8  77.6  68.1  57.5  50.0  69.3
2005  49.5  52.6  56.7  65.5  73.5  83.6  85.1  86.8  83.7  68.8  60.8  48.1  67.9
2004  48.5  45.8  61.8  66.2  74.4  79.4  83.5  81.5  78.4  72.1  57.1  48.9  66.5
2003  43.7  45.6  56.2  66.9  75.5  79.0  86.3  86.7  74.5  68.7  59.4  49.6  66.0
2002  47.6  47.0  54.8  68.4  72.0  80.1  83.2  84.7  79.6  64.9  53.8  47.7  65.3
2001  42.7  50.2  51.8  67.8  74.2  80.3  86.7  84.9  74.7  65.0  59.7  49.5  65.6
2000  50.6  57.3  61.0  65.3  76.6  80.7  87.3  90.2  80.4  69.7  49.8  39.4  67.4
1999  48.6  55.5  56.4  67.9  73.8  82.1  86.1  90.2  79.0  69.2  62.9  51.5  68.6
1998  48.4  51.1  54.9  63.8  78.5  85.5  91.6  87.8  83.6  69.5  57.6  47.0  68.3
1997  44.0  49.5  58.3  60.4  70.1  78.8  84.9  83.1  80.2  67.2  52.1  45.6  64.5
1996  43.1  52.1  53.3  64.2  79.7  82.5  86.1  82.5  74.5  66.9  54.9  49.2  65.8
1995  48.1  52.5  56.9  64.2  73.2  79.9  85.5  85.5  76.6  67.8  54.8  47.4  66.0
1994  45.7  48.6  59.0  65.9  71.5  84.1  83.9  84.0  76.3  67.3  57.9  49.0  66.1
1993  45.1  49.0  56.1  63.3  71.9  81.7  87.3  87.5  78.2  63.8  51.6  49.5  65.4
1992  46.9  54.4  59.0  66.0  71.1  79.4  84.3  80.2  77.7  69.5  52.7  49.9  65.9
1991  42.8  53.7  59.8  67.4  75.4  81.0  85.0  82.5  75.2  68.1  51.7  50.3  66.1
1990  51.8  53.9  57.7  64.0  73.4  84.0  82.5  84.6  80.0  66.4  59.8  44.0  66.8
1989  50.0  42.2  56.7  66.4  74.3  77.9  82.8  82.3  74.7  69.0  58.2  39.0  64.5
1988  42.2  47.1  56.0  64.5  72.8  80.4  85.3  87.9  79.2  65.7  58.1  49.1  65.7
1987  44.5  50.8  53.9  65.0  75.1  79.6  83.4  86.5  77.1  66.5  55.7  47.3  65.5
1986  48.8  51.2  60.2  67.2  71.5  80.8  86.4  83.4  80.2  65.7  52.4  46.1  66.2
1985  37.8  45.0  60.8  67.2  74.0  80.2  84.4  87.6  77.7  67.6  56.3  42.3  65.1
1984  39.3  50.9  56.3  63.7  73.7  82.5  85.5  85.8  76.1  67.0  54.6  52.6  65.7
1983  43.4  48.5  54.5  60.6  69.5  77.3  83.6  84.9  77.1  67.8  57.3  34.8  63.3
1982  44.6  44.5  59.8  62.5  72.5  79.2  84.6  86.7  78.1  67.0  55.6  49.2  65.4
1981  44.6  48.9  55.7  69.2  70.5  80.3  85.9  83.4  76.2  66.1  57.5  47.3  65.5
1980  45.5  46.6  54.2  63.1  75.0  87.0  92.0  88.5  80.3  65.4  54.9  49.4  66.8
bobuhito
  • 231
  • 1
  • 5
  • I don't think it can be done with any useful confidence values. Weather is very chaotic, and small inputs could have drastic results. Daily temperature forecasts are often out by more than a degree F. –  Nov 12 '19 at 22:00
  • @bobuhito why not just use a climate model ? –  Nov 13 '19 at 00:50
  • https://www.weather.gov/hun/climateforecast might be a starting point –  Nov 13 '19 at 00:52
  • The atmospheric temperature depends, basically, on: Incoming radiation (Sun mainly, which can be not me easy modeled), humidity (depends on atmosphere, so it can be modeled, altitude, biome and winds. Just with temperature values it is not possible to do this predits. If you can model all of it, then you will model it. – David García Bodego Nov 13 '19 at 07:23
  • @gansub: Because climate is not weather. – jamesqf Nov 13 '19 at 18:02
  • @BarryCarter I learned a lot from that website and it seems they mainly use sea currents and sea temperatures to get long-term forecasts, but that is not so useful for Dallas. I'm still hoping for someone to go out on the limb here and answer with a forecast for Dallas! – bobuhito Nov 13 '19 at 19:08
  • You could contact them directly. I know they publish all sorts of complex weather models, but I 'm not that deeply into weather stuff. Obviously, they have some way of predicting Dallas will be warmer in the next 1-3 months. –  Nov 13 '19 at 19:15
