August 2016 Projected Temperature Anomalies from NCEP/NCAR Data

Guest Post By Walter Dnes

In continuation of my Temperature Anomaly projections, the following are my August projections, as well as last month’s projections for July, to see how well they fared.

Data Set Projected Actual Delta
HadCRUT4 2016/07 +0.793 +0.736 -0.057
HadCRUT4 2016/08 +0.754
GISS 2016/07 +0.86 +0.84 -0.02
GISS 2016/08 +0.85
UAHv6 2016/07 +0.327 +0.389 +0.062
UAHv6 2016/08 +0.334
RSS 2016/07 +0.407 +0.469 +0.062
RSS 2016/08 +0.374
NCEI 2016/07 +0.9575 +0.8719 -0.0856
NCEI 2016/08 +0.9206

The Data Sources

The latest data can be obtained from the following sources

Miscellaneous Notes

At time of posting, all 5 monthly data sets were available through July 2016. The NCEP/NCAR re-analysis data runs 2 days behind real-time. Therefore, real data through August 29th is used, and the 30th and 31st are assumed to have the same anomaly as the 29th.

August will be the 13th consecutive month that sets a new record for that specific calendar month. I.e. August 2015 was the hottest August in NCEP/NCAR data to that time; September 2015 was the hottest September to that time; October 2015 was the hottest October to that time, etc. NCEP/NCAR data goes back to January 1948.

The global NCEP/NCAR anomaly (HadCRUT/GISS/NCEI) and the UAH-proxy subset anomaly were little-changed from last month. Most of the globe cooled slightly, while Antarctica warmed significantly. Because the RSS analysis only reaches down to 70°S, RSS missed the Antarctic warmth, and its proxy NCEP/NCAR anomaly fell noticably.

dailyanom

The graph above is a plot of recent NCEP/NCAR daily anomalies, versus 1994-2013 base, similar to Nick Stokes’ web page. The trendlines are as follows…

  • Black – The longest line with a negative slope goes back to early August, 2015, as noted in the graph legend. This is near the start of the El Nino, and nothing to write home about. Reaching back to 2005 or earlier would be a good start.
  • Green – This is the trendline from a local minimum in the slope around late 2004, early 2005. To even BEGIN to work on a “pause back to 2005”, the daily anomaly has to drop below the green line.
  • Pink – This is the trendline from a local minimum in the slope from mid-2001. Again, the daily anomaly needs to drop below this line to start working back to a pause to that date.
  • Red – The trendline back to late 1997. Again, the daily anomaly needs to drop below this line to start working back to a pause to that date.
Advertisements

85 thoughts on “August 2016 Projected Temperature Anomalies from NCEP/NCAR Data

  1. I don/t intend to be critical, but how significant is the difference between 0.0327 and 0.334 which is 0.007
    UAHv6 2016/07 +0.327 +0.389 +0.062
    UAHv6 2016/08 +0.334

    Why do we claim such accuracy when the measurement error is probably orders of magnitude greater?

    • Why do we claim such accuracy when the measurement error is probably orders of magnitude greater?

      a) I’m trying to give an “apples-to-apples” comparison with the 5 data sets. That means matching them digit-for-digit.

      b) The ultimate counter-argument is that removing a few digits of precision will leave me only able to project +1 or 0 or -1, which would be pointless (sorry about that).

      There is a differemce between error margin and precision of the calculation. I recognize that the error margin is rather high. Currently, I’m happy to get within +/-0.1 of the actual value.

    • “Why do we claim such accuracy when the measurement error is probably orders of magnitude greater?”

      Its not an accuracy claim.

      lets keep it simple.

      I have a scale

      It has a precision of 1 pound.

      You step on the scale

      200 lbs

      you step off

      you step on again

      201 lbs

      Now I ask you the question

      Please predict your actual weight if I weigh you on a perfect scale.

      I want you to MINIMIZE the error in your prediction.

      Go!

    • If the people paid to study this weren’t making a point out of 1/100 K they would have say there isn’t anything happening – in any direction.

