Background/Question/Methods   Global change impacts on the biosphere involve a complex mixture of factors and will act on large spatial scales and long time scales. Long-term assessments and predictions of these impacts on local, regional and global environments are typically conducted by models, which are both based on and tested against knowledge gained from experiments. Such experiments are restricted in space (small test plots), time (mostly < 10 years) and the degree of complexity that can be taken into account (number of global change drivers tested, species composition, interaction between plants, symbionts and pathogens, atmospheric and soil feedback). Consequently, the validity of model predictions depend on how well results from these experiments are scaled up in time, space and complexity.

Results/Conclusions   Here, we show that the relative size of an effect, regardless of the nature of both the treatment (e.g. increased CO2 or temperature) and the studied parameter (e.g. biomass production or plant transpiration), tends to decrease with increasing spatial and temporal scales of the study, as well as with increasing levels of complexity on which the experiment is conducted. Modelling studies tend to overestimate effect sizes with increasing scale and complexity, either due to inadequate up-scaling of experimental results or due to insufficient mechanistic representation of the involved processes. The general pattern of reduced effect size with increasing spatial and temporal scale as well as complexity may result from an increasing probability of mutual cancelling of effects at larger scales and in more complex (realistic) systems. Our findings have far-reaching implications for the interpretation and conclusions drawn from existing experiments as well as the design of future experiments and models. In particular, short-term experimental results from artificial, homogenous systems are likely to overestimate long–term and large–scale responses of the biosphere to global change.