Terrestrial Water Cycle Acceleration-Deceleration: Non-Binary and Space-Time Divergent

Changes in the terrestrial water cycle are often discussed as either an acceleration or a deceleration of the cycle. However, different combinations of precipitation, runoff, and evapotranspiration changes are possible, and it is largely unknown which combinations actually occur around the world. We quantify water flux changes and their combinations from 1980–2000 to 2001–2020 based on: (a) observational data for 3,614 hydrological catchments with worldwide distribution; (b) a new ensemble of machine learning (ML) models, trained and tested on data for these catchments and applied globally; and, comparatively, (c) four alternative data sets for water flux changes from 1981–1995 to 1996–2010 in 1,561 catchments worldwide. The changes in precipitation, runoff, and evapotranspiration are mostly in opposite directions, with 51 ± 7% of the catchments or land area (based on (a–b); 56 ± 4% based on (c)) experiencing acceleration or deceleration in two fluxes and the opposite in the third. Unidirectional changes in all water fluxes are observed only in 27.5 ± 2.5% and 21.5 ± 4.5% of the catchments or land area (based on (a–b); 23.5 ± 6.5% and 19.5 ± 4.5% based on (c)) for full deceleration and full acceleration, respectively. Different terrestrial water fluxes thus concurrently decelerate and accelerate at both local and global scales. Interpretation of the ML modeling further shows different driver-impact relationships for the water flux changes over time than across space. This space-time difference challenges the usefulness of space-for-time substitution approaches for temporal flux changes. The ML model ensemble developed in this study offers a promising approach for addressing this challenge.