Revised Sunspot Data

Credits

Kindly note that all the sunspot data used in my own calculations were downloaded from (SILSO data/image, Royal Observatory of Belgium, Brussels). 

New way to check and report sunspot data implemented

The Royal Observatory of Belgium changed over to a new way to capture and report sunspot number on 1 July 2015. The Press release is linked for your convenience.

On the website of the  Royal Observatory of Belgium there are links to the reports and papers published by the work groups that  that led to this change. Take some time to read the reasons for the change. There are numerous reasons for the changes, but in the end it all boils down to one element, the humans and their errors that captured and will continue to capture the sunspot data.

Reasons for change

It is not my intention to list all the reasons  - strong and valid as they are - for the changes, but the following important ones caught my eye:

1. Change in instrumentation. Well it is pretty obvious that if you use a stronger telescope - with its appropriate adjustments, you will not just point your telescope to look at the sun will you? -- you will see more detail and more sunspots.
2. Change in eyesight of observer. This I found very sad and painful to read.  Image the poor dedicated man who recorded the sunspot numbers for more than 30 years with an ever failing sight, counting less and less sunspots as time goes by. I have great admiration for him and can only applaud the work group, led at times by Lief Svalgaard, for the sympathetic way they described this issue. 

My findings

I downloaded the revised sunspot series at the beginning of August 2015 and plotted the new sunspot series against the data from the old sunspot series using the LOESS - as described previously - function in Excel. The same Y-axis maximum of 350 sunspots is used in all the graphs I prepared to make it easier to compare.

Conclusion

The big difference between the earlier sunspot data and the most recent sunspot cycles have disappeared. We cannot talk of the 20th Century super cycles anymore. 

Types of Sunspot Cycles

Types of Sunspot Cycles?

Introduction

I have yet to see sunspot cycles classified. From the work I did with the LOESS manipulation of sunspot data it is evident that at least two type of sunspot cycles exist. There are sunspot cycles with a single peak and there are sunspot cycles with double peaks, almost like the humps of camels. The D-type sunspots, with a single hump, are the most common kind, and there are, to date, also a few examples of a B-type, double peaked, or hump sunspot cycles.   So I will use the zoological terms to classify camels, Dromedary - one peaked sunspot  cycle and Bactrian - double peaked sunspot cycle.

Dromedary Sunspot Cycle

Figure 1 - Typical Dromedary type sunspot cycle

Figure 1 which shows a plot of the curve fitted to sunspot cycle 21 is an example of the most common type of Dromedary curve fitted to sunspot observation data.

Bactrian Sunspot Cycle

Figure 2- Typical Bactrian type sunspot cycle

Figure 2 which shows a plot of the curve fitted to sunspot cycle 22 is an example of the rather more uncommon Bactrian curve fitted to sunspot observation data.

Discussion

It appears as if our current sunspot cycle 24 is also going to be a Bactrian type.

Revised dates for sunspot cycle peaks

Revised sunspot cycle peak dates

Introduction

Using the LOESS data to plot sunspot cycle data is the topic discussed in my previous blogs. I show that the current practice of using a moving average on sunspot data is probably not ideal and that using the LOESS procedure yields better results. 

A spin-off from using the LOESS procedure to calculate the shape of a sunspot cycle is that the peak of the sunspot cycle is calculated to be earlier than that reported using the moving average approach. I will list the revised sunspot cycle peaks that I obtained using the LOESS procedure.

Revised Sunspot Cycle Peak dates using the LOESS procedure

I did all the calculation in Excel and used the inbuilt maximum function to locate the peak of the sunspot cycles that I had data for. The results were summarized and are shown in the following tabulation:-

Table 1 - Sunspot Cycle Peak Dates

In Table 1 the revised sunspot cycle peak dates using the actual raw sunspot data (Sun Spot Count), the revised method of calculating the sunspot cycle shape (LOESS 1000) and moving average method (Ma 390) are shown. The results is a lot of dates that need to be carefully inspected to make any sense. To make this information easier to understand I created the following tabulation.

Table 2 - Deltas in Days

Note the average is calculated using the full sunspot cycles 7 to 23

In Table 2 I show the differences (deltas) in days between the various ways I used to calculate the peak date for the sunspot cycle. I used the LOESS 1000 sunspot peak date as the reference. A negative number is the number of days the other way to calculate the sunspot cycle is later than the date of the LOESS 1000 date, and using the same logic, a positive  number is the number of days the other method gives an earlier date.

Discussion

On average the moving average method of establishing the sunspot cycle peak calls the peak 184 days later than that of the LOESS method. This is approximately a  six months difference.

On average the moving average method calls the peak of the sunspot cycle 6 months too late.

How will this impact on studies that rely on sunspot data to make projections? I am not sure but will start looking whether I can find any recent papers and see what is the influence.

Lastly, sunspot cycle 24 is still ongoing and incomplete and ignored in the average calculation, as is Sunspot cycle 6.

More fun with Sunspot data and Excel

Excel has a neat trend-line function that can be used when plotting data. I did some exercises using the trend-line function in Excel and to compare the output with the results obtained from the LOESS function. The data set that I used is in this example is that of the sunspot cycle 23, downloaded from the Royal Observatory of Belgium.

Lets look at the raw sunspot data for sunspot cycle 23 plotted in Excel:-

The raw data points

Using the polynomial trend-line we get the following:-  

Trend-line fitted using polynomial to the 6th power  

Now I superimpose the curve obtained by plotting the LOESS 1000 days function on the graph:-

The LOESS 1000 curve, Sunspot 23 raw data and trend-line
 fitted to the sunspot raw data.

What is my conclusions?

1. Using the inbuilt Excel trend-line facility with the polynomial set to the 6th power yields a curve that is remarkably similar to the curve obtained from the LOESS function set to 1000 data points (days).
2. If the Sunspot 23 data is subjected to the LOESS function set to 1000 days  a curve that is very close to the inbuilt curve fitting functionality of Excel is obtained.
3. I cannot get similar results using the moving average data.
4. As an Engineer I would put more trust in the output from the LOESS 1000 function to describe the shape and outline of  Sunspot 23 than the results obtained from the moving average method of calculation.

Why? It just fits neater, and that is always a good indication that I am on the right track.

Sunspot Cycle 24 update for year-end 2013

Note the source of the sunspot data is:-

WDC-SILSO, Royal Observatory of Belgium, Brussels.

The Royal Observatory of Belgium have updated their website and it is now possible to download the sunspot data in a csv file format.That saves a lot of time for people who want to use the data in  spreadsheets, thank you to the good folk at that institution.

Progress of the sunspot cycle 24.

Note that by year-end the sunspot cycle did not come to a close as the sun is still merrily firing away on all cylinders. So, at this stage we are still experiencing sunspot cycle 24. It appears as if this cycle is going to a bit longer than the forecasts made by some experts, and given the complexity of the issue, that should come as no surprise.

Sunspot Cycles as a set

Up to now I showed plots of the recorded sunspot cycles individually. I decided to do some plots of the sunspot cycles on a continuum. The limitations in Excel stops me from plotting all the series in one go, that is using the daily spot counts as data.

Raw data with observed counts

17 April 2013

Local Regression v the Gaussian Mask

Introduction

On my previous post local-regression-loess-and-sun-spot I went to great lengths to explain how local regression works in practice. To my great delight I discovered that the UK Met office is using their own version of local regression. 

But, instead of calling it local regression they call it a Gaussian mask.

See if you can spot a difference or a similarity?

Met office gaussian mask

So, instead of referring to this process as a local regression, and acknowledging the work done by the Americans, a new term to describe the technique was created