Summary Semi‐arid ecosystems play an important role in regulating the dynamics of the global terrestrial CO2 sink. These dynamics are mainly driven by increasing inter‐annual precipitation variability. However, how ecosystem carbon processes respond to changes in precipitation is not well understood, due to a lack of substantial experimental evidence that combines increased and decreased precipitation treatments. This study, a 3‐year field manipulation experiment with five precipitation levels conducted in a semi‐arid steppe, examined the impacts of increased and decreased precipitation on ecosystem CO2 (GEP: gross ecosystem photosynthesis; ER: ecosystem respiration; NEE: net ecosystem CO2 exchange), water exchange (ET: evapotranspiration), and resource use efficiency (CUE: carbon use efficiency; WUE: water use efficiency). We found that decreased precipitation reduced ecosystem CO2, water exchange and resource use efficiency significantly, while increased precipitation did not cause significant influence on them. That is, they responded more sensitively to decreased precipitation. Soil water availability was the most important driver determining changes in GEP, ER and ET. Changes in NEE, CUE and WUE were predominately regulated by soil temperature. Photosynthesis at leaf and ecosystem levels showed significantly greater sensitivity to changed precipitation than respiration and ET, and therefore determined the trends of net carbon uptake and resource use efficiency. This study highlighted an asymmetric response of ecosystem carbon and water processes to altered precipitation. This is potentially important for improving our understanding of how possible future changes in precipitation will affect the carbon cycle. Taking this asymmetric response into consideration will inevitably reduce uncertainties in predicting the dynamics of the global carbon cycle. A http://onlinelibrary.wiley.com/doi/10.1111/1365-2435.12836/suppinfo is available for this article.

Elucidating the variation of allocation pattern of ecosystem net primary productivity (NPP) and its underlying mechanisms is critically important for understanding the changes of aboveground and belowground ecosystem functions. Under optimal partitioning theory, plants should allocate more NPP to the organ that acquires the most limiting resource, and this expectation has been widely used to explain and predict NPP allocation under changing precipitation. However, confirmatory evidence for this theory has mostly come from observed spatial variation in the relationship between precipitation and NPP allocation across ecosystems, rather than directly from the influences of changing precipitation on NPP allocation within systems. We performed a 6‐yr five‐level precipitation manipulation experiment in a semiarid steppe to test whether changes in NPP allocation can be explained by the optimal partitioning theory, and how water requirement of plant community is maintained if NPP allocation is unaltered. The 30 precipitation levels (5 levels × 6 yr) were divided into dry, nominal, and wet precipitation ranges, relative to historical precipitation variation over the past six decades. We found that NPP in both aboveground (ANPP) and belowground (BNPP) increased nonlinearly as precipitation increased, while the allocation of NPP to BNPP (fBNPP) showed a concave quadratic relationship with precipitation. The declined fBNPP as precipitation increased in the dry range supported the optimal partitioning theory. However, in the nominal range, NPP allocation was not influenced by the changed precipitation; instead, BNPP was distributed more in the surface soil horizon (0–10 cm) as precipitation increased, and conversely more in the deeper soil layers (10–30 cm) as precipitation decreased. This response in root foraging appears to be a strategy to satisfy plant water requirements and partially explains the stable NPP allocation patterns. Overall, our results suggest that plants can adjust their vertical BNPP distribution in response to drought stress, and that only under extreme drought does the optimal partitioning theory strictly apply, highlighting the context dependency of the adaption and growth of plants under changing precipitation.