Reconstruction of Zonal Precipitation From Sparse Historical Observations Using Climate Model Information and Statistical Learning. Issue 23 (8th December 2022)

• Record Type: Journal Article
• Authors: Egli, Marius; Sippel, Sebastian; Pendergrass, Angeline G.; de Vries, Iris; Knutti, Reto
• Is Part Of: Geophysical research letters. Volume 49:Issue 23(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-12-08
• Subjects: precipitation -- reconstruction -- statistical learning -- infilling -- large ensemble -- hydrological cycle; Geophysics -- Periodicals; Planets -- Periodicals; Lunar geology -- Periodicals; 550
• DOI: 10.1029/2022GL099826
• Languages: English
• ISSNs: 0094-8276

Abstract

Abstract: Future projected changes in precipitation substantially impact societies worldwide. However, large uncertainties remain due to sparse historical observational coverage, large internal climate variability, and climate model disagreement. Here, we present a novel reconstruction of seasonally averaged zonal precipitation metrics from sparse rain‐gauge data using regularized regression techniques that are trained across climate model simulations. Subsequently, we test the reconstruction on independent satellite data and reanalyzed precipitation, and find a large fraction of historical zonal mean precipitation (ZMP) variability is recovered, in particular over the Northern hemisphere and in parts of the tropics. Finally, we demonstrate that the reconstructed ZMP trends are outside the variability of pre‐industrial control simulations, and are largely consistent with the range of historical simulations driven by external forcing. Overall, we illustrate a novel way of estimating seasonally averaged zonal precipitation from gauge data, and trends therein that show a signal very likely caused by human influence. Plain Language Summary: When studying changes in the global water cycle due to climate change it is instructive to study precipitation along constant latitudes (zonal mean), as the average amount and seasonality of precipitation differ strongly across latitudes. When trying to calculate the zonal mean from observations, we face the problem that observations do not exist for many locations at the latitude in question since there may be no precipitation gauges, and the number and locations of gauge stations changes over time. Here we present a method to reconstruct the zonal mean precipitation (ZMP) from spatially incomplete observations, by training a statistical model to predict the zonal mean from only the observed grid cells directly. Our reconstructions show high similarity to satellite‐based estimates of ZMP. Further, we find a trend in these reconstructions when analyzing the pattern of all zonal trends together, which is very likely caused by human influence. Key Points: Detection and attribution of change in the water cycle is difficult due to sparse observations, model uncertainty, and internal variability We reconstruct the inter‐annual variability of zonal mean precipitation from gauge data using regularized regression techniques We demonstrate that the observed multi‐decadal zonal water cycle changes lie within the range of historical climate model simulations