Comparison of precipitation projections of CMIP5 and CMIP6 global climate models over Yulin, China

This study compared precipitation projections of CMIP5 and CMIP6 GCMs over Yulin City, China. The performance of CMIP5 and CMIP6 GCMs in replicating Global Precipitation Climatology Centre (GPCC) precipitation climatology of the city was evaluated using different statistical metrics. The best performing GCMs common to both CMIP5 and CMIP6 were finally selected and subsequently downscaled to GPCC resolution using linear scaling method to assess spatiotemporal changes in precipitation in the basin. The study revealed BCC.CSM1.1(m), IPSL.CM5A.LR, MRI.CGCM3, and MIROC5 of CMIP5 and their equivalents BCC-CSM2-MR, IPSL-CM6A-LR, MRI.ESM2.0, and MIRCO6 of CMIP6 as the most suitable GCMs for the projection of precipitation in Yulin. This study revealed changes in precipitation in the range of −14.0 to 0.0% and −22.0 to 0.2% during 2021−2060 for RCP4.5 and SSP245 scenarios, respectively. The precipitation was projected to decrease more during 2061–2100 for both the scenarios. The highest decrease of −29.7 to −22.0% was projected by MRI-ESM-2-0 for SSP2-45, while −28.0 to −20.0% by MIROC5 for RCP4.5. For RCP8.5 and SSP5-85 scenarios, precipitation was projected to decrease in the range of −17.0 to −2.0% and −32.0 to 0.0%, respectively, during 2021–2060 by most of the GCMs. An increase in precipitation up to 12.3% was projected only by IPSL-CM5A-LR for RCP85 for this period. A further decrease in precipitation was projected by all GCMs during 2061−2100 for both RCP8.5 and SSP5-85 scenarios. The highest decrease was projected by MIROC5 (−40.2 to −29.0%) for RCP8.5 and IPSL-CM6A-LR (−40.2 to −26.0%) for SSP5-85. Overall, the results revealed a higher decrease in precipitation in Yulin City by CMIP6 GCMs compared to that projected by their corresponding GCMs of CMIP5 for both scenarios. This study can be of significance in the planning and mitigation of climate change as it gives insight into the expected changes in precipitation and the possibility of the choices of the best performing GCMs.


