Dr Roy Spencer has released the final 2017 month's figures for the UAH Global Temperature satellite data. Here I'll give some analysis of the year and see how well my predictions did.
December 2017
The anomaly for December was 0.41°C, compared with the 1981 - 2010 base period. This is a larger anomaly than November, and is the 2nd warmest Decembers on record.
This means the 12 month rolling average continues to rise slightly.
The average anomaly for 2017 finished at 0.375°C. This was the third warmest on record, beaten only by 2016 and 1998, both of which were strong El Niño years. 2017 was the warmest non-el Niño year.
Forecasts
In November I was forecasting 2017 would be 0.365 ± 0.028°C, with a 98.3% chance of finishing 3rd warmest. The actual average was 0.375°C, 0.01°C warmer than forecast, but well within the 95% confidence interval.
The following graph shows how the forecast would have changed each month. The blue dashed line shows the actual anomaly for 2017. The gray band shows the 95% prediction interval for each month.
The forecast underestimated the actual result every month, but the result was within the 95% band each month, though only just for some months. The fact that 2017 finished higher than forecast reflects the surprising warming seen in later months in the satellite data.
The least accurate forecast was in January with a forecast of 0.29°C, 0.085°C below the result. The actual cumulative average would have been a better forecast in each month, which effectively means my model was continually expecting temperatures to decrease when in fact they continued to rise.
Comparison With Previous Method
Last year I used a simple method to forecast the annual temperature. This simply compared the year to date value with previous years.
This year I tried a slightly more complicated model, that included both the year to date value and the year on year trend trend. Testing had suggested this should be a better fit, but I was interested to see how the two methods compared this year. This somewhat messy graph compares the old and new methods against the result for 2017.
The new method was consistently better than the old. I wouldn't read too much into this, as a lot will depend on how close any specific year is to the trend. In this case 2017 was quite a bit warmer than the trend would suggest, but at the same time the start of the year was a lot cooler than the end. Both methods were expecting the year to stay cooler than it did, but the new method just won out.
No comments:
Post a Comment