Moonage Daydream

Ken Ring believes that the moon's influence on our weather is so strong that he can forecast the weather for New Zealand, Australia and the United Kingdom for years in advance, based on calculating how the moon moves in its orbit around the earth. At first glance it seems to be a bit like working out the tables that predict how tides will behave. Unfortunately for Ken, the moon's influence on the atmosphere is much, much smaller than on the sea (it's a gravity/mass thing, very well explained by Bill Keir of the Auckland Astronomical Society here). His method, as he describes it, can't work, so his forecasts must be rubbish. But are they? His books sell well enough for Random House to keep coming back for more. The popularity of Ring's almanac suggests that at least some of his readers think that his forecasts have value. Of course, astrology books also sell by the truckload…

However, what Ken does to arrive at his forecasts is irrelevant when all we want to do is to see if they work. I'm treating Ken's "lunar method" as a black box. He plugs in whatever he plugs in, and out pop the very detailed forecasts he publishes every year. I am simply going to look at what he predicted for 2006, and see how things turned out in real life.

"1. Firstly there is no claim to get every forecast right 100% of the time, but about 80-85% seems reasonable, the same as the metservices claim."

That's the accuracy Ken expects to achieve, as stated on his Appraisals and Surveys page [here]. He then lays down some conditions that - he claims in the interests of fairness - assessments of his forecasts should meet. He wants to be allowed a 24 hour error, which I take to mean a day either side of his forecast weather event. Four times a month there is a "potential skewing" of 2 to 3 days, and the summer cyclone season is apparently difficult too. Rainfall is hard, and he believes that if a rain event misses his forecast location by up to 60 miles, he should be given a "hit". With the pressure maps he prints in the almanac, he wants readers to look for "3-4 day trends" rather than an exact fit to reality. And finally, he expects that his forecasts should only be compared to other forecasts for the same day made at the same time - in other words, years in advance.

I thought about Ken's caveats very carefully. Could I do any meaningful analysis of his weather predictions and take into account all his caveats? Let's consider a forecast for rain in Christchurch on a Friday. He expects to be allowed 24 hours error, so he could claim rain on Thursday or Saturday as a success. But there are the "potential skewing" events to consider. They happen four times a month - roughly once a week - so perhaps he could claim that any rain in Christchurch on Tuesday, Wednesday, Thursday, Friday, Saturday, Sunday or Monday was confirmation of a successful forecast for rain on Friday. That's rather a lot of leeway.

I quickly realised that attempting to look at the detail of Ken's forecasts would be a thankless task. For a start, it would be a huge amount of work to take the pressure charts he prints and compare them to what actually happened. Does a Ring forecast of high pressure over the north Tasman count as successful if there's a high to the east of NZ, or if the high pops up in the right place but 3 days early or late? Too much work (I do have a life, actually), and too much wiggle room for Ring. The same applies to his daily forecasts in words - he provides two for each day. One in the "monthy summary" at the start of each month's section in the almanac, and one next to the map. Neither provide enough information for a ready measurement against reality.

There are two forecasts, however, that can be related to actual events. Ken provides summary tables of rainfall and sunshine hours for each month, and provides "estimates" of figures for 32 North Island and 25 South Island locations. Thus he predicts that Auckland will receive 79mm of rain in September, and 101 hours of sunshine. Here are figures that can be compared with actuals, and - even better - because we are dealing with whole months, the precise timing of weather events is much less important to the "skill" of the forecast. I therefore set about assembling the data for the comparison…