    • @caretaking is Exactly Correct, hundredths of a degree is the throw-away part for our most accurate, well placed digital sensor, and those are as rare as hen’s teeth. in mercury thermometers the tenth of a degree is the throw away part.

      • There are too many inaccuracies to count. It would be better if the metric used was actually a measure of atmospheric heat content, which is after all the value claimed to be of concern. So show the graph in kilojoules per kilogram, It will be seen once you try to use the correct metric that the entire ‘average of an intensive variable’ approach is unsupportable, let alone arguing about differences of thousandths between measures of the incorrect units.

    • I don/t intend to be critical, but how significant is the difference between 0.0327 and 0.334 which is 0.007

      If your a global warming nut the difference is about $100 Billion dollars.

  2. I am confused as to the point of these posts? Even I could manage to predict the average
    temperature for a month given 28 days of data. And anybody with a knowledge of Taylor
    series would realise that any smooth function can be approximated by a straight line over a
    sufficiently small interval (which is, as I understand it, how Walter Dnes is doing his projections).

    • The point is to get the estimates NOW, rather than waiting a couple of weeks for GISS and NCEI, or 4 weeks for HadCRUT.

      • So there’s people that are in such a hurry to know what was the average global temperature last month? I hope it is not an insane obsession and they have some real reason that I cannot imagine.

    • I must confess that whilst I always read this series, I too am confused as to the point.

      I can see no advantage of being ahead of the game by days or weeks. Why not simply wait until mid to end September to see what the August data set reveals. We will within the next days get the UAH release.

      Of more interest to me is whether the data sets are accurate and whether they are noting anything of substance.

  3. The north of 80° arctic temperature is usually pretty close to the average during the melt season. This year we’ve had a bump up in the last ten days or so. 1998 also had similar temperatures at this time of year. link I find this surprising because I thought the melting ice would moderate the temperature.

      • Nick

        Let’s get real. There is right now 4.5 million sq. km of ice in the Arctic Basin – that’s half of Canada’s [the world’s second largest country] land area and within +/- 2 1981-2010 SD.

        Take out the Arctic cyclone induced 2012 outlier minimum and you’ll notice that 2007 was the turning point when losses stopped – not only that, we know from several sources that there is an increase in all important multi year ice.

        This on the 10th anniversary of Gore CAGW/CACG propaganda movie…

      • Maurice Ewing was of the opinion that open water N of 80˚ was in fact the cue for for cooling as ocean-borne heat got the chance to ’emit’ into the atmosphere at sub zero temperatures rather than be insulated by an ice sheet.

      • Whilst not quite the latitudes you are talking about, there does seem to be plenty of ice up there.

        See: http://polarocean.co.uk/

        This little Irish boat should be given the Freedom of Westport. It has travelled through the Irish Sea, North Sea, Norwegian Sea, Barents Sea, Kara Sea, Laptev Sea, East Siberian Sea, Chukchi Sea and now the Beaufort sea. Each sea has had its quirks. Laptev the hardest with the ice, and then closed behind us. But, for some odd reason, I thought the Beaufort would be the easiest. None of it. In terms of hard sailing, this is proving difficult. I want my shower in Tuk, but it’s so slow.

    • and the Arctic ice cover is already second lowest, and perhaps heading lower

      (since 1979, not the whole Holocene)

      • If the Greenland Ice Cores are anything to go by, the present amount of ice must be high in comparison with the past. See:

      • “the present amount of ice must be high in comparison with the past”
        No, the amount in 1855 (last data point) may have been high in comparison with earlier times.

      • Nick

        I am talking about earlier times, eg., the peaks of the Medieval, roman and Minoan Warm Periods noted in the data plot.

    • Aug 2016 will be 0.07 deg warmer than Aug 1998 making it hottest Aug on record using Wood for Trees Index (WTI)

      • Interesting but meaningless other than for CAGW/CACC propaganda purposes.

        Winter 2009-2010 there was snow on the beaches on the Cote d’Azur [ France] and winters 2010-2014 the UK was covered by snow and ice top to bottom, and the past five years or so have seen scores of new ltemp minima and snow/ice records set across North America. Early August this year there was frost on the ground and new temp minima in central Germany. Not to mention more ice in Antarctica.

