Evaluation and Projection of Climate Change in the Second Songhua River Basin Using CMIP6 Model Simulations

Abstract

The aim of this study is to evaluate the performance of the Global Climate Model (GCM) of the Coupled Model Intercomparison Project Phase 6 (CMIP6) in historical simulations of temperature and precipitation. The goal is to select the best performing GCMs for future projection of temperature and precipitation in the Second Songhua River Basin under multiple shared socioeconomic pathways (SSPs). Interannual variability skill (IVS) and Taylor diagrams are used to evaluate the spatiotemporal performance of GCMs against temperature and precipitation data published by the China Meteorological Science Commons during 1956–2016. In addition, five relatively independent models are selected to simulate the temperature and precipitation for 2021–2050 using Hierarchical Clustering. The selected models are CMCC-ESM2, EC-Earth3-Veg-LR, IPSL-CM6A-LR, MIROC-ES2L, and MPI-ESM1-2-HR. The projected results find that SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios show an increasing trend of future annual mean temperature and precipitation. However, for annual precipitation, there is a mixed state of increase and decrease among different models on the seasonal scale. In general, future temperature and precipitation changes still show a trend of growth and uneven distribution in the Second Songhua River Basin, which may be further accelerated by human activities.

1. Introduction

The Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) indicates that human activities have exacerbated regional climate change and led to more extreme weather events [1]. The Coupled Model Intercomparison Project (CMIP) is a project that evaluates the performance of GCMs in simulating historical, present, and future climate change under different scenarios [2,3]. CMIP6 is the latest version of CMIP and has superior simulation capability and projection accuracy compared to previous versions [4]. The CMIP6 collects several simulations of GCMs currently used to interpret past and future climate changes [2,5]. The latest GCMs improve spatial resolution, improve the parameterization scheme of physical processes, and increase the representativeness of climate change feedback. GCMs are the most advanced and effective tools for studying historical and future climate [6]. However, uncertainties caused by the model structure, initial conditions, boundary conditions, calibration process and other factors may still affect the GCM outputs [7]. In order to ensure the credibility of the GCMs in projecting future climate elements, the performance of the GCMs in simulating the historical climate needs to be evaluated before applying the GCMs to verify that the GCMs are able to correctly simulate the physical mechanisms and trends of the climate system [8,9,10].

Previously, several studies have been conducted to evaluate the capability of GCMs from the old CMIP versions in simulating temperature and precipitation over China. Compared with the CMIP3 GCMs, the CMIP5 improved the simulation of temperature and precipitation, but they still underestimated the precipitation intensity [11,12,13]. In addition, based on the studies of CMIP5 GCMs, it is found that the temperature will increase significantly and the precipitation will increase relatively in the future of the 21st century [14,15,16,17]. The modeling capability of CMIP5 GCMs has been extensively assessed, but little attention has been paid to the spatiotemporal performance of the new CMIP6 GCMs over China.

Climate change is mainly characterized by the variability of temperature and precipitation, which is also an important criterion for evaluating the accuracy of GCMs [18,19]. A wide range of studies have been conducted to evaluate the performance of CMIP in simulating temperature and precipitation changes around the world [20,21,22,23]. Additionally, some studies’ comparative evaluations of CMIP6 and CMIP5 GCMs for the simulation of temperature and precipitation have been conducted in different regions of the world. Compared with CMIP5 GCMs, CMIP6 has improved in simulating extreme temperature and precipitation, and found that there have been improvements in spatial simulation ability in some individual CMIP6 GCMs relative to CMIP5 [24,25]. In another study, Sonia I. Seneviratne and Mathias Hauser [26] analyzed the predicted changes in extreme temperatures and heavy precipitation in the CMIP5 and CMIP6 multimodel ensembles, and the results revealed that the CMIP6 GCMs have better simulation capabilities. Some studies have reported that downscaling and bias revisions of the CMIP6 GCMs are necessary due to factors such as different resolutions, which may affect its performance in different regions. Zhu et al. [17] argue that CMIP6 GCMs perform differently for different regions and therefore, before using the output of CMIP6 GCMs, it is necessary to evaluate whether they can simulate local temperature and precipitation performance.

The Second Songhua River basin, located in the middle and high latitudes, is sensitive to climate warming. In recent years, extreme precipitation and drought events have greatly impacted its economic and social development and the safety of its inhabitants. As one of the important grain production bases in China, the basin has been seriously threatened by the rapid rise of temperature and the frequent occurrence of extreme precipitation events. Therefore, it is crucial to conduct scientific research on future climate change in the basin to provide guidance for future grain production.

