Monday, February 15, 2016

Prospects for 2016

We have very little information yet about 2016, but we do have a possible pattern in the 1997/8 El Nino. I started thinking about this post when I noted that surface measures either diminished or advanced by a little from December to January. I thought that might be a deviation from the 1997/8 pattern, but I found that it was similar then.

Satellite data rose by quite a lot, but less than in 1997/8, though it started from a warmer base. I showed a plot of RSS in that time here. So in this post, I'll develop those plots for a range of indices, showing more detail of the current evolution. I'm thinking also about the prospects for 2016 being a record year. I think it is quite likely, though satellite measures are less well placed, and more volatile. So there is an interactive plot below the jump.

I'll also take the opportunity to digress on some updates to the data pages. The two active plots that show monthly data are here and the trend viewer. I have rearranged the display for better ordering of datasets, and to make it easier to make additions. I have added, or restored, UAH V5.6, UAH V6 TMT, and a newish TMT measure from NESDIS, NOAA. I think there will be a general shift to the Mid-Troposphere level, partly because of declining infrastructure to support TLT.

For the active plotter, I have changed so that if you ask for data in trendback more, it shows the trends as plotted, not the time series.

I have added an axis on the diagonal of the trend viewer. This is to help identify the various events (mainly ENSO) that it shows.

So here is the plot - use the button at the top to cycle through the datasets. It shows months of 1998 in red, 1997 from July in blue, and of 2015/6 in cyan. The annual means for 1998,2010 and 2015 are also marked. I'll comment below:

GISS shows a 12 month plateau for 1997/8, from Sep to August, and then a modest decline. There was a peak in February, but January rose little from December. This time the pattern is similar, but nearly 0.5°C higher. So it will almost certainly be warmer than 1998. The test is beating 2015, where it is now about 0.25°C ahead. Looking at the plateau shape of 1998, if that is followed again, a record year, possibly by a lot, is very likely. Much the same can be said for HADCRUT and NOAA.

For RSS (and UAH) 1998 was the warmest year. 2016 starts out with January about 0.1°C warmer. 1998 had two months (Feb and April) quite a lot warmer, but otherwise also pretty much a plateau starting in January. 2016 has ascended that plateau, and only has to stay there. On the other hand, the end year drop-off in 1998 was greater, and if that came earlier or stronger, 2016 woyld not be warmer. Again, this also applies to UAH

Land temperatures (Bestla and CRUTEM) warmed a lot in late 2015, and are far ahead of the 2015 average. In both cases there is a strong peak in 1998 Feb, but otherwise mostly flat, with warm Mar-Aug, down at the end. A record is very likely here.

The SST patterns (HADSST, NOAAsst) are similar but more even, again with a 12-month plateau. SST for 2015 was generally warm, so a record is more of a challenge. I December, SST was about 0.1°C ahead of 2015. So a record is still favored, but there isn't much in reserve if there is any early cooling.


  1. To me the biggest potential difference is where the PDO is now versus then. In 1998 the positive phase of the PDO had been in decline for a long time, whereas the current PDO could be in the initial stages of a positive phase. It's jumped up in January. If the PDO index goes up and stays up throughout 2016, then 2016 could actually produce a very high anomaly. If the PDO cascades downward, which it did in 1998, then 2016 is going to have a harder time knocking off 2015.

  2. Nick,

    The Problem is you assume the way that 2016 takes is the same like 1998 and 2010, this is for now very unlikly. In the same time where 1998 goes up for La-Nina (in respect to recent upper ocean heat-content in the tropical pacific), 2016 does not, it seems more like a central-pacific-El-Nino-Pattern over the rest of 2016, this is because upper wind-pattern cause even more WWBs (West-Wind-Bursts) in the central pacific basin.