        You don’t see that acknowledged or discussed in the MSM – it doesn’t fit the climate establishment’s narrative.

      • tetris on September 1, 2016 at 6:15 am

        Interesting but meaningless other than for CAGW/CACC propaganda purposes.

        Wow. A person writing with a slightly arrogant undertone should imho produce relevant and menaningful comments.

        You make the typical skeptic mistake of confounding local and global, by giving us a sequence of cold spots here and there whilst everybody talks here about the globe’s mean temperature.

        It is, inverted, the same mistake as those made by persons convinced that the 1930’s were the warmest decade on Earth, and 1934 even the warmest year since GISS’ beginning, what even isn’t true for USA (it was in fact 1901).

      • A little correction: 1901 is indeed a USA top temp, but only in mothly records (°C, absolute):

        1. 1901 7 25.46
        2. 1936 7 25.07
        3. 2012 7 25.03
        4. 1934 7 24.86
        5. 2006 7 24.68

        For yearly records, USA top temps look like this:

        1. 2012 14.13
        2. 1998 14.01
        3. 1921 13.80
        4. 1880 13.76
        5. 1934 13.74

        Source: NOAA’s GHCN, unadjusted variant

      • If you make the warranted adjustment for the UHI effect and other land use changes, 2012 and 1998 fade into the pack.

        Naturally, NASA’s UHI adjustment Al-Gore-ithm actually warms the raw data instead of cooling them, as it ought to do.

  4. We run statistical models…. and this is what they say…

    Aug Temps will come in slightly higher than July… mostly noise level.

    Hadcrut +.01
    GISS +.06
    RSS +.01
    UAH +.05
    WTI +.03

    Our forecast for the next 12 months is -0.23 deg as unwinding continues from the recent super el nino.

    Our forecast for the next 5 years is -0.15 deg… so that indicates a return to slight warming after the post el nino cool down.

    Also, we find that El Ninos can cause huge global temperature spikes but La Ninas tend to produce modest cool dips… not inverse downward spikes. So the Ninos are short spikes up while Ninas are more modest, longer lasting but modest dips.

    Skill scores for our one and 5 year forecasts are quite high with Rsq > 0.50

    • Now would be a good time to bet somebody that the temperature average for 2021 will be lower than in 2016. The only thing that would derail that would be another super el nino at just the wrong time.

      I would estimate that prob T2017 < T2016 = approx 90%
      I would estimate that prob T2021 < T2016 = approx 80%

      The 1-5 year time frame is dominated by mean-reversion factors and after the Feb 2016 temp spike, we have a lot of mean reverting left to do.

      • Only a fool would make accept such a bet given the large intrinsic fluctuations in the annual
        temperatures. How about a bet that the five year period from 2021 will be warmer than any five
        year period in the 20th century?

      • Re:Willis

        You dont seem to understand a thing I have said…

        I would be happy to send you the monthly forecasts back to the beginning of the model data… Along with the equations that generated the forecasts

        The skill on in-sample data is obvious beyond 99.9% certainty. There is no out of sample data since it new.

    • Mary Brown August 31, 2016 at 8:23

      We run statistical models…. and this is what they say…

      Thanks for that, Mary. While that looks impressive, without a citation or a reference to show ALL of your past predictions and their outcomes it is impossible to know whether it is significant or not …

      It is also totally unclear how you are measuring the “Rsq” of the forecast. What exactly are you measuring the R^2 OF?

      Please excuse my skepticism, but you are far from the first person to show up here with a fistful of anecdotes about their whiz-bang method … you desperately need to back them up.

      w.

      • “ALL of your past predictions and their outcomes it is impossible to know whether it is significant or not …”

        That’s pretty much the same with all climate forecasts. The sample size is so small that they are not bounded by reality like weather forecasts. In weather forecasting, you can’t just keep busting too hot for 30 years and not calibrate your models. Apparently, in climate forecasting, that’s not a problem.

        Our short-term climate models are pretty straightforward and work quite well. They are vulnerable to el nino spikes (underforecast them) and clueless about volcanoes (aren’t we all). They have no value at climate outcomes beyond 5 years.