Introduction
Climate change has been an issue of concern over several decades due to its devastating impacts. It has increased the risks of flooding (Manawi et al. 2020;Rahman et al. 2019), occurrences of droughts (Alamgir et al. 2019;Ayugi et al. 2020a;Shiru et al. 2019a), heatwaves (Kang and Eltahir 2018;Khan et al. 2019;Satyanarayana and Rao 2020), and ecosystem damages (Kim et al. 2019;Pérez-Ruiz et al. 2018). Many sectors including water resources (Sharafati et al. 2021, agriculture (Islam and Nursey-Bray 2017;, energy, and health among others have also been affected by the changing climate. Like many other parts of the globe, China is also experiencing the impacts of climate change. Flooding is a common occurrence in different parts of China including cities like Yulin (Huang et al. 2015;Yang et al. 2017), and it often causes damages to sectors such as power, agriculture, 1 3 and health and sometimes leads to loss of lives (He et al. 2018). Flood occurrences in Yulin City date back to 1961 (Yin et al. 2020) with about 149 rainstorms in the area from 1971 to 2016 (He et al. 2018). The Shaanxi area to which Yulin belongs is sensitive to climate change (Yin et al. 2020) and has been recently affected by climate change (Dai et al. 2013;Ma and Fu 2003). In 2003, heavy precipitation caused river flooding leading to the destruction of 17,438 acres of farmlands, destruction of roads, and damages to about 1,458 houses (He et al. 2018). Heavy precipitation affected more than 150,000 people in 2012 with 15 people missing and 11 deaths, destruction of roads, and disruption of aviation due to the heavy rain (China-Daily 2012). The disaster caused a direct economic loss in the Yulin and Yan'an regions up to 134 million yuan ($21 million). Similarly, the flood in 2016 affected more than 21,000 people and evacuation of about 1200 persons (He et al. 2018). The flood led to the damage of more than 850 houses, destruction of more than 1100 km of roads, and damages to 30 bridges and culverts and caused 11 landslides and other related geological disasters. The economic implication of the disaster was about 150 million yuan. The region is also known to be affected by droughts Yin et al. 2020).
Understanding the expected changes in climate is crucial for the areas susceptible to disasters in order to develop adaptation and mitigation plans against climate change. This is particularly important using the recently released global climate models (GCM) of the Coupled Model Intercomparison Project Phase 6 (CMIP6). In addition, as global climate models (GCMs) structures can be sources of uncertainties in climate change projections, evaluations of the different GCMs are crucial for the choice of the model(s) or ensemble models for the projection of climate in an area. It is also important to assess how the CMIP6 differs in projection from the previous Coupled Model Intercomparison Project Phase 5 (CMIP5) in order to streamline the existing adaptation measures based on the predecessor Song et al. 2021).
GCMs are developed by different institutions and have different performances in different parts of the globe (Chen et al. 2017). Hence, the selection of the most realistic ones for a reliable projection of climate for a region is crucial. The assessment of the performances of GCMs has been conducted using different statistical indices (Rivera and Arnould 2020; Sreelatha and Anand Raj 2019). However, due to contradictory outputs from different statistical measures, supporting such outputs with other measures can be beneficial in reaching a compromise.
As GCMs are characterized by coarser spatial resolutions, their applications in climate projections and climate change impact studies at local and regional scales can be unreliable (Onyutha et al. 2016;Pour et al. 2018;Salman et al. 2018). Therefore, they are required to be downscaled (Almazroui et al. 2020a;Sa'adi et al. 2019;Shiru et al. 2020) using either the dynamical downscaling or the statistical downscaling (SD) methods. The SD method is known to have the advantages of computational efficiency and cost-effectiveness, the possibility of incorporation of observations directly into methods, and provision of point-scale climatic projections from GCM scale (Fowler et al. 2007). Studies have found different SD methods suitable for downscaling of climate data. For example, quantile perturbation was found to be skillful in removing biases in precipitation over Norway (Gudmundsson et al. 2012), Kenya (Ayugi et al. 2020b), and Belgium (Tabari et al. 2021). Linear scaling was found to outperform power transformation, general quantile mapping, and gamma quantile mapping methods over Syria (Homsi et al. 2020) and Nigeria (Shiru et al. 2019b). Over Costa Rica, Mendez et al. (2020) found that empirical quantile mapping, delta method, and power transformation marginally outperformed parametric quantile mapping techniques of gamma quantile mapping and gamma-pareto quantile mapping methods. The performance of bias correction methods is dependent on location; therefore, selection of method for an area is crucial ).
An array of statistical metrics was used to select the best performing GCMs for the projection of precipitation in the study area. Equivalent GCMs of the CMIP5 and CMIP6 were selected for the comparison of projections. The selected GCMs were downscaled using linear scaling method. The methodology proposed in this study employs only the GCMs which can reliably simulate the exiting climate of the study area and, thus, capable to provide a trustworthy comparison of CMIP5 and CMIP6 projects. It is expected that the comparison of precipitation projections of CMIP5 and CMIP6 would help in streamlining the existing adaptation measures formulated based on CMIP5 projections or deriving new measures based on new scenarios of CMIP6. This will be very beneficial in the mitigation of the risks of disasters such as flooding and droughts in the area. So far, the introductory part of this study has been given. Henceforth, the study area and datasets are discussed in Section 2, methodology in Section 3, results in Section 4, discussion in Section 5, and conclusions in Section 6.

Study area
The study area, Yulin (Figure 1), is located in the northern Shaanxi province of China (longitude: 107° 15ʹ 47ʺ−111° 14ʹ 44ʺ E; latitude: 36° 49ʹ 07ʺ−39° 34ʹ 47ʺ N). Yulin covers a total area of 385 km by 263 km (Zha et al. 2008). The terrain of the area descends from 1907 m in the east to 585 m in the west above the mean sea level. Yulin has a semi-arid temperate continental monsoon type climate which is characterized by dry and little precipitation in spring and winter and high precipitation during summer and autumn (Liu et al. 2020). The annual average precipitation in Yulin is around 300-500 mm (Wang et al. 2012). The annual average temperature in the area is 9.6°C (Wang 2016).