    So the Nino 3.4 double Peaks because of the strong WWBs in early January and no wonder is, that CFSv2 is forecasting above El-Nino-Threshold to the end of 2016 or look here: and in contrast (but typical for a Central-Pacific El-Nino) the Nino1.2 goes down to La-Nina-Threshold

    If you looking for Windanomalies this Pattern is forecast to get stronger: and in context to 1998 just look at the ocean heat content

    And the negativ anomalies in 2016 will become less or will resplaced positive by the new WWB and due possible new downwelling-Kelvinwave

    So i would not assume that the downkick in Global Temperature in the End of 2016 is strong as its 1998

  3. That's a really nice plot, showing in detail what we could intuitively assume concerning temps from satellites and stations.
    By the way, you said earlier Hadcrut may be colder in january. That will effectively be the case, as JMA, which is very similar to Hadcrut (same stations) just released 0,91°C, a drop from december (+1,04°C). Maybe NOAA will be in between. Arctic unbelievable warming in january (+7°C) probably the cause of the difference, as Hadcrut and JMA are not covering the region.

  4. Nice graphs. So far 15/16 has followed 97/98 reasonably well. Like 1997 surface temperatures rose sharply at the end of 2015, so 2016 is now lapping much cooler readings from early 2015. That almost guarantees a record in 2016 even if the nino decay in 2016 differs somewhat from 1998.


  5. So let's start to make predictions for the 2016 temperature. I will do it for the GISS global anomaly, by regression.
    So I predict for Feb 2016 +1.2°C.
    For the year 2016 I predicted +1.00°C after the Dec 2015 data were out. With the Jan 2016 data are now out I predict now 0.99°C.

    1. For Feb, looking at the NCEP/NCAR index, it will need to get warmer than so far this month. Which it easily could do.
      For 2016, yes, your prediction looks quite achievable.

    2. I'll go with GISS 0.99 to 1.00 for 2016. I'm basing this on a guess that the 2016 ENSO follows the 1998 portion of the 1997-1998 ENSO. Anomalies should then be in the range of those of October '15 - February '16 until August, for the rest of the year, the anomalies should be in the range of the 2015 average.

    3. I have updated my regression model to avoid over-fitting, using the AICc.
      The new model gives for GISS 2016 1.02+-0.10°C before the Feb data are out.
      With the GISS Feb 2016 data it is 1.04+-0.10°C for the year 2016.
      For GISS Mar 2016 the model gives now 1.06+-0.25°C, the old over-fitted model predicted a strong cooling to 0.95°C

    4. After the GISS March data are out I update my prediction for GISS 2016 to 1.10+-0.08°C.
      Unbelievable high.

    5. After the GISS April data are out I update my prediction for GISS 2016 to 1.07+-0.08°C. The seasons are
      MAM 1.14+-0.04°C
      JJA 1.02+-0.13°C
      SON 1.09+-0.16°C

    6. Interesting - warmer in SON than JJA?

    7. Yes, probably due to changing annual cycle since 1951-1980. From May to August around 1.0, then rising in Sep and Oct, and may be a large drop in Dec. I should note, that I tried to predict MEI the same way and there is no hint of La Nina, at least until Nov (The El Nino decays after May). But the prediction skill is not very good more then two or three month from now.
      If there is really a La Nina developing this year, the temperature could be lower.

    8. I have improved my regression model using PCA, full MEI data and QBO (there was a visible 2 year period in the residuals for the full year).
      My prediction using the new model is now:
      MAM 1.16+-0.05°C
      JJA 0.99+-0.12°C
      SON 1.05+-0.13°C
      2016 1.09+-0.06°C
      If 2016 really became so hot, then it will take many years to break the 2016 record.

    9. "I have improved my regression model using PCA, full MEI data and QBO (there was a visible 2 year period in the residuals for the full year)."

      Uli, OK, now you have my interest. MEI (ENSO) data is a lot different than QBO data. The QBO data is very close to a 2.33 year period, and that period does show up secondarily in the ENSO cycle, but ENSO is overall a longer cycle than that so the applicability is subtle.

      Where are you working out your model? I do everything at the site and

      As a disclaimer, I am not the least interested in prediction, because that is only a side-effect of getting the physics correct. And I really am not going to waste my time waiting for predictions to come true! There are many other ways to verify the applicability of a model.

    10. I have to retract my improved model from May 21, because of strange, probably erroneous, behavior. F.e. most month go down, but the year goes up, and the result would change completely, if I do a prediction for the remaining month only. I have to rework it. So I go back to the previous model, where the last results posted at May 18.

