Two Curve Fitting Methods Applied to Co2 Flask Data

Digital filtering and harmonic regression curve fitting techniques are applied to CO2 flask data from four stations in North America (Pt. Barrow, Alert, Sable Island and Cape St. James) to evaluate these two different methodologies in terms of growth rate and seasonal cycle in the atmospheric CO2 concentration. Both methods agree relatively well in producing long-term atmospheric CO2 trend at each of the monitoring stations, as well as in capturing relatively large interannual variations in the annual growth rate. Furthermore, they both agree in indicating the dependency of the variation in the seasonal amplitude on the seasonal minimum concentration. The digital filtering technique is able to capture the local temporal variation in CO2 measurements much better than the harmonic regression method, although in some cases this variability is exaggerated in the digital filtering approach. The harmonic regression approach tends to smooth out the data, with much of the power in the very long period oscillations. Also, the timing of the occurrence of the seasonal minimum calculated by the digital filtering method tends to be earlier than that calculated by the harmonic regression method, although both methods do not indicate any major secular change in the timing. The overall assessment of the two methods applied to the CO2 flask data underscores the importance of using more than one curve fitting method before any conclusions can be drawn from the flask data about the interannual variability in the trend and seasonal cycle of the atmospheric CO2 concentration. © 1997 John Wiley & Sons, Ltd.