Extended Data Fig. 2: CMIP-based predictions of past variability in tropical lower stratospheric water vapour using two other reanalysis temperature datasets.
From: Response of stratospheric water vapour to warming constrained by satellite observations
Black: monthly mean predictions of past Δqstrat anomalies (relative to the respective seasonal cycles), using the CMIP-based functions provided with a MERRA-2 and b JRA-55 temperature data. We also show SWOOSH observational data for the same period (red), with the dots indicating the timing of the 315 months used in our calculations. The same months were selected from MERRA-2/JRA-55 for the CMIP-based qstrat predictions. The blue dashed line indicates the averaged predictions using the cross-validated ridge functions learned from the 50 combinations of SWOOSH and MERRA-2/JRA-55 data, if MERRA-2/JRA-55 is used again as the consistent input. The comparison with the SWOOSH time series (red) itself underlines that these ridge regressions represent a large fraction of the SWV variance, as evident from high r2 scores of 0.89 for a and 0.79 for b (see for example ref. 58 for a detailed explanation of this time series performance metric) and Pearson correlation coefficients of 0.96 and 0.90, respectively.
