On 26/07/2018 21:55, Alastair wrote:
On Thursday, 26 July 2018 18:45:05 UTC+1, N_Cook wrote:
On 26/07/2018 16:55, JGD wrote:
I'm afraid that I was always taught that curve fitting is good (with
caveats) for interpolation but to be avoided as far as humanly possible
for extrapolation.

(Because there is always a serious danger that a model that appears to
be a good fit over a limited range of data can become - potentially -
absurdly wrong the further the curve is pushed beyond the available
data. This is especially so if the model equation includes some sort of
power function and/or is not grounded in some credible physical
hypothesis.)

This obviously presents a real difficulty for forecasts of climate
change and related parameters where - short of becoming time travellers
- there is no choice but to try to extrapolate into the future. But it
needs to be done with real caution if the parameter values are to be at
all useful or credible, whatever the nominal SSR might suggest.


Fair enough and here
https://www.aviso.altimetry.fr/en/da...sea-level.html
they say
"for scientific and statistics reasons, period under 5 years are not
significant."
so what is the point of graphing out this info.
Last year it seemed to me, that concattenating Jason1+2+3 SLR plots,
whatever character SLR had, it was not linear, and so my concattenation,
allowing for filter-effect discontinuites at 1/2 and 2/3 junctions
http://diverse.4mg.com/jason1+2+3r.jpg
I've not even found a proper academic concattenation of J1+2+3 outputs.
They and other academics are still putting a linear "fit " to the curve.
The situation at the end of last year looked a lot worse, due to filter
effects etc, but the curve is still not linear best fit for northern
nemispherw spring+summer months added data.
Just my effort to get a more rational handle on the SLR later this
century. The curve-fit projection will increase again , due to the
filters ,again this autumn/winter, but hopefully it will all
give a better idea than linear, after a few years of doing this

Nick,

I think you are doing sterling job. As you say, the conventional approach is to use a linear fit which has all the drawbacks explained by JGD. By presenting several projections you are describing various alternatives, only one at most of which can be correct. Keep up the good work!

Cheers, Alastair.


I've not updated that plot I referred to above, probably even less
difference in recent months. What I did not expect was only 2 pixels of
vertical difference between those 3 curved plots, almost imperceptible,
but of course project on through the century and very different outcomes.
It looks as though the next but one Jason3 public update may return to
the more elevated cycle again. It does not look as though the R*R
ranking order is going to change before that return to more positive
territory and 4/3 indicial power law will continue to be the best fit.