Fitting a quadratic model to an incomplete number of cycles of a periodic function can give the false impression of an “acceleration” or deceleration.

Here, synthetic data was made using a folded cosine and an upward linear trend, similar in size and form to the global sea level record reported by Jevrejeva et al 2014.

A quadratic model was fitted to the data.

The result slightly reduces the real linear trend and fits a spurious upward acceleration.

This is because of the incomplete cyclic part towards the end producing an upwards bias.

Final set of parameters Asymptotic Standard Error

======================= ==========================

a = 0.00178066 +/- 0.0002117 (11.89%)

b = 1.73492 +/- 0.03279 (1.89%)

c = -120.217 +/- 1.065 (0.8858%)

A similar demonstration of fitting a linear model to incomplete cycles:

https://climategrog.wordpress.com/?attachment_id=209

Code to reproduce the above graph using gnuplot 4.x

### simulate jevrejeva tide gauge quad fit since 1860 cos1(x)=a1*abs(cos(2*pi*(x-yz)/p1)) yz=1945; p1=120; a1=50; lin(x)=m_lin*(x-1940) m_lin=2.25 set key top left Left reverse plot [1860:2010] lin(x)+cos1(x) set table "tide_sim.dat" ; replot; unset table # write sim data to file quad(x)=(a*(x-1860)+b)*(x-1860)+c fit [1860:2010] quad(x) "tide_sim.dat" using 1:2 via a,b,c plot [1860:2010] lin(x)+cos1(x) title "50mm abs_cos + 2.25 mm / yr linear ", quad(x) title "quadratic fit" set title "Fitting Quadratic Acceleration to Cosines" set label 1 sprintf("fitted quadratic coeff x 2 = %.4f mm.yr^-^2",a*2) at graph 0.05, 0.8 set label 2 sprintf("fitted quadratic coeff = %.2f mm.yr^-^1",b) at graph 0.05, 0.75 replot # Final set of parameters Asymptotic # Standard Error ======================= ========================== # a = 0.00178066 +/- 0.0002117 (11.89%) # b = 1.73492 +/- 0.03279 (1.89%) # c = -120.217 +/- 1.065 (0.8858%) #