Oscillations and Time Trends in Stratospheric Ozone Levels

Abstract

A functional data analysis approach is presented to study altitude-dependent patterns of ozone variation in data from a sequence of ozonesonde flights. Ozonesondes are balloon-based instruments that measure ozone as the balloon ascends through the troposphere and lower stratosphere. This article concentrates on variation in the altitude range of 15.5–30.5 km, in January–July in 1967–1998. Ozonesonde flights originating at a mid-latitude site, Hohenpeissenberg in Germany, are studied. Estimates are obtained of altitude-dependent nonlinear time trends in ozone partial pressures, together with ozone variation associated with the quasi-biennial oscillation (QBO), an atmospheric process thought to influence global ozone transport. Both methodological and scientific contributions are made. The data analysis approach combines dimension-reduction basis function approximations with low-dimensional spline-based models on the basis function coefficients. This provides an efficient and flexible approach for studying complex time/altitude variation in the ozone partial pressure profiles, including nonlinear time trends. In contrast, the standard approach uses multiple linear regression models to estimate linear or piecewise-linear ozone time trends separately for each altitude level. Scientific results include identifying clear bimodalities (in altitude) in the estimated QBO components of variation in certain months. These may relate to planetary wave transport of ozone from the tropics to mid-latitudes. Additional empirical evidence of an 11-year cycle in ozone levels also is provided, possibly linked with a solar cycle. However, ozone peaks at Hohenpeissenberg do not always coincide with the timing of the maxima of the 11-year solar cycle. This may indicate a transport-related lag in ozone maxima at some altitudes. The estimated QBO features are robust to the presence or absence of the data quality “correction factors” commonly used in ozonesonde studies. However, the nonlinear time trend components show greater sensitivity to these.