        The Rsq are simply from the regression analysis… forecast temps vs obs temps. About half the variation is explained by the short term statistical model. Data in the model goes back to 1911… although the equations are different depending on the available input.

        Here are a couple of graphics showing some of the historical data and forecasts

        One year forecasts vs observed

        Error bars on the one year forecasts. Interesting to see the biggest error times.

        Oh, and

        “… you desperately need to back them up.”

        Not really. I’m not trying to prove anything. But we do have a really good idea where temps are headed over the next few years. Eventually, we may expand our systems and add longer range forecasts and publish a record of the forecasts in real time. Competing against existing GCMs should be easy.

      • Thanks, Mary. Unfortunately, you haven’t answered the questions. It’s easy to post up a graph, but without a record of when it was posted we can’t tell whether you just made the “forecast” for 2015 last week … your forecast looks good, but anyone can make up a graph.

        Please be clear that I’m not accusing you of anything but a lack of clarity … for example, you call it “one year forecasts vs observed”, but it is unclear whether you made month-by-month forecasts one year ahead, or you forecast the entire upcoming year at once, or something else.

        For example, it APPEARS as though you are forecasting 2015 temperatures, including the El Nino … and the claimed date of your post is September 2013.

        Sorry, I’m not buying that for one moment. I do NOT believe that you were able to exactly forecast the timing (and do well on the size) of the 2015 El Nino back in September 2013.

        w.

      • RE: Willis Eschenbach … I’m not trying to convince anybody of anything. The model is new and anybody can fit anything to the past. The Irish Population is highly correlated to global temperature.

        But, it obviously works for 1-5 years forecasts. Time will tell. We’ll know for sure about the time I’m dead.

        The model makes one year forecasts, once a month. Right now, the forecast is -0.23 deg for the next 12 months.

        Willis ….
        “For example, it APPEARS as though you are forecasting 2015 temperatures, including the El Nino … and
        the claimed date of your post is September 2013.”

        I’m not sure what you are referring to. The model clearly way underforecast the Feb 2016 el nino highs so it hardly nailed that. The graph is hard to read…because the forecasts are always just 1 year out so they obviously stay with the overall trend. It is easier to understand looking at the raw numbers.

        The model did forecast > .10 deg one year warming each month from for much of 2015…clearly seeing the el nino bounce coming. And in Dec 2015 it started forecasting year over year cooling. In Feb 2016, it was screaming a -0.50 one year drop. That is likely to come true by Feb 2017 since temps are already down .42 from the Feb peak.

        Also, interesting to me was that the biggest bust was the Pinatubo year. That suggests the volcano DID cool the planet briefly but significantly… but that needs a much closer look. A similar bust happened in 1963 when Mt. Agung blew. Could be coincidence.

        I’ll send you the code. I think you’ll find it interesting. It’s all quite new and hardly documented for publishing… but it’s not our first rodeo.

      • Mary Brown September 1, 2016 at 12:30 pm

        RE: Willis Eschenbach … I’m not trying to convince anybody of anything. The model is new and anybody can fit anything to the past. The Irish Population is highly correlated to global temperature.

        But, it obviously works for 1-5 years forecasts. Time will tell. We’ll know for sure about the time I’m dead.

        No, it doesn’t “obviously” work for 1-5 year forecasts. You CLAIM, without a single scrap of evidence, that it works for 1-5 year forecasts.

        I’m sorry, but claims of wonderful performance just like yours can be found all over the web. Yes, you’ve put up some pretty graphs, including a graph dated 2013 that exactly predicts the El Nino in 2015-16 … bad news. I don’t believe that in 2013 you predicted the El Nino. Call me skeptical, but I don’t buy a word of that.

        Which is why I asked for details such as a link to a dated forecast, which you have not supplied. Be clear that I’m not saying you are wrong. I’m saying you have not come anywhere close to establishing your claims.

        Best regards,

        w.

      • “No, it doesn’t “obviously” work for 1-5 year forecasts. You CLAIM, without a single scrap of evidence, that it works for 1-5 year forecasts.”