This study aims to evaluate the spatiotemporal performance of CMIP6 GCMs and select the optimal model to accurately project the future temperature and precipitation changes in the Second Songhua River Basin. The first part evaluates and screens the spatiotemporal simulation capability of the GCMs for temperature and precipitation in the basin. The second part considers multiple SSPs and selects the best-performing model to project future temperature and precipitation in the basin during 2021–2050. This study can provide insight into the capability of GCMs in simulating temperature and precipitation in northeastern China and provide a basis for future climate projective studies of the northeastern region. In addition, the application of the best-performing CMIP6 GCMs can help to understand the changes of precipitation and temperature in the basin, and provide a basis for formulating timely measures to cope with climate change and reduce the hazards caused by climate change.

2. Data and Methods

2.1. Study Area

The Second Songhua River basin (124°36′—128°50′ E, 41°44′—45°24′ N) is located in Jilin Province, China, and covers an area of 78,180 km². The basin has a temperate continental monsoon climate with an average annual temperature of 3–5 °C and an average annual rainfall of 650–750 mm. Most of the rainfall occurs between June and September, accounting for 72% of the yearly total. Rainfall gradually decreases from southeast to northwest.

2.2. Data

2.2.1. Meteorological Data

The spatial distribution of the study area and the 17 selected hydrological stations is shown in Figure 1. Observed daily precipitation, and maximum and minimum temperature data within the study area are available for 17 hydrological stations in the basin from 1956 to 2016, which are from the China Meteorological Science Sharing Network (http://data.cma.cn (accessed on 19 August 2023)).

2.2.2. Climate Model Datasets and Emission Scenarios

The Scenario Model Intercomparison Project (ScenarioMIP) provides multi-model climate projections for CMIP6 [4]. These climate projections will be driven by the SSPs, and related to the Representative Concentration Pathways (RCPs). In this study, we used three emission scenarios as major future climate change scenarios: the high-emission scenario SSP5-8.5, the medium-emission SSP2-4.5, and the low-emission SSP1-2.6.

The climate model simulations used in this study were derived from the historical (1961–2010) and future (2021–2050) simulations of all publicly available GCMs from the CMIP6. The detail of the models is shown in

Table 1. Details of CMIP6 models used in the study.

No	Models’ Name	Country Name	Institute ID	Spatial Resolution (Lon × Lat)
1	ACCESS-CM2	Australia	CSIRO	144 × 192
2	ACCESS-ESM1-5	Australia	CSIRO	145 × 192
3	BCC-CSM2-MR	China	BCC	160 × 320
4	CanESM5	Canada	CCCMA	64 × 128
5	CMCC-ESM2	Italy	CMCC	192 × 288
6	CNRM-CM6-1	France	CNRM	128 × 256
7	CNRM-ESM2-1	France	CNRM	128 × 256
8	EC-Earth3	European Union	EC-Earth-Consortium	256 × 512
9	EC-Earth3-Veg	European Union	EC-Earth-Consortium	256 × 512
10	EC-Earth3-Veg-LR	European Union	EC-Earth-Consortium	160 × 320
11	HadGEM3-GC31-LL	UK	MOHC	144 × 192
12	INM-CM4-8	Russia	INM	120 × 180
13	INM-CM5-0	Russia	INM	120 × 180
14	IPSL-CM6A-LR	France	IPSL	143 × 144
15	MIROC6	Japan	MIROC	128 × 256
16	MIROC-ES2L	Japan	MIROC	64 × 128
17	MPI-ESM1-2-HR	Germany	MPI	192 × 384
18	MPI-ESM1-2-LR	Germany	MPI	96 × 192
19	MRI-ESM2-0	Japan	MRI	160 × 320
20	NorESM2-LM	Norway	NCC	96 × 144
21	NorESM2-MM	Norway	NCC	192 × 288
22	UKESM1-0-LL	UK	MOHC	144 × 192

. In order to resolve the differences in spatial resolution and facilitate comparison, the observed meteorological station data and the model simulation data with different resolutions were interpolated at a 0.125° × 0.125° resolution using the distance inverse interpolation method and the bilinear interpolation method, respectively.

2.3. Method of Model Evaluation

2.3.1. Simulation Effectiveness Evaluation Methods

For the evaluation of the model’s time simulation ability, this paper uses the Interannual Variability Index (IVS) as a measure.

$$IVS={\left(\frac{ST{D}_{\mathrm{m}}}{ST{D}_{\mathrm{o}}}-\frac{ST{D}_{\mathrm{o}}}{ST{D}_{\mathrm{m}}}\right)}^2$$

(1)

where STD_m is the standard deviation of each grid simulation datum in the time series and STD_o is the standard deviation of the measured data of each grid in the time series. The smaller the difference between the simulated interannual variability and the measured interannual variability, that is, the closer the IVS is to 0, the better the simulation results and the stronger the model’s time simulation ability. In order to characterize the temporal simulation capability of the GCMs, the IVS of all grid points were weighted and averaged to obtain a total IVS value.