      I do not actually use QBO data but it discovered a 2 year tic toc in the residuals of my model. So I incorporated a simple 2 year periodicity which improves the fit.
      My model is not a physical one and originates from a very simple regression model which tries to predict the GISS temperature for the actual year earlier using the temperature of the month that is already known. It is purely statistical and has no physical meaning. I simply want to know early, if the actual year becomes a new record or not.

    11. I tried to improve the old regression model, results for the year 2016.
      old version from May 18: 1.07+-0.08°C
      marking volcanic years: 1.06+-0.07°C
      using 2 year period: 1.04+-0.06°C (best estimate now)
      I'm curious about odd and even years are so different, odd years are usually warmer except for Feb and Mar, where even year are warmer in this two month.

    12. Uli said:
      "I do not actually use QBO data but it discovered a 2 year tic toc in the residuals of my model. So I incorporated a simple 2 year periodicity which improves the fit.
      My model is not a physical one and originates from a very simple regression model which tries to predict the GISS temperature for the actual year earlier using the temperature of the month that is already known. It is purely statistical and has no physical meaning. I simply want to know early, if the actual year becomes a new record or not."

      Nothing wrong with using a purely mechanistic fit. If there actually is a biennial signal, it doesn't matter how you discover it.

      Recently some scientists found a 2 year signal in earth deformation parameters extracted from GPS data.

      Pan, Yuanjin, et al. "The Quasi-Biennial Vertical Oscillations at Global GPS Stations: Identification by Ensemble Empirical Mode Decomposition." Sensors 15.10 (2015): 26096-26114.

      And these guys aren't even climate scientists or geophysicists!

      Read the latest posts on my blog, including this one

      There's no question that there is a strict biennial signal underlying ENSO. The only tricky thing about is the nature of the metastability. Since there is no real favoritism between odd-year and even-year cycles, how the behavior locks into it is really the nagging question.

    13. After the May values are out a new prediction.
      Much faster, but less accurate regression model.
      Old model for last month 1.04+-0.06°C. New model for last month 1.02+-0.07°C. So it is lower due to the new model.
      New prediction using the new model.
      JJA 0.89+-0.13°C
      SON 1.00+-0.15°C
      DJF 0.79+-0.21°C
      2016 1.01+-0.07°C
      The prediction of MEI values are still bad for more than a view month, but now negative values are predicted from September. So it seems more open, if the year 2016 will be above 1.0°C or not. It is not due to a single effect but due to a combination of effects that accumulates over the last 2 month, each lowering the predicted value by 0.01 to 0.02°C.

    14. I've run the old model for data until May16.
      I got for the year 2016 1.02+-0.06°C.
      I also experimented with a model which uses also NH data to predict global temperature anomaly. The first run got for the year 2016 (global) 0.97+-0.06°C.
      Depending on the inclusion of different data sets I got now values between 0.96 and 1.04°C for the best estimate.

    15. After the large (SH) cooling in June to 0.79°C(global) the new model gives now 0.98+-0.06°C for the year 2016.

    16. Uli - looks like July is rebounding, and La Nina odds are waning. Both PDO and AMO remain solidly positive. .98 ℃ is plenty high.

    17. After the July data are out the new model gives now 1.00+-0.05°C for the year 2016.

    18. After the August data are out the new model predicts now 1.03+-0.04°C for the GISS anomaly of the year 2016.
      For SON 2016 it predicts 1.01+-0.13°C. So this also, may be higher than the record last year (SON 2015 was 0.96).

    19. After the September data are out the new model predicts now 1.00+-0.03°C for the anomaly of the year 2016.

    20. For SON 2016 it predicts 0.95+-0.08°C.

    21. Well, so far in 2016 no La Niña, though it's still possible. The AMO has stayed high; despite the decay of El Niño, the JIASO PDO remains positive. Uli, your model is good. What are the odds of 2017 staying close to 2016? Looks possible.

    22. The prediction for 2017 is now 0.88+-0.14°C.

    23. After the September data are out the new model predicts for GISS anomaly
      J-D2016 0.99+-0.03°C (year 2016)
      D-N2016 1.02+-0.02°C
      J-D2017 0.88+-0.14°C (year 2017)

    24. Uli - J-D2017 0.88+-0.14°C (year 2017)

      Which would make 2017 the 2nd warmest year in the record... would do little harm to Gavin Schmidt's models versus observations graph.

