Evaluation of bias correction methods in improving the direct model output of temperature in CMIP over Iran

Document Type : Research Article


1 M.Sc. Student of Climatology, Department of Geography, Ferdowsi University of Mashhad, Mashhad, Iran.

2 Associate Professor of Climatology, Ferdowsi University of Mashhad,

3 Postdoctoral Researcher of Climatology, Ferdowsi University of Mashhad, Department of Geography, Mashhad.


The general circulation models (GCMs) are state-of-art tools available to investigate the response of climate system to external and internal forcing. They are used to predict/project climate in seasonal to decadal time scales. The general circulation models (GCMs) have more or less biases, and bias correction methods are one of the techniques used to correct their biases. The Coupled model intercomparison project phase 6 (CMIP6) has been widely used to simulate the historical period and project the future climate. However, due to the uncertainty of the models and their coarse resolution, GCMs are not directly used to assess the impacts of climate change. Therefore, to reduce the uncertainty in CMIP6 models, bias correction is necessary in the first step. This study evaluates three methods of bias correction including, Linear Scaling, Variance Scaling Of Temperature, Empirical Quantile Mapping, Quantile mapping using a smoothing spline and Empirical Robust Quantile Mapping for two variables of minimum and maximum temperature against 46 synoptic stations in Iran during 1980-2014 using the EC-Earth3-CC. To evaluate direct model output (DMO) and bias correction methods, we used three metrics including root-mean-square error (RMSE), percent bias (PBIAS), index of agreement (d), and interannual variability skill score (IVS). The results showed that the direct model output of the EC-Earth3-CC model has a cold bias (underestimation) for both minimum and maximum temperature in all climate zones of Iran, as well as the area-averaged of the country. The bias correction methods examined in this research have significantly reduced the bias in Iran. It is found that the bias decreased to 51.8 percent for the minimum temperature in the highlands of Azerbaijan, northeastern Iran, as well as parts of the Alborz and Zagros mountains. In general, the results of the three methods are close and they are not much different from each other. Based on the analysis of RMSE values, bias correction methods significantly reduced RMSE in comparison with DMO. So that the value of this metric in DMO has been more than 2 oC in most of Iran's climate zones, while the use of bias correction methods has reduced the error value to less than 1 oC. Also, bias correction methods have increased the index of agreement (d) by more than two times in average climate zones. Since the EC-Earth3-CC DMO has a good performance in depicting interannual climate variability (IVS) and is close to the observations, this has caused the DMO to not differ greatly from the results of using bias correction methods such as linear scaling. Finally, the bias correction methods used in this research estimate the maximum temperature with higher accuracy than the minimum temperature. There is no single bias correction method that provides the best performance in all regions. Therefore, each of these methods has its own advantages and limitations, which are caused by spatiotemporal differences and local geographical features.


Main Subjects