Combining field experiments and predictive models to assess potential for increased plant diversity to climate-proof intensive agriculture

Agricultural production systems face increasing threats from more frequent and extreme weather fluctuations associated with global climate change. While there is mounting evidence that increased plant community diversity can reduce the variability of ecosystem functions (such as primary productivity) in the face of environmental fluctuation, there has been little work testing whether this is true for intensively managed agricultural systems. Using statistical modeling techniques to fit environment–productivity relationships offers an efficient means of leveraging hard-won experimental data to compare the potential variability of different mixtures across a wide range of environmental contexts. We used data from two multiyear field experiments to fit climate–soil–productivity models for two pasture mixtures under intensive grazing—one composed of two drought-sensitive species (standard), and an eight-species mixture including several drought-resistant species (complex). We then used these models to undertake a scoping study estimating the mean and coefficient of variation (CV) of annual productivity for long-term climate data covering all New Zealand on soils with low, medium, or high water-holding capacity. Our results suggest that the complex mixture is likely to have consistently lower CV in productivity, irrespective of soil type or climate regime. Predicted differences in mean annual productivity between mixtures were strongly influenced by soil type and were closely linked to mean annual soil water availability across all soil types. Differences in the CV of productivity were only strongly related to interannual variance in water availability for the lowest water-holding capacity soil. Our results show that there is considerable scope for mixtures including drought-tolerant species to enhance certainty in intensive pastoral systems. This provides justification for investing resources in a large-scale distributed experiment involving many sites under different environmental contexts to confirm these findings.