Gauge-based gridded precipitation
The GPCC Full Data Monthly Product Version 2018 at 0.5° of the Deutscher Wetterdienst (Schneider et al. 2018) was used in this study as the reference data. The GPCC precipitation product among most other precipitation products has the advantages of (1) good data quality for hydrological studies, (2) availability for a longer period, (3) better performance as being developed using the highest number of collected precipitation records, and (4) completeness of time series for the recent decades (Ahmed et al. 2017;Spinoni et al. 2014). Some studies have found the GPCC to outperform many other gridded data in many parts of the globe including China, the region to which the study area of this study belongs. In the neighboring Pakistan, Ahmed et al. (2019b) evaluated four gridded precipitation data, namely Asian Precipitation Highly Resolved Observational Data Integration Towards Evaluation (APHRODITE), Global Precipitation Climatology Centre (GPCC), Center for Climatic Research-University of Delaware (UDel), and Climatic Research Unit (CRU) for arid, semi-arid, and hyper-arid regions. Though the study revealed variations in the performances of the products, GPCC precipitation data was reported to perform better than the other products at all climatic regions. Over Central Asia, three gridded datasets Global Precipitation Climatology Centre (GPCC) V7, Climatic Research Unit (CRU) TS 3.22, and Willmott and Matsuura (WM) precipitation datasets were compared with the observed ). Among them, GPCC was found to have the highest correlation and lowest bias with the observed. Over China, GPCC was found to outperform Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Record, PERSIANN-CDR, and the Climate Hazards Group Infrared Precipitation with Station data version 2.0 CHIRPS) (Wei et al. 2019). The GPCC was also found to have the ability of replicating the precipitation properties over mainland China (Yu et al. 2020). Therefore, this study uses the monthly precipitation data of the GPCC for the period 1961-2005. This period was considered since the CMIP5 data is only available till the 2005 period. So a later period of up to 2014 for which the CMIP6 is available was not considered in the study. Data were collected from a total of 100 grid points to cover the whole Yulin. The location of the grid points is shown in Figure 1.

Global climate models
This study uses the historical and future simulations of GCMs of CMIP5 and CMIP6. The CMIPs are sets of globally coordinated GCM simulations which comprise historical and future climate simulations assembled from different climate modeling groups. The CMIP5 offers significant improvements compared to the CMIP3 (Taylor et al. 2012). The CMIP5 comprises four scenarios called representative concentration pathways (RCPs). In the CMIP6, the four RCP scenarios of CMIP5, RCP2.6, RCP4.5, RCP6.0, and RCP8.5 have been updated as shared socioeconomic pathways (SSPs) scenarios, SSP1-2.6, SSP2-4.5, SSP4-6.0, and SSP5-8.5, respectively. Each GCM also considers 2100 radiative forcing levels. The GCMs of CMIP6 are developed through improved emission scenarios, land use data, physical processes, and model parameterization to provide more realistic projections of future climate (Eyring et al. 2016;O'Neill et al. 2016). In this study, historical and future simulations of 10 GCMs of CMIP5 and CMIP6 were considered. The GCMs were chosen based on their availability from the same institution. The first ensemble members for both CMIP5 and CMIP6 were considered and downloaded from https:// esgf-node. llnl. gov/ search/ cmip5/ and https:// esgf-node. llnl. gov/ proje cts/ cmip6/, respectively. The GCMs chosen for this study are provided in Table 1.

Methodology
The methods used in this study are described in this section. For an unbiased comparison of model performance, the GPCC and the precipitation simulations of all GCMs were re-gridded to a uniform resolution of 1°×1° (latitude × longitude) using bilinear interpolation to have a uniform resolution. Bilinear interpolation is often conducted for transforming spatially coarse GCM data into finer data through GCM data interpolation from the four nearest neighboring grid points (Ahmed et al. 2019a;Almazroui et al. 2020a). After selection of GCMs, the selected GCMs and the GPCC precipitation data were re-gridded to 0.25° and used for spatiotemporal projection of precipitation.

Statistical indices
The ability of the different GCMs in reproducing the properties of GPCC precipitation at all the 100 grid points of the study area was assessed using four statistical indices: normalized root mean square error (NRMSE), percentage of bias (Pbias), volumetric efficiency (VE), and coefficient of determination (R 2 ). Besides, probability density function and spatial relationship of the mean monthly precipitation of the different GCMs were compared with GPCC precipitation to assess their abilities in replicating the precipitation climatology of the study area. Details about the statistical metrics used in this study are as follows. The expressions used to describe statistical metrics used here: x pred,i and x obs,i are the i-th GCM and GPCC data respectively.
The magnitude of the errors in estimation can be summarized by NRMSE (Willmott 1982). It is a normalized statistic that provides a relative magnitude of the residual variance to the variance of the measured data. Smaller NRMSE values (preferably zero) indicate better performance of the model. NRMSE is defined as follows: The tendency of GCM to under-or overestimate the GPCC precipitation is measured using Pbias. Model performance is better when the Pbias is closer to zero. The Pbias is a statistical metric that gives the estimate of overestimation or under prediction of a model (Wagena et al. 2018). The evaluation of Pbias is conducted as follows.
The VE measures the ratio between GCM and GPCC precipitation volumes over a period, where a VE value of 1 indicates a perfect estimation. It can be calculated using the following equation.
The R 2 can be defined as the square of Pearson's productmoment correlation coefficient (i.e., R 2 = r 2 ) describing the proportion of the total variance in the GPCC precipitation which is explainable by GCM (Legates and McCabe Jr 1999). R 2 values can range between −1.0 and 1.0, in which the higher absolute value indicates a better agreement. Computation of R 2 is as follows.

