全球变化知识资源中心

Conventional methods of modelling impacts of future climate change on crop yields often rely on a limited selection of projections for representing uncertainties in future climate. However, large ensembles of climate projections offer an opportunity to estimate yield responses probabilistically. This study demonstrates an approach to probabilistic yield estimation using impact response surfaces (IRSs). These are constructed from a set of sensitivity simulations that explore yield responses to a wide range of changes in temperature and precipitation. Options for adaptation and different levels of future atmospheric carbon dioxide concentration [CO2] defined by representative concentration pathways (RCP4.5 and RCP8.5) were also considered. Model-based IRSs were combined with probabilistic climate projections to estimate impact likelihoods for yields of spring barley (Hordewn vulgare L.) in Finland during the 21st century. Probabilistic projections of climate for the same RCPs were overlaid on IRSs for corresponding [CO2] levels throughout the century and likelihoods of yield shortfall calculated with respect to a threshold mean yield for the baseline (1981-2010).

Results suggest that cultivars combining short pre- and long post-anthesis phases together with earlier sowing dates produce the highest yields and smallest likelihoods of yield shortfall under future scenarios. Higher [CO2] levels generally compensate for yield losses due to warming under the RCPs. Yet, this does not happen fully under the more moderate warming of RCP4.5 with a weaker rise in [CO2], where there is a chance of yield shortfall throughout the century. Under the stronger warming but more rapid [CO2] increase of RCP8.5, the likelihood of yield shortfall drops to zero from mid-century onwards.

Whilst the incremental IRS-based approach simplifies the temporal and cross-variable complexities of projected climate, it was found to offer a close approximation of evolving future likelihoods of yield impacts in comparison to a more conventional scenario-based approach. The IRS approach is scenario-neutral and existing plots can be used in combination with any new scenario that falls within the sensitivity range without the need to perform new runs with the impact model. A single crop model is used for demonstration, but an ensemble IRS approach could additionally capture impact model uncertainties.