    25. GISS June through October average is also 0.88 as temperatures appear to have stabilized after the large spring drop.


    26. 2016 had a warm start and temperature drops after spring. 2017 may the other way around. If 2017 becomes really as warm as 2015 depends if the late rise really happens.
      The model predicts
      DJF17 0.78+-0.17
      MAM17 0.83+-0.19
      JJA17 0.84+-0.16
      SON17 1.02+-0.17

    27. Perhaps should continue here rather than going OT at ATTP's.

      2017 at 0.88 following 2016 at 0.99, so a 0.11C drop. That seems plausible to me. I feel like I can equally well come up with reasons why the drop will be on the small and large side. On the large side is the Arctic, which has seen extreme warmth this year. Regression to the (rising) mean would dictate a decent probability of a large fall in temperature, possibly more than 1C in 2017, which would translate to around -0.1C for the global mean, somewhat secular from a typical post-El Nino drop.

      For that reason I'm content to consider a range rather than preferring any particular estimate. About 90% of that range would sandwich 2017 in between 2014 and 2015 as 3rd warmest year.

      Typically that pattern of cold SST anomalies in the West and Central North Pacific would be a classic positive PDO indicator, though I don't know if it's as simple as that.

      I really can't find any reason to trust ENSO forecasts at all.

    28. There are two apparent reasons: religion and politics.

    29. November seems to be warmer than October according to
      so it depends on December how close we come to 1.00 for 2016. Remarkable is the very warm air above the Arctic ocean and very cold air in Siberia.
      I have tried to test the robustness of the predictions by choosing different start years of the data set. While the DJF prediction seems more robust, long time may not so, so I got best estimates from 0.78 to 0.88 for 2017, generally the 2017 prediction is lower, if I use less data and higher if I use all data from 1950 on since the starting of MEI time series.

    30. Yes, according to the local NCEP/NCAR index, November could be (just) the warmest since April.

    31. The WUWT prediction for GISS November is .95 ℃. If correct, that would make .84 ℃ the floor for December... assuming all other months remain the same.

      I can't imagine it. .99C on GISS means high anomalies for all 12 months. ... - me feb 2016

      Well, I can imagine it now.

    32. New predictions of my regression model using the Nov16 temp data.
      Dec16 0.75+-0.17°C
      J-D16 0.99+-0.02°C (year 2016)
      DJF17 0.77+-0.16°C
      MAM17 0.82+-0.19°C
      JJA17 0.84+-0.16°C
      SON17 1.02+-0.17°C
      J-D17 0.87+-0.14°C (year 2017)
      The value for the year 2017 depends on the number of data used. All data since 1950 is standard. For less data starting with the year of the first column, the results are:
      1950 J-D17 0.87+-0.14°C
      1960 J-D17 0.83+-0.13°C
      1970 J-D17 0.82+-0.10°C
      1980 J-D17 0.78+-0.07°C
      The formal error goes down, because of fitting less data points.

    33. GISS data are out for the whole year 2016. The value is 0.99°C, which seems very high a year ago.
      The same as my regression model predicted in Jan 16 (and in Nov 16).
      During the even warmer first part it predicted over 1.0°C. Generally the months Aug to Nov seems rather high in the prediction (for 2016 and now also for 2017)
      The model did predict only slightly negative values for MEI in the second half of 2016 but no La Nina. Maybe this could also count as not so bad.
      For the year 2017 the prediction is now 0.88+-0.12°C (using all data since 1950).

    34. Uli,
      Yes, the regression method seems good, especially in January :).
      I suggest you start a new sub-thread; it's getting hard to reply.

    35. Yes Uli, you pretty much nailed it. I think mommy nature (PDO) had your back all year. .88 ℃ for 2017? What if there is an El Niño starting in the fall of 2017... odds still low, but improving. First telling observation will be OHC for last quarter of 2016. If it is already going back up, maybe Western Pacific recharge won't take long.

    36. Yes, I will start a new sub-thread for the new year. Except if there is a whole new thread for 2017 ...

      But let me post a comparison of the old and new model here.

