Prediction of climate change impacts on cotton yields in Greece under eight climatic models using the AquaCrop crop simulation model and discriminant function analysis

• AquaCrop was applied for cotton simulation in 7 areas of Greece.
• 3 Time periods of 8 climate models derived from the A1B SRES were used.
• Stepwise discriminant analysis was used to assess the efficiency of climate models for yield prediction.
• Two climate models (DMI and C4I) were chosen for estimating the cotton yield change.
• Western Greece could be more favoured to climate change than other areas.

The impact of climate change on cotton yields in seven main arable crop sites in Greece (Agrinio, Alexandroupoli, Arta, Karditsa, Mikra, Pyrgos, Yliki) was investigated. The FAO AquaCrop (v.4) water driven model was used as a crop growth simulation tool under eight climatic models (HadRM3, C4I, REMO-MPI, ETHZ, CNRM, DMI-HIRHAM, KNMI, SMHI) based on IPPC's A1B emission scenario. The mean values of the models ensemble for temperature were +1.8 °C until 2050 and +4 °C until the end of the century. The respective values for precipitation were −11% and −24%. The research was applied over three periods, 1961–1990, 2021–2050 and 2071–2100. AquaCrop was calibrated for 2006 and validated for 2005 and 2007 using the field data from the experiments carried out in Karditsa (Central Greece). Root Mean Square Error for yield and biomass was 0.17 and 0.49 t/ha, respectively, while Index of Agreement was 0.93 and 0.94. AquaCrop was run using the Growing Degree Day mode in order to account better for the temperature variations. However, it gave erratic results for some specific climatic models (SMHI, KNMI, CNRM) in some years within the period 1961–1990. A tendency towards increasing yields by the end of the century was detected for the majority of the climate models, especially in Western Greece (Arta, Agrinio, Pyrgos) and Northern Greece (Mikra, Alexandroupoli). The efficiency of the eight models for yield predictions in the seven sites was assessed by means of a discriminant function analysis. On the account of their function coefficients over the seven sites, it was found that the models DMI and C4I explained consistently a great proportion of variation among the three time periods whereas the models ETHZ, SMHI and KNMI were more efficient only in the periods 1961–1990, 2021–2050 and 2071–2099, respectively. By running the models DMI and C4I the relative impacts of climate change on seedcotton yield in the different areas were predicted and the results were discussed on the account of the corresponding changes in precipitation, temperature and crop evapotranspiration. These results will be useful for future irrigation planning in the study areas.