Explaining Global Increases in Water Use Efficiency: Why Have We Overestimated Responses to Rising Atmospheric CO2 in Natural Forest Ecosystems?

The analysis of tree-ring carbon isotope composition (δ13C) has been widely used to estimate spatio-temporal variations in intrinsic water use efficiency (iWUE) of tree species. Numerous studies have reported widespread increases in iWUE coinciding with rising atmospheric CO2 over the past century. While this could represent a coherent global response, the fact that increases of similar magnitude were observed across biomes with no apparent effect on tree growth raises the question of whether iWUE calculations reflect actual physiological responses to elevated CO2 levels. Here we use Monte Carlo simulations to test if an artifact of calculation could explain observed increases in iWUE. We show that highly significant positive relationships between iWUE and CO2 occur even when simulated data (randomized δ13C values spanning the observed range) are used in place of actual tree-ring δ13C measurements. From simulated data sets we calculated non-physiological changes in iWUE from 1900 to present and across a 4000 m altitudinal range. This generated results strikingly similar to those reported in recent studies encompassing 22 species from tropical, subtropical, temperate, boreal and mediterranean ecosystems. Only 6 of 49 surveyed case studies showed increases in iWUE significantly higher than predicted from random values. Our results reveal that increases in iWUE estimated from tree-ring δ13C occur independently of changes in 13C discrimination that characterize physiological responses to elevated CO2. Due to a correlation with CO2 concentration, which is used as an independent factor in the iWUE calculation, any tree-ring δ13C data set would inevitably generate increasing iWUE over time. Therefore, although consistent, previously reported trends in iWUE do not necessarily reflect a coherent global response to rising atmospheric CO2. We discuss the significance of these findings and suggest ways to distinguish real from artificial responses in future studies.