Detectability of forced trends in stratospheric ozone

The continued monitoring of the ozone layer and its long-term evolution leans on comparative studies of merged satellite records. Such records present unique challenges due to differences in sampling, coverage, and retrieval algorithms between observing platforms, leading to discrepancies in trend calculations. Here we examine the effects of optimal estimation retrieval algorithms on vertically resolved ozone trends, using one merged record as an example. We find errors as large as 1 % per decade and displacements in trend profile features of as much as 6 km altitude due to the vertical redistribution of information by averaging kernels. Furthermore, we show that averaging kernels tend to increase the length of record needed to determine whether vertically resolved trend estimates are distinguishable from natural variability with good statistical confidence. We conclude that trend uncertainties may be underestimated, in part because averaging kernels misrepresent decadal to multi-decadal internal variability, and in part because the removal of known modes of variability from the observed record can yield residual errors. The study provides a framework to reconcile differences between observing platforms, and highlights the need for caution when using merged satellite records to quantify trends and their uncertainties.