عنوان مقاله [English]
The Intergovernmental Panel on Climate Change (IPCC) has observed that climate variable changes associated with global warming are affecting different sector of human society. Changes in precipitation and temperature are anticipated as direct driving factors because they are the main factors that impact regional hydrological processes.
The general circulation models (GCMs) are the most important tool for predicting the future climate change, which can reproduce important processes about global and continental scale atmosphere and project future climate under the different scenarios.
The use of climate model simulations is now widespread, but there's still a risk in using these data, given the existing biases as well as the inability of these models to represent regional topography and land-sea contrast properly due to the coarse resolution of GCMs which makes local climate projection a huge challenge. Teutschbein and Seibert (2012) state that simulations of temperature and precipitation using GCMs often show significant biases due to systematic model errors or discretization and spatial averaging within grid cells, which hampers the use of simulated climate data as direct input data for climate change studies. Bias correction procedures are used to minimize the discrepancy between observed and simulated climate variables on a daily time step so that the corrected simulated climate data match simulations using observed climate data reasonably well. Many bias correction methods, ranging from simple scaling techniques to the rather more sophisticated distribution mapping techniques, have been developed to correct biased GCM outputs.
This study aims to select the best model and bias correction method based on different statistical metrics to downscale the daily precipitation and Maximum and Minimum temperature (Tax and Thin) outputs of selected CMIP6 models for Mashhad and Goldmkan synoptic stations, which are located in the Kashaf Rroud basin.
1. Materials and Methods
For this study, daily observations of precipitation, minimum temperature and maximum temperature during 1989-2019 of two synoptic stations in the Kashaf Roud basin were used. Over the last few years, the Copernicus database has been of great help to researchers in the field of climate change studies for the preparation and aggregation of climate data, observational data, satellite data, etc. This database has enabled researchers to pre-select models as well as specify the desired time and place. The historical CMIP6 data have been download from the Copernicus Climate Data Store (CDS).
First, the precipitation and temperature time series were qualified. RHtests-dlyPrcp and RHtestsV4 packages in the R software environment were then used to test the homogeneity of the precipitation and temperature daily time series.
CMhyd (Climate Model data for hydrologic modeling) is a software that has been used to extract and bias-correct data obtained from the selected global climate models using three statistical methods e.g., linear scaling (LS), distribution mapping (DM) and delta change (DC). The bias, RMSE, Pearson, Spearman and Kendall correlation coefficients and the Taylor diagram were used to determine the accuracy of the results of each model and to select the best method to downscale the daily and monthly outputs of GCMs. There are two sets of data in the CMIP6 simulations:
2-1. Historical data
Historical data of CMIP6 covers the period of 1850-2014, which can be used as a reference period to compare and verify the performance of each GCM.
2-2. Scenarios data
SSP scenarios provide different pathways for future climate forcing. They typically cover the period 2100-2006.
2. Results and Discussion
In this study, daily precipitation and temperature time series from two synoptic stations, Mashhad and Golmakan, located in the Kashaf Roud basin, were first used as primary quality control. Then the homogeneity of these data was tested. Maximum and minimum temperatures and precipitation in Golmakan were homogeneous. Mashhad was homogeneous for precipitation and minimum temperature time series, but for maximum temperature it had a change point on 09/30/1994, which was homogenized in the series and then used. The large-scale data of precipitation and temperature of three models have been scaled down to the level of the stations using three statistical methods and have been corrected for bias. Root-mean-square error (RMSE), correlation coefficients (Pearson, Spearman and Kendall) and bias (RB) were calculated to determine the accuracy and performance of each model and each exponential downscale method. In a further step, the daily time series of these data were converted to monthly data, then all validation process were done for monthly data in order to make a better and more comprehensive decision based on these results.
Among downscale results of rainfall and temperature using LS, DM and DC methods; in all models and at both stations for daily and monthly rain; LS method showed better results. However, the DM method gives better results for the maximum and minimum temperature data (except for the minimum temperature in the ACCESS-CM2 model, where the LS method is slightly better for both Mashhad and Golmakan). With a difference of 1.6 mm in the DM method, the largest bias in the monthly precipitation data for Mashhad is observed in the MIROC6 model.
The results of this research show that LS for precipitation and DM for temperature will have higher accuracy in the Kashaf Roud basin. MRI model performed better for precipitation and maximum temperature using LS method, however ACCESS performed better for minimum temperature in these stations. Exponential down scaling using the DM method gives better results for precipitation in Mashhad for the MRI model and in Goldmkan for the ACCESS model. MRI gives better results at Mashhad and MIRO at Goldmkan. The DM method also simulates more accurate results at both stations of the MRI model for the minimum temperature. Therefore, the MRI model is ranked first and ACCESS and MIRO are ranked second and third, respectively, if we want to select a comprehensive model among these three models that has the best simulations of precipitation and temperature.