Trend Value Calculations

[mcrossley]

I think Cumulus calculates trend values (temperature and pressure) by the difference between the value three hours ago and now, then dividing by three to obtain a rate per hour? Is this the meteo 'standard' for this calculation? It would seem that as we have lots of intermediate data that a least-squares fit may be more representative, but I guess historically the data points would be few and far between hence the simple approach.

[steve]

The standard (in the UK at least) is the change over the last three hours, yes. I can't find anything specific about how it should be calculated, but as historically this has been done manually, you would assume the method is simply to take the difference between the figure now and that three hours ago.

If you were to calculate a genuine rate of change over the same period, then I agree you might want to do something a bit more sophisticated than that.

[mcrossley]

Thanks Steve, just thinking about analysing data in a database. Least-squares is probably the simplest of the more sophisticated solutions (still a straight line fit) though the current gradient of a polynomial fit would be more representative of what is actually happening. You begin to wish you'd paid more attention to statistics at school!

[DaveNZ]

Mark, what exactly are you trying to calculate? I can't imagine a low order polynomial would fit temperature data well except on days with a steady rise or decline.

I just use the temperature difference between now and 1 hour ago as the temp trend value on my website.

[mcrossley]

Just playing around with ideas. Quite often the tend figure is the opposite of what is happening as short term variations affect the simple calc.