How runoff will change as atmospheric CO2 rises depends upon several difficult to project factors, including CO2 fertilization, lengthened growing seasons, and vegetation greening. However, geologic records of the hydrological response to past carbon cycle perturbations indicate large increases in runoff with higher CO2. We demonstrate that the fact that the Earth has remained habitable since life emerged sets a lower-bound on the sensitivity of runoff to CO2 changes. The recovery of the Earth system from perturbations is attributed to silicate weathering, which transfers CO2 to the oceans as alkalinity via runoff. Though many factors mediate weathering rates, runoff determines the total flux of silicate-derived cations and hence the removal flux of excess CO2. Using a carbon cycle model that parameterizes weathering as a function of rock reactivity, runoff, temperature, and soil CO2, we show that recovery from a perturbation is only possible if the lower-bound for the sensitivity of runoff to atmospheric CO2 is 0%/K. Using proxy data for the Paleocene-Eocene Thermal Maximum, we find that to match the marine d13C record requires a runoff sensitivity greater than 0%/K and similar to estimates of the modern runoff sensitivity derived from an ensemble of Earth system models. These results suggest that the processes that enhance global runoff are likely to prevail over processes that tend to dampen runoff. In turn, that the Earth has always recovered from perturbations suggests that, though the runoff response is spatially complex, global discharge has never declined in response to warming, despite quite varied paleogeographies.

Satellite data reveal widespread changes in Earth’s vegetation cover. Regions intensively attended to by humans are mostly greening due to land management. Natural vegetation, on the other hand, is exhibiting patterns of both greening and browning in all continents. Factors linked to anthropogenic carbon emissions, such as CO2 fertilization, climate change, and consequent disturbances such as fires and droughts, are hypothesized to be key drivers of changes in natural vegetation. A rigorous regional attribution at the biome level that can be scaled to a global picture of what is behind the observed changes is currently lacking. Here we analyze different datasets of decades-long satellite observations of global leaf area index (LAI, 1981–2017) as well as other proxies for vegetation changes and identify several clusters of significant long-term changes. Using process-based model simulations (Earth system and land surface models), we disentangle the effects of anthropogenic carbon emissions on LAI in a probabilistic setting applying causal counterfactual theory. The analysis prominently indicates the effects of climate change on many biomes – warming in northern ecosystems (greening) and rainfall anomalies in tropical biomes (browning). The probabilistic attribution method clearly identifies the CO2 fertilization effect as the dominant driver in only two biomes, the temperate forests and cool grasslands, challenging the view of a dominant global-scale effect. Altogether, our analysis reveals a slowing down of greening and strengthening of browning trends, particularly in the last 2 decades. Most models substantially underestimate the emerging vegetation browning, especially in the tropical rainforests. Leaf area loss in these productive ecosystems could be an early indicator of a slowdown in the terrestrial carbon sink. Models need to account for this effect to realize plausible climate projections of the 21st century.