Downscaling of precipitation of selected GCMs of CMIP5 and CMIP6
The linear scaling (LS) method was applied for the downscaling of the selected GCMs. LS (Lenderink et al. (2007) uses the monthly correction values obtained from the difference in GPCC and GCMP simulated monthly precipitation for the reference period. The monthly scaling factor is then applied to raw GCM data. The monthly precipitation, P, is corrected using the following equation: P o is the monthly mean of GPCC precipitation, whereas P s is the monthly mean of the GCM simulated precipitation. LS method requires less information such as only monthly data to calculate the scaling factor (Lafon et al. 2013) and, thus, widely used for precipitation downscaling. (2)

Results
This section presents the results obtained from the projections of the precipitations of the CMIP5 and CMIP6 GCMs over Yulin using four statistical indices for evaluating their performances and LS method for downscaling.

Performance assessment of GCMs using statistical metrics
The results of the statistics used for the performance evaluation of different historical GCMs are presented in Table 2. It shows a variation in the performances of different GCMs. For CMIP5, the GCMs with the best performance metrics are ACESS1.3, BCC.CSM1.1(m), IPSL. CM5A.LR, and MIROC5. The GCMs of the same institutions also showed good performances for CMIP6 except for ACCESS.ESM1.5 which showed a relatively poor performance. With the exception of some of the models of the CMIP6, e.g., GISS-E2-1-G, there was a generally better performance of the models over the CMIP5. This indicates that there has been an improvement in many of the models of the CMIP6. It also showed that some of the models have become poorer for the CMIP6, e.g., INM-CM4-8. This emphasizes the needs for model evaluations before their usage for climate projection.

Spatial relationship between GCMs and GPCC precipitation
The ability of different GCMs in replicating the spatial distribution of GPCC precipitation for the study area is presented in Figure 2. The performances of the GCMs were found to vary widely in reproducing the GPCC precipitation. Among CMIP5 GCMs, the highest overestimations were by CanESM2 and NorESM1-M, while an underestimation was by MRI-CGCM3 followed by IPSL-CM5A-LR in some parts. For CMIP6, GISS-E2-1G showed the highest overestimation of precipitation. Generally, the GCMs with better performance metrics (Table 2) showed better skills in replicating the precipitation climatology of GPCC for the study area. As seen from Figure 2, CMIP5 GCMs, the CMIP6 GCMs generally showed better performances in replicating the spatial properties of precipitation of the GPCC over the study area indicating improvements in many of its models except some like GISS-E2-1-G which showed significant overestimation and MRI-ESM2-0 which showed significant underestimation.

Comparison using probability density function
The PDFs of monthly GCM precipitation were compared with the PDF of GPCC precipitation for the study area. The results for CMIP5 and CMIP6 are presented in Figure 3a and b, respectively. Figures show that most GCMs were able to replicate the precipitation properties of the GPCC especially the skewness. However, the distribution of precipitation was found better for the GCMs which showed a better performance in terms of statistical metrics presented in Table 2. Also, some of the models of the CMIP6 showed better replication of the PDFs indicating that some of the CMIP5 GCMs have been improved in the newly released CMIP6 GCMs. Figures show that CanESM2 of the CMIP5 and GISS-E2-R of the CMIP6 have the least replicability abilities of the PDF of the GPCC.