  • 3
    I'm extremely skeptical that weather can be forecast with any usability at that range due to the uncertainties and errors that propagate and accumulate rapidly over time. Even the long-term climate outlook (i.e. generally warmer/colder, wetter/drier than average) is difficult and problematic. Anecdotally, I've tracked one of the major forecast corporations' very long-term predictions several times, and each time the quality rapidly degrades past ~10 days out, with cooling and warming trends becoming completely out of phase with reality - i.e., worse than just assuming the climatic average. – dplmmr Nov 14 '19 at 00:18
  • @jamesqf - so weather forecasts are valid beyond 15 days ? –  Nov 14 '19 at 00:34
  • @jamesqf https://www.weather.gov/hun/climateforecast this site says " Long-range climatological forecasts are produced by the Climate Prediction Center (CPC)," –  Nov 14 '19 at 00:35
  • 2
    @gansub: Not really all that good beyond a week or so, really. It's those darned butterflies in the Amazon, you know :-) And a climate forecast won't tell you what the weather is going to be like. E.g. looking at El Nino will give you some idea about the climate in California next winter, but it's not much use in deciding whether to plan a skiing trip for President's Day weekend next year. – jamesqf Nov 14 '19 at 18:28
  • 2
    @jamesqf that was the point i was trying to make. Beyond two weeks nobody runs weather models. It is climate models such as cwrf to get likely trends. Yes you cant plan a ski trip that far out. –  Nov 15 '19 at 00:32
  • Weather isn't stock market, it doesn't follow trends (climate does). Two months weather forecasts don't exist. Medium range forecasts run up to 10 days, but you'll have to compete with a few supercomputers and the people that design and run them, as well as data collection and assimilation. Which is rather futile for an individual. Long range forecasts are rather experimental, and very coarse. –  Jun 13 '20 at 17:16
  • @a_donda Fine, you can give up, but the point of my question is that weather is deterministic. I want to hear why Dallas had a hotter-than-expected January and I'm sure someone could explain it (maybe based on some water current) in hindsight. – bobuhito Jun 13 '20 at 17:57
  • Yes that's what we're trying to transport. Weather is not determinsitic. No complex system is. To handle these things probability was introiduced in math in ancient times. And we're in the balpark of 50% for a 10 day weather forecast (depends). Determinism is a belief, not a working principle. https://en.wikipedia.org/wiki/Statistics –  Jun 13 '20 at 19:17
  • @bobuhito: Here's am introduction into a rather recent numeric weather model: https://www.dwd.de/EN/research/weatherforecasting/num_modelling/01_num_weather_prediction_modells/icon_description.html (ICON model, German Wetterdienst). The ECMWF has texts on medium range (~10 days forecasts). I may be wrong, but i think the NOAA's GFS model is rather dated and under revision or to be replaced, but I'm sure you can find something there, too. And listen to guys like gansub, I believe they know what they're talking about. –  Jun 13 '20 at 19:39
  • From that lecture you may get an impression how difficult it is to predict weather for such a small area as a single city several days into the future, and what processing capacity and data heap is needed to start from. –  Jun 13 '20 at 19:42

0 Answers0