The multi-scale structure of atmospheric energetic constraints on globally averaged precipitation

Abstract

Abstract. This study presents a multi-scale analysis of cross-correlations based on Haar fluctuations of globally averaged anomalies of precipitation (P), precipitable water vapor (PWV), surface temperature (T), and atmospheric radiative fluxes. The results revealed an emergent transition between weak correlations at sub-yearly timescales (down to ∼5 days) to strong correlations at timescales larger than about ∼1–2 years (up to ∼1 decade). At multiyear timescales, (i) Clausius–Clapeyron becomes the dominant control of PWV (ρPWV,T≈0.9), (ii) surface temperature averaged over global land and over global ocean (sea surface temperature, SST) become strongly correlated (ρTland,SST∼0.6); (iii) globally averaged precipitation variability is dominated by energetic constraints, specifically the surface downwelling longwave radiative flux (DLR) (ρP,DLR≈-0.8) displayed stronger correlations than the direct response to T fluctuations, and (iv) cloud effects are negligible for the energetic constraints in (iii), which are dominated by clear-sky DLR. At sub-yearly timescales, all correlations underlying these four results decrease abruptly towards negligible values. Such a transition has important implications for understanding and quantifying the climate sensitivity of the global hydrological cycle. The validity of the derived correlation structure is demonstrated by reconstructing global precipitation time series at 2-year resolution, relying on the emergent strong correlations (P vs. clear-sky DLR). Such a simple linear sensitivity model was able to reproduce observed P anomaly time series with similar accuracy to an (uncoupled) atmospheric model (ERA-20CM) and two climate reanalysis (ERA-20C and 20CR). The linear sensitivity breaks down at sub-yearly timescales, whereby the underlying correlations become negligible. Finally, the relevance of the multi-scale framework and its potential for stochastic downscaling applications are demonstrated by deriving accurate monthly P probability density functions (PDFs) from the reconstructed 2-year P time series based on scale-invariant arguments alone. The derived monthly PDFs outperform the statistics simulated by ERA-20C, 20CR, and ERA-20CM in reproducing observations.