Temperatures July Update - Now With BEST

All the data sets have released for July so it's time for another look at how the statistical predictions are coming along. This month I've added BEST to the data sets, and I've been trying to improve the appearance and colors of my graphs, but it's still a work in progress.

Here's what 2016 looks like so far, month by month. (All temperatures have been rebased to the 1981 - 2010 period.)

July saw a pause in the rapid drop of the last few months, with most sets either rising slightly or dropping by only a small amount.

Here's how the projection of the expected annual temperature has changed for each set - as always this is expressed as the difference to the previous record.

Each data set is continuing to slowly drop. It's interesting to see how much of a split there is between the various sets. The two hot satellite sets (the old non-beta UAH, and the new RSS) continue to be the sets projected to beat the record by the largest amount, whilst the two cold satellite sets are projected to beat the record by the least amount. All of the terrestrial sets are between these two. Again it's important to remember these are showing the difference between the expected 2016 value and the previous record, which is different for each dataset. In particular the satellites are compared to the exceptional 1998 value, whilst the surface data is compared with 2015.

The changing probabilities of each set beating their respective record years are shown here:

UAH beta 6 is down to 78% and RSS 3.3 is at 90%. Both down 2 percentage points. Others remain close to certain, with HadCRUT now at 97% and NOAA at 99%.

As always, I don't agree with these probabilities. In particular I think it's very unlikely that UAH beta 6 will beat the record. An interesting point regarding UAH beta 6 is that the average for the first 7 month of 2016 is almost exactly the same as the average of the first 7 months of 1998. 0.584 °C for 2016, compared with 0.580 °C for 1998. Thus the next 5 months will have to be cooler than the last 5 months of 1998 for UAH beta 6 to not beat the record.

Here's the comparison of 2016 to 1998 along with a dashed line representing the average temperature required over the rest of the year for 2016 to equal 1998.

UAH can only drop slightly from July before it fails to beat the record. This will depend on

Comparisons with Other Record Years

The assumption is that 2016 will most resemble 1998 as far as changes over the year are concerned. Both were strong El Niño years, and show a very strong spike in temperatures especially in the satellite records.

Here's what all the different data sets did during the last great El Niño year of 1998. (All anomalies based on the 1981 - 2010 period.)

To make this clearer here's 1998 showing the average of the 4 surface sets, and the 4 satellite sets.

One obvious point is that the satellite data was much higher throughout the year than the surface data. This illustrates just how hot 1998 was in the satellite records.

What's also interesting is the way the surface data drops so quickly in September, whilst the satellite data shows an even bigger drop in November.

So how well would this forecasting method have worked in 1998? Using only data up to 1998, this is how the margin graph would have looked for 1998.

I'm surprised how good this looks. Note that the figure for December is the final result, not a projection. Here's how the probabilities looked.

Less surprising given how warm the start of 1998 was by July all data sets would have been near certain of beating the record.

Lets do the same with 2010. This was another El Niño year, though not a strong one. It broke the record in all the surface data, and produced another massive spike in satellite data, but not enough to beat 1998.

I'd say this would have been a good set of forecasts taken from July say, the two sets that were given little chance failed to beat the record, the one set (NOAA) with a strong chance did beat the record, and of the other five being given between around 40 - 75% chance, 3 did beat the record and 2 didn't.

Finally, if you haven't had a enough, here's last year, which was a record in all the surface data, and not a record in all the satellite data.

For all the data sets the forecast underestimated the annual temperature throughout the year, with the July forecasts being around 0.1 °C below the actual value. But the probability forecasts were not bad, with all the satellite data showing no chance by July, and all the surface sets having greater than 70% chance of a record.