Inappropriateness of space-for-time and variability-for-time approaches to infer future dryland productivity changes

Drylands are among the most susceptible ecosystems to global climate change. It has been suggested that a future surface drying trend would largely reduce gross primary productivity (GPP) in drylands, given that water is the dominant factor controlling the spatial distributions (i.e., space-for-time analogy) and inter-annual fluctuations (i.e., variability-for-time analogy) of dryland GPP. However, whether these approaches using spatial and inter-annual diagnostics are valid to infer longterm dryland GPP remains unknown. In this study, we evaluate whether space-fortime and variability-for-time approaches, which are based on the empirical scaling between GPP and dryness, are able to capture future changes in dryland GPP as simulated by 18 Earth system models (ESMs). Using observational data during 1958-2014, we identify a strong coupling between dryland GPP and the annual aridity index (Al, the ratio of precipitation to potential evapotranspiration) over both spatial and inter-annual scales. This GPP-Al scaling is used to predict future GPP changes throughout the 21st Century based on the future Al changes projected by ESMs. The space-for-time, and variability-for-time approaches predict an overall decrease of dryland GPP by -23.66 +/- 10.93 (mean +/- 1 standard deviation) and -3.86 +/- 2.22 gC m(-2) yr(-1), respectively, in response to future surface drying, however, the ESM projections exhibit a strong dryland GPP increase (+81.42 +/- 36.82 gC m(-2) yr(-1)). This inconsistency is because the space- and variability-based approaches, which rely on the spatial or short-term GPP-Al relationships, cannot capture the slowly-evolving but key determinant of dryland GPP changes over multi-decadal or longer timescales, which, in ESMs, is the ecosystem physiological response to rising CO2. Our study questions the validity of "the drier the less productive" hypothesis rooted in the space-for-time and variability-for-time theories, and the implementation of such theories to constrain future ecosystem changes.