Comparison of GCMs in reproducing monthly mean precipitation
The mean monthly GCM precipitation was compared with mean monthly GPCC precipitation for the period 1961−2005. Obtained results for the GCMs of CMIP5 and CMIP6 are presented in Figure 4a and b, respectively. There was a variation in the estimation of GPCC precipitation by the GCMs, especially during the wet season (May to October). CanESM2 and NorESM1-M of CMIP5 overestimated the precipitation for all the months. Overestimation by CanESM2 was 125mm in September, while overestimation by NorESM1-M was 85mm in August. MRI-CGCM3 which has the highest underestimation of the GPCC underestimated it by 57mm during the month of August. For the CMIP6, overestimation was by GISS-E2-1-2G and underestimation by MRI-ESM2-0 and IPSL-CM5A-LR, especially during the wet period. GISS-E2-1-2G overestimated precipitation by highest amount of 130mm during September. MRI-ESM2-0 and IPSL-CM5A-LR have highest underestimation of 20mm and 25mm during July and August, respectively. Though most of the CMIP6 GCMs were found to overestimate the GPCC precipitation, overall they were found more capable in replicating the mean monthly precipitation of GPCC compared to CMIP5 GCMs. This indicates that there have been improvements in many of the GCMs of the CMIP6.

Selection of the best performing GCMs
The performances of the GCMs based on statistical indices and replication of PDF and spatial precipitation distribution patterns were considered in selecting the highest performing models. Based on the statistics, BCC.CSM1.1(m), IPSL.CM5A.LR, and MIROC5 were the better performing GCMs for both CMIP5 and CMIP6. Besides, ACCESS1.3 showed an overall better performance in comparison to other CMIP5 GCMs, while its equivalent under CMIP6 showed a poor performance. As CMIP6 is a more recent simulation, the MRI.CGCM3 for CMIP5 and its equivalent GCM for CMIP6, MRI.ESM2.0, were prioritized as the fourth model for projection of precipitation over the study area.

Projection of precipitation from the selected GCMs of CMIP5 and CMIP6
The spatial projections of precipitation for the study area by CMIP5 GCMs for RCP 4.5 and CMIP6 GCMs for SSP2-45 for two future periods, 2021-2060 and 2061-2100, are presented in Figure 5. Large heterogeneity in precipitation changes was projected by different GCMs for RCPs and SSPs scenarios and the two projection periods. During 2021-2060, the highest decrease in precipitation was projected by MRI-ESM-2-0 for SSP2-45, while the highest decrease for RCP 4.5 was projected by MIROC5. For the same period, BCC-CSM2-MR projected an increase in precipitation by 1.2% at the extreme west of the study area. Percentage change in precipitation was in the range of −14.0 to 0.0% for RCP4.5, while it was −22 to 1.2% for SSP2-45. During 2061-2100, all the GCMs projected decreases in precipitation, with the highest decrease of −29.7 to −22.0% by MRI-ESM-2-0 for SSP2-45 and −28.0 to −20.0% by MIROC5 for RCP4.5. The lowest decrease during this period was projected by MRI-CGCM3 and BCC-CSM2-MR for RCP4.5 and SSP2-45 scenarios, respectively. The spatial distribution of projected precipitation in the study area by CMIP5 GCMs for RCP8.5 and CMIP6 GCMs for SSP5-85 for the periods 2021-2060 and 2061-2100 are presented in Figure 6. Compared to RCP 4.5 and SSP2-45 scenarios, RCP8.5 and SSP5-85 scenarios showed higher decreases in precipitation. During 2021-2060, the projected decrease in precipitation was in the range of −17.0 to −2.0% by the CMIP5 GCMs for RCP8.5, while the decrease was projected between −32.0 and 0.0% by the CMIP6 GCMs for SSP5-85. Increases in precipitation of 0.0-12.3% were noticed only for IPSL-CM5A-LR for RCP85. During 2061-2100, the decrease in precipitation was projected highest by MIROC5 (−40.2 to −29.0%) for RCP8.5 and IPSL-CM6A-LR (−40.2 to −26.0%) for SSP5-85.

