Beyond the ordinary metrics on the evaluation of historical Euro-CORDEX simulations over Türkiye: the mutual information approach

Vazifehkhah S., Kahya E., Gao W., Delju A.

Theoretical and Applied Climatology, vol.153, no.1-2, pp.829-851, 2023 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 153 Issue: 1-2
  • Publication Date: 2023
  • Doi Number: 10.1007/s00704-023-04492-3
  • Journal Name: Theoretical and Applied Climatology
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, IBZ Online, PASCAL, Aqualine, Aquatic Science & Fisheries Abstracts (ASFA), Artic & Antarctic Regions, BIOSIS, CAB Abstracts, Environment Index, Geobase, Index Islamicus, INSPEC, Pollution Abstracts, Veterinary Science Database
  • Page Numbers: pp.829-851
  • Istanbul Technical University Affiliated: Yes


Previous studies applied a large variety of performance metrics to evaluate the global and regional climate model simulations; however, there is still a huge debate to justify the rationale for the chosen metrics. The performance of sixty Euro-CORDEX temperature and precipitation simulations was investigated in temporal and spatial approaches through various common metrics over Türkiye. In addition, several mutual information (MI) methods based on the information theory were evaluated as state-of-the-art alternative metrics and compared with the applied common metrics. Based on the average of outputs over the ensemble of the driving models for the annual temperature, the MBE, MASE, MRAE, and NSE are presenting a similar pattern on the rank of the simulations. The MPI with 0.3 on MBE, NCC with 2.7, 2.9, and − 10.4 on MASE, MRAE, and NSE, respectively, and IPSL-LR and NOAA with 0.1 on the modified KGE represented the least errors. The ICHEC with respective 15, 0.1, 1.01, and − 4 for the MBE, MASE, MRAE, and NSE presented the lowest errors for the similar above-mentioned analysis except for the precipitation. The MPI and CNRM with 0.37, 0.37, and 0.08 obtained the highest outcomes on the KNN, mixed and noisy KNN, respectively. We conclude that the ability of mutual information to capture nonlinear relationships is very beneficial. Finally, it is also suggested to undertake these analyses for other hydroclimatic variables for future studies to gain a comprehensive insight into the performance of MI methods.