        Actually, I have all the code, all the predictors, and all the forecasts which I have offered to share. The predictors are well known atmospheric variables, the statistics are straightforward, and the statistical results show skill significant beyond 99.9%.

        But you don’t seem interested in that… just dismissive and combative.

        So nevermind.

      • Mary, thanks for your kind offer to send me the code that you used. I’d love to see it, please send it to me (willis) at surfacetemps.org.

        I would also, however, like to see a record of your forecasts and when you made them. Your graphs are less than informative.

        Next, you say “The model makes one year forecasts, once a month. Right now, the forecast is -0.23 deg for the next 12 months.” But in the graphs you posted, you do NOT show a forecast for the coming year. Instead you show month-by-month forecasts, which is confusing.

        All the best, I look forward to receiving your code.

        w.

        PS—You keep talking about “we” and “our model” … who is the “we”.

      • Mary Brown has been kind enough to share her model with me, big props to her. It is a nine parameter model … something I’m not fond of, but that’s not the real problem.

        The real problem is that to judge whether a forecast is “skillful” or not, it’s not enough to just look at the results. One common method is to compare the results to what is called a “naive forecast”. This is a forecast done by merely extending a simple trend of the data itself.

        The bad news is … the “naive” forecast outperforms her model, with a standard deviation (scatter) of the residuals (errors) for her model being 0.20, and that of the “naive forecast” being 0.18 … and that is without any attempt to improve the naive model by using a specified-length trend. That’s using the linear trend from the start of the data, unaltered.

        I also suspect that I could do better by using the trend of the gaussian average of the data, which is another kind of “naive” model that depends solely on the extension of a linear trend into the future.

        Again, my thanks and respect to her for being willing to have her model put to the test.

        w.

    • I assume those numbers are versus July. The only major disagreement I have is RSS. The problem is that RSS covers 82.5°N to 70°S, and will miss the warming in Antarctica. RSS should be down a bit in August, according to my numbers. I run 3 sets of NCEP/NCAR data. A global set for HadCRUT/GISS/NCEI and separate subsets to emulate the coverage of the UAH satellite data and the RSS satellite data.

      • Thanks for the article. I thought we were the only people who forecast this stuff :-) …. Our model is not that sophisticated, but it gets us pretty close to where the end of month values will come in.

    • Geronimo says…

      “Only a fool would make accept such a bet given the large intrinsic fluctuations in the annual
      temperatures. How about a bet that the five year period from 2021 will be warmer than any five”

      Our models forecasts average WTI temp 0.34 next 5 years which is higher than any previous 5 in the consensus global temperature database. So, as the brief el Nino spike fades, the overall slow warming is likely to continue.

      Last five years were 0.28. That’s up about 0.70 deg since 1950 at the dawn of the fossil fuel era. Warming has been about a degree per century rate

  5. Guest Post By Walter Dnes

    In continuation of my Temperature Anomaly projections, the following are my August projections, as well as last month’s projections for July, to see how well they fared.

    Walter, thanks for your post. However, today is August 31st … and the date of publications of your predictions for August also August 31st.

    On what planet is making predictions for August on the thirty-first of August an actual thing?

    You have also linked to your “projections for July” … which were published on July 31st.

    What am I missing here?

    Next, a very common way to assess the skill of a forecast is not to compare it to observations, but to compare it to what might be called a “naive forecast”. This is a forecast made by simply extending the trend. Particularly in heavily autocorrelated data, such as is common in climate science, such a naive forecast is often very accurate. This is because autocorrelation means that the present is like the past, and thus the future will be like the present.

    So you still have some work to do. You need to link to the place where we can verify WHEN you made the prediction. It’s not that we don’t believe you … it’s that science is not about belief, and we’ve been scammed too many times to take anything on trust.

    Then you need to figure out how you are going to evaluate your predictions. The method you are now using (comparison to observations) is totally inadequate. For example, is it impressive if the sun rises at 6:02 am today, and I predict it will rise at 6:03 am tomorrow? I mean, I can get it almost exactly to the minute, is that a skillful prediction?

    Well … no. We could do better with a simple linear extrapolation of the trend, and likely get it within a few seconds.