      The old model was really simply multilinear regression of monthly temperature and MEI data, so the prediction was lucky. Then I experimented much and get the new model approx. in June 2016.
      The model is better but for a prediction for more than a few months (or short periods) the error bars are still large.
      Let's compare the predictions of the old and new model, for 2016 with data up to 2015.
      With the data to the end of December 2015 I got with the old, over-fitted model: 1.00+-0.18 for the year 2016. The new model I use now would have given 0.99+-0.13 with slightly lower error bars.

      Could I've predicted the extraordinary record of 2016 earlier? Using the data only up to the month in the first column I get (I've used the GISS and MEI data from Dec16 now, they may have been different from the past)

      Prediction for year 2016
      Month | Old model | New Model
      Jan15 | 0.72+-0.22 | 0.80+-0.18
      Feb15 | 0.72+-0.22 | 0.80+-0.18
      Mar15 | 0.74+-0.22 | 0.83+-0.17
      Apr15 | 0.73+-0.21 | 0.83+-0.16
      May15 | 0.73+-0.22 | 0.85+-0.15
      Jun15 | 0.78+-0.21 | 0.88+-0.15
      Jul15 | 0.77+-0.21 | 0.88+-0.15
      Aug15 | 0.85+-0.20 | 0.94+-0.14
      Sep15 | 0.86+-0.20 | 0.94+-0.14
      Oct15 | 0.92+-0.20 | 0.94+-0.14
      Nov15 | 0.95+-0.20 | 0.94+-0.14
      Dec15 | 1.00+-0.18 | 0.99+-0.13

      The same for 2017
      Prediction for year 2017
      Month | Old model | New Model
      Jan16 | 0.73+-0.23 | 0.87+-0.19
      Feb16 | 0.76+-0.23 | 0.87+-0.19
      Mar16 | 0.78+-0.23 | 0.86+-0.18
      Apr16 | 0.79+-0.22 | 0.87+-0.16
      May16 | 0.79+-0.22 | 0.88+-0.15
      Jun16 | 0.76+-0.22 | 0.88+-0.15
      Jul16 | 0.77+-0.21 | 0.87+-0.15
      Aug16 | 0.84+-0.20 | 0.89+-0.14
      Sep16 | 0.81+-0.20 | 0.87+-0.14
      Oct16 | 0.82+-0.20 | 0.87+-0.14
      Nov16 | 0.82+-0.20 | 0.87+-0.14
      Dec16 | 0.83+-0.18 | 0.88+-0.12

    37. I think Nick has to start a topic: Prospects for 2017.

    38. Yes, I'll start a new topic. I'll need a few hours to write some content.

    39. Yes Nick, do that, it will be appreciated..
      You will find some material here:

      Metoffice has 2017 0.02 C below 2015, Uli suggests 2017 0.01 C above 2015..

    40. That's not a bad idea looking at Met Office's forecasts. Last year, Met Office predicted +0,95°C above 20th century average. That was the central forecast. NOAA ended with +0,94°C.

  6. I can't imagine it. .99C on GISS means high anomalies for all 12 months. If La Nina arrives late fall, there are going to be some low anomalies. ENSO is going to decay. So the Nino SST numbers are going to crash. The AMO is probably coming down as well. There is only one thing that can offset that, and that would be a strongly positive PDO. As in, unheard of numbers. That's what's called a Hail Mary.

    1. Met Office predicts +0,95°C above 20th century average (WMO standard : Giss+NOAA+Met/3). That's the central forecast mean for the whole year 2016. Upper estimation is +1,07°C. Met Office's forecasts are always near the mean anomaly predicted.

    2. JCH, 1.00 C is not unlikely. Giss temperatures rose by 0.16 C from 1997 to 1998. Watching the various el Nino forecast plumes, I'm not convinced that the present Nino will crash, but rather make a soft landing in neutral territory..
      We will know by the end of Spring. In 1998 Nino 3.4 passed zero by June 1, and continued downward quite quickly (OISST weekly).

    3. Olof - I have a similar hunch about ENSO's outcome for 2016, but ENSO seems to defy prediction. It could be far different... powerful la Nina. Anyway, as for the long term, February 2016 is looking to get warmer.

