A Demonstration of the Scientific Value of GRUAN Data: the use of GRUAN Uncertainty Estimates in Trend Analyses

One of primary goals of the GCOS (Global Climate Observing System) Reference Upper Air Network (GRUAN) is to provide vertical profiles of reference measurements suitable for reliably detecting changes in global and regional climate on decadal time scales. A key attribute of reference quality measurements, and hence GRUAN data, is that each datum has a well characterized and traceable estimate of the measurement uncertainty. Considerable effort is invested in GRUAN operations to describe/analyze all source of measurement uncertainty to the extent possible, quantify/synthesize the contribution of each source of uncertainty to the total measurement uncertainty, and verify that the evaluated net uncertainty is in agreement with the required target uncertainty. However, if the climate science community is not sufficiently well informed on how to capitalize on this added value, the significant investment in estimating meaningful measurement uncertainties is largely wasted. The purpose of this presentation is to discuss and demonstrate how the GRUAN measurement uncertainties should be used in determining long-term trends in upper air climate data records. The presentation will include discussion of the construction of linear least squares regression models for trend analysis, boot-strapping approaches to determining uncertainties in trends, dealing with the combined effects of auto-correlation in the data and measurement uncertainties in calculating the uncertainty on trends, determining whether trends (with their uncertainties) derived from two different data sets are statistically significantly different, best practice for determining seasonality in trends, interpretation of derived trends and in particular how to interpret derived trends when the basis functions in the regression model are not orthogonal. The presentation will include analyses of existing GRUAN data but will also include demonstrations of these various concepts using synthetic data. The presentation will include a discussion of how to characterize the additional uncertainty in trends when the climate data record is known to have a discontinuity. Finally the presentation will include a discussion of when and how uncertainties on the measurements become the determining factor in estimates of the uncertainty in derived trends.