Discussion
Climate change remains a major challenge in many parts of the globe as they have devastating impacts on several sectors. Yulin considered for this study like many other parts of the globe has been affected by climate change impacts. The climate of the area has had increased warming tendencies which has led to droughts in the area in recent times (Wang et al. 2012;Yan 2015). This has resulted into desertification in addition to human activities (Wang et al. 2012). The increase in temperature of nearly 0.04 °C/year in the recent times , which is higher than the average temperature rise over China (0.5 ~ 0.8 °C during last 100 years), has been reported to have affected the water resources of the area with difference in supply and demand for water being observed (Xiao-jun et al. 2015). Many studies have shown that the expected changes in climate will result in increased temperatures and more erratic precipitations in many parts of the world. Projection of precipitation under CMIP5 in Nigeria showed that while precipitation will increase in some parts of the country, particularly the semiarid and arid regions where they are usually low, the other parts where they used to be higher will experience decreases (Shiru et al. 2019b). CMIP5 GCMs was used in Syria and showed that precipitation would increase by up to 87% in some parts and decrease by up to −85% in the coastal areas (Homsi et al. 2020). Many other studies have also shown both increase and decrease in precipitation in many parts of the world using CMIP5 GCMs (Iqbal et al. 2020;Narsey et al. 2020;Shiru and Park 2020;Ullah et al. 2020).
Studies conducted using the recently released CMIP6 GCMs have also shown such changes in precipitation. A study conducted over South Asian countries showed that the annual mean precipitation will increase by 27.3% in India,18.9% in Bhutan,26.4% in Pakistan,19.5% in Nepal,25.1% in Sri Lanka,and 17.1% in Bangladesh in the last part of the century under SSP5-8.5 scenario (Almazroui et al. 2020c). Over Africa, projected precipitation under CMIP6 showed that while the northern and the southern parts of Africa are expected to witness a reduction in precipitation, the central parts are expected to have increases of 6.2%, 6.8%, and 9.5% for SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios, respectively Almazroui et al. 2020b). In China, studies showed variation in the projected precipitations for both CMIP5 and CMIP6 with some studies showing mostly increases, while others showed a simultaneous increase and decrease depending on the region. The annual precipitation was projected to increase significantly relative to the present day for CMIP5 (Chen 2013). The study revealed that the increase in precipitation in the Northwest of China is primarily due to the increase in light showers, while the increases in precipitation in the north and northeastern parts are due to an increase in medium precipitation. The increases in precipitation are expected in the southern parts of China due to an increment in heavy precipitation. Over China, a study conducted using CMIP5 showed both increases and decreases in daily precipitation under different warming conditions (Zhou et al. 2019). Increase in warming of 1.5°C was projected to cause an increase in the frequency and intensity of precipitation in Northeast China, North China, and the Qinghai-Tibet Plateau, while there The projection of the changes in precipitation over northwestern China for CMIP6 (Su-Yuan et al. 2020) showed that there would be lesser warming compared to that previously expected, which would affect the patterns in precipitation changes. Unlike in the historical period (1850-2014) when the rate of warming was 0.05°C per decade, the annual mean temperature is projected to increase up to 0.06°C, 0.26°C, and 0.59°C per decade for SSP126, SSP245, and SSP585, respectively, for the period 2015-2099. The total annual precipitation for the area is projected to increase by 5.6, 6.4, and 8.0 mm/decade for SSP126, SSP245, and SSP585, respectively.
In this study, both increases and decreases in precipitation are projected for the study area. Zhou et al. (2019) projected increases in precipitation over China, except some northwestern parts where Yulin belongs. This supports the

Conclusions
This study compares the projections of precipitation by CMIP5 and CMIP6 GCMs over Yulin City of China. Different statistical and graphical metrics were used in assessing the ability of 10 GCMs in replicating the precipitation properties of the study area. Finally, four GCMs having the highest ability in replicating the properties of GPCC precipitation were selected for the projection of precipitation over the study area. The study revealed that ACESS1.3, BCC. CSM1.1(m), IPSL.CM5A.LR, and MIROC5 of CMIP5 and their equivalents in CMIP6, BCC-CSM2-MR, IPSL-CM6A-LR, MRI.ESM2.0, and MIROC6, have better abilities in replicating historical precipitation properties over Yulin and were therefore selected for projection of precipitation. Projection of precipitation showed an overall decrease in Fig. 6 Spatial distribution of projected precipitation by the GCMs of CMIP5 and CMIP6 during 2021-2060 and 2061-2100 for RCP8.5 and SSP5-85 scenarios precipitation over Yulin by the GCMs of both CMIPs. The decrease in precipitation would be more in the far future compared to the near future. The expected decreases in precipitation can increase the frequency and intensity of droughts in the area. The findings of the study can be used as a guide in the development of adaptation and mitigation measures against climate change in the area. In the future, more GCMs common to both CMIP5 and CMIP6 can be employed (when they will be available for CMIP6) for the selection of best performing GCMs. Besides, a more reliable approach can be utilized for the selection of GCMs to avoid the dispute in selection using conventional statistical metrics.

Data availability
The authors can support all relevant data if requested.

Code availability
The authors can support if necessary.

Declarations
Ethics approval This article does not contain any studies with human or animal participants performed by any of the authors.

Consent to participate
The authors declare that they have consent to participate in this paper.

Consent for publication
The authors declare that they have consent to publish in this journal.

Conflict of interest
The authors declare no competing interests.