  7. My rough GISS projection for 2016 is 1.06 which is the average of the last three months of 2015. Nick's chart above shows that this simple method worked well in 97/98. The 1.06 estimate is on track so far in 2016 with temperatures increasing slowly in January and February above their late 2015 values towards a nino peak sometime in the first half of 2016.


  8. I'll take 0.9 (0.85-0.95). I think 2015 had a boost from the abortive 2014 El Nino. By contrast 1997 and 1982 were in the recovery from major volcanoes. So I see 2016 as only slightly warmer, and don't exclude it being slightly cooler. I'll be very surprised if it doesn't beat 2014 though.

    Kevin C

  9. Hi, I don't know if anyone else has been following the temperature data for the Aqua satellite. For channel 6 there is an unprecedented spike at the moment see .

    I don't believe this is spurious as both channels 7 and 8 show a similar spike.

    I gather the Aqua satellite data is not being used for UAH or RSS (correct me if am wrong) but the UAH vv6 beta 5 correlates with the Aqua channel 6 with a correlation coefficient of R2 = 0.76 see and .

    Todays value is 4.6 standard deviations above the mean. The distribution of Aqua anomalies has significant kurtosis (see' but still this is quite remarkable.

    It will be fascinating to see how this translates into this month's UAH and RSS anomalies. I also wonder if this is related somehow to the amazing GFS (greater than 7 C at the moment) anomalies for the Arctic regions.

    1. Interesting - it's a huge rise. The surface NCEP/NCAR index is also very high at the moment.

  10. It looks like February will be the Nino peak just like in 1998. month-to-date+forecast has reached the end of February and suggests that temps will be 0.13 C up from January. This is supported by the reanalysis of Weatherbell, and NCEP/NCAR, a few days behind, will likely follow..

    The satellites had double peaks in 1998, the later in April slightly higher than February. UAH v5.6 January 2015 was quite close to the old monthly record from 1998, and I believe that this record will fall in Feb 2016.
    I'm not sure about new records in RSS 3.3 and UAH v6, they are handicapped by 0.2-0.3 C compared to other troposphere and surface indices..

  11. In like a lamb; out like a ram?

  12. It looks like the spike in Aqua 6 that I mentioned above did translate into a mammoth UAH temperature of 0.83 C. Causing much consternation over at .

    By the way the spike has regressed back towards the mean but is still well into record territory. It is currently 50% greater compared to any data prior to February 2016.

  13. Hi Nick,

    have you seen that Werner Brozek has just posted some feeble article at WUWT claiming to be co-authored by you ?

    This seems to be on account of quoting a comment you made on the site a while back.

    The article if pretty feeble and I doubt you had anything to do with its writing. It seems like Brozek is just trying to steal some cred by using your name.

    I would imagine A.Watts will correct that if you object. He is pretty strong about misrepresentation.

    The joke is even using your name on that site gets a comment held back for moderation checking.

    Just thought you might want a heads up, in case you had not seen it.


    1. Thanks, Greg. Actually Werner and I correspond from time to time, and although I didn't see this article in advance, I thought I was accurately reported. FWIW, I would have waited until RSS came out before posting.

  14. I guess that substantiates the old adage that if you lie down with dogs, you will wake up with fleas.

  15. Hello Nick,

    I actually did not know when my article would get published until an hour before it did, hence the very short notice. I purposefully rushed things when I could and did not want to wait for RSS to come out first because then it would have lost all interest. It would have been like reading all about the pre polling results after the election. Who would still have been interested? As it turned out, RSS came out a few hours after that article was published and I could verify that your prediction was correct.
    Thank you very much for backing me up! I really appreciated it! I thought some people were making a big deal about nothing, however we will word things differently next time I use your excellent graphics and insights!

    Best Regards

    Werner Brozek

    1. Thanks, Werner. I think the eventual resolution was probably best. Exciting times in the troposphere!

    2. Hello Nick,
      Weekly Climate and Energy News Roundup #217 includes the following:
      Long Satellite Pauses Ending (Now Includes January Data)
      Guest Post by Werner Brozek, Excerpted From Nick Stokes, Edited by Just The Facts: WUWT, Mar 2, 2016