    Please take this in the intended spirit, which is not to tear your work down but to encourage you to improve it. You might enjoy this look at forecast analysis, or this discussion of the mean square skill score (MSSS) forecast analysis method. Another relevant work is “Forecast and analysis assessment through skill scores“. The TL;DR version is this:

    There is plenty of established literature on assessing the skill of forecasts … reading it would greatly improve your work.

    Best regards, and thanks for all your work,

    w.

    • “You need to link to the place where we can verify WHEN you made the prediction.”
      The link on the second line is to the previous post, where the July predictions were made. And that post has a link to the previous, and so on. Walter is not predicting NCEP/NCAR for August – 29 days of data are in. But there is still prediction, informed by NCEP/NCAR, in terms of what GISS etc will say for August.

    • Walter, thanks for your post. However, today is August 31st … and the date of publications of your predictions for August also August 31st.

      On what planet is making predictions for August on the thirty-first of August an actual thing?

      You have also linked to your “projections for July” … which were published on July 31st.

      What am I missing here?

      Note that I use the word “projections”, not ‘predictions”. NCEP/NCAR temperatures are influenced by the same atmosphere as the 5 data sets. They are co-dependant variables. Using the NCEP/NCAR data to the 29th (it runs 2 days behind) I project the value of the 5 data sets for the month. I do NOT claim to be making a meteorological or climatological forecast. Rather, I try to get a preliminary estimate out a bit earlier than the official release dates.

  6. I’m just going down to the Pacific Ocean right now…to walk the dog. (it’s 4.40pm here)
    I’ll let you know if it’s a hundredth of a degree warmer or colder…or a mm higher or lower, as soon as I get back.

  7. Thermalization explains why CO2 (or any other noncondensing ghg) has no significant effect on climate.

    Global average water vapor is increasing. This has a warming effect which is countering the on-going cooling effect of dwindling numbers of sunspots and declining average sea surface temperature (declining temperature phase of the net of ocean cycles). Sunspot time-integral plus net of all ocean cycles plus effect of water vapor increase provides a 98% match to measurements 1895-2015 at http://globalclimatedrivers2.blogspot.com

    Increasing water vapor increases the probability of heavy rain and flooding.

    Switching from coal to natural gas adds water vapor.

  8. UAH for August is up a little at +0.44degC.

    The July figure was +0.39.

    Similar happened following the Super El Nino of 1998. Temperatures started falling then went uip a little and then fell in earnest.

    Presently the La Nina which may well follow the strong 2015/16 El Nino, has yet to take hold. It will be interesting to see what the next 6 to 9 months has in store.

    • The running, centred 13-month average in UAH, shown as the red line on Dr Spencer’s chart, has topped 1998 for the first time (now 0.48 versus the 0.46 peak in 1998): http://www.drroyspencer.com/wp-content/uploads/UAH_LT_1979_thru_August_2016_v6.jpg

      That running 13-month figure will rise again in September unless the UAH anomaly comes in below 0.25C. The red line in Dr Spencer’s chart also gives a good example of the point Bindidon makes below. The longer term lower troposphere temperature response to La Ninas that follow big El Ninos is typically not as strong as it is to La Ninas that follow less strong El Ninos.

      • Irrespective of whether 2016 beats 1998 as the warmest calendar year, this latest UAH anomaly update sets a new warmest continuous 12-month period record. Sept 2015 – Aug 2016 averaged 0.496C, beating the previous joint record of 0.483C, set during the 12 months ending May then June 1998.

        If the September UAH anomaly is 0.240 C or above then this new warmest continuous 12-month record will be beaten again.

  9. Late or weak La Niña. Who knows.

    But the 2016 Hurricane Season sure looks like another dud.

    4,000 -days becoming likely!

  10. I’ve been through a lot of UK summers and whatever anybody may try to say, this year’s summer was awful both less warm than most and wet. I’m a motorcyclist and my riding has been restricted this year. When I look at this year compared with the mid 70’s and 1998 it has been a big nothing. I’ve also had none of the very warm nights where sleep was impossible or where I could do long rides overnight that were hallmarks of warm summers.

Comments are closed.