Comparing climate change impacts on cereals based on CMIP3 and EU-ENSEMBLES climate scenarios

Abstract

This study presents the simulated impact of climate change on cereal production for multi-model ensembles of global and regional climate models (GCMs and RCMs). The study aims are (i) to assess the sensitivity of impacts to the use of climate scenarios from specific ensembles of GCMs or RCMs, and (ii) to quantify the uncertainty in predicted impacts resulting from differences in climate models from a multi-model ensemble. The impact on cereal yield (for maize and winter wheat) was assessed under climate projections of an ensemble of 15 low-resolution GCMs from the Coupled Model Intercomparison Project phase 3 (CMIP3) and an ensemble of nine high-resolution RCMs from the EU-ENSEMBLES project. Climate projections for the middle of the 21st century (2031–2050) were stochastically downscaled to local-scale weather scenarios for two locations in the Flemish Region (Belgium) using the LARS-WG weather generator. The local weather data were applied as input in two crop simulation models: AquaCrop for maize and Sirius for winter wheat. The study showed mostly positive climate change effects on maize and winter wheat in the region. Higher yield increases were simulated under climate scenarios with less pronounced temperature increases, mainly following a less pronounced reduction of the length of the growing season. For maize, a crop that is prone to water stress and hence vulnerable to changes in precipitation, simulated impacts differ between climate scenarios from CMIP3 versus EU-ENSEMBLES ensembles. For simulated impacts on wheat, inter-model variation within an ensemble was more important than inter-ensemble variation. The dominance of inter-ensemble, inter-model or spatial variation depends on the crop of study and the relation between the regional climate and agronomic conditions. Although it can only be hypothesized that EU-ENSEMBLES-based scenarios may be more advanced than CMIP3-based scenarios for agricultural assessments due to their better performance in representing sub-GCM-grid topography-dependent changes and distribution of dry and wet days, the study results demonstrate definitely that the choice for one or another ensemble of climate models (with different resolution) adds to the overall uncertainty in impact assessments, even when the climate projections are downscaled to the local level via statistical inference. Using scenarios from RCMs driven by a limited number of GCMs is discouraged as they would probably not give a representative range of possible impacts.