A Taylor diagram can intuitively judge the correlation coefficient (R), the ratio of standard deviation (σ) and the center root mean square error (E′) between the model simulation results and the measured data on one diagram. This has been widely used in evaluating simulation ability. R represents the correlation of spatial distribution between simulation results and measured data. σ compares the discrete degree of spatial distribution between simulation results of each grid and measured data. E′ represents the similarity between simulation results and measured data.

The calculation method of R is shown in Equation (2):

$$R=\frac{\frac{1}{N}{\displaystyle \sum _{n=1}^N\left(f_n-\overline{f}\right)}\left(r_n-\overline{r}\right)}{{\sigma }_f{\sigma }_r}$$

(2)

where f_n is the multi-year average value of each grid model simulation, r_n is the multi-year average value of each grid measured; N is the number of grids; $\overline{f}$ is the regional mean of the simulated multi-year average, and $\overline{r}$ is the regional mean of the measured multi-year average; σ_f, σ_r is the standard deviation of f and r.

$${\sigma }_{\mathrm{f}}=\sqrt{\frac{1}{N}{\displaystyle \sum _{n=1}^N{\left(f_n-\overline{f}\right)}^2}}$$

(3)

$${\sigma }_r=\sqrt{\frac{1}{N}{\displaystyle \sum _{n=1}^N{\left(r_n-\overline{r}\right)}^2}}$$

(4)

The ratio of standard deviations (σ) can be calculated from Equation (5):

$$\sigma =\frac{{\sigma }_f}{{\sigma }_r}$$

(5)

The calculation method of central root mean square error (E′) can be expressed as

$$E^{\prime }={\left\{\left.\frac{1}{N}{\displaystyle \sum _{n=1}^N{\left[\left(f_n-\overline{f}\right)-\left(r_n-\overline{r}\right)\right]}^2}\right\}\right.}^{\frac{1}{2}}$$

(6)

Taylor index S is introduced to quantitatively evaluate the spatial simulation ability of the model. The calculation method is shown in Equation (7):

$$S=\frac{4\left(1+R\right)}{{\left(\sigma +{\sigma }^{-1}\right)}^2\times \left(1+R_{\max }\right)}$$

(7)

where R_max is the maximum value of the correlation coefficient. The closer the S value is to 1, the better the spatial simulation ability of the model is.

2.3.2. Similarity Test of Climate Model Simulation Results

Many models have the same parameterization process or belong to the same class. This can result in non-independence and correlation between models, leading to similarities in their simulation results. The number of consistent projected results can affect the credibility of the conclusions [27]. Therefore, this study examines the similarity of climate simulation results and explores the similarity of simulated temperature and precipitation between models [28]. The specific steps are referenced in the literature [29,30].

2.3.3. Climate Model Screening Methods

Considering the similarity of their climate simulation results, the 22 CMIP6 GCMs were divided into several relatively independent groups. From each group, representative models with better simulation effects were screened [29], and the specific process of climate model screening in this study is as follows.

(1)

Based on the results of precipitation hierarchical clustering, a threshold value was determined to categorize the precipitation into groups.

(2)

If there is only one model in the group, then select it directly.

(3)

If there is more than one model in the group, then select the model with the best simulation.

2.3.4. Correction of Model Simulation Results

In view of the bias in global model simulation results, the quantile map method is used to revise temperature and precipitation at the same frequency [31]. The principle of the quantile map revision is to find a transfer function to revise the model output at the same frequency. For the climate variable x [32], the mathematical expression of the method is shown in Equation (8):

$${\tilde{x}}_{m-p.adjst.}=F_{o-c}^{-1}(F_{m-c}(x_{m-p}))$$

(8)

where x_m−p and ${\tilde{x}}_{m-p.adjst.}$ are the simulated values before and after the model revision, respectively, and F_o−c and F_m−c are the cumulative probability density functions of the measured and model simulated values, respectively.

Li et al. [33] argued that this method is only applicable when the climate variables obey distributions that do not change significantly over time. To overcome this limitation, evaluating that the bias of the model in simulating the current climate is the same as the bias in simulating the future climate, it is proposed that this bias be used for the revision of the future simulated values, and Equation (8) is rewritten as

$${\tilde{x}}_{m-p.adjst.}=x_{m-p}\cdot F_{o-c}^{-1}(F_{m-p}(x_{m-p}))/F_{m-c}^{-1}(F_{m-p}(x_{m-p}))$$

(9)

$${\tilde{x}}_{m-p.adjst.}=x_{m-p}+F_{o-c}^{-1}(F_{m-p}(x_{m-p}))-F_{m-c}^{-1}(F_{m-p}(x_{m-p}))$$

(10)

Among them, Equations (9) and (10) are used for the revisions of precipitation and temperature, respectively.

To address the issue of interpolation spreading when using empirical functions for revisions, a 4-parameter beta distribution function and a hybrid function (consisting of a step function and a 2-parameter gamma distribution function) are proposed as the distribution functions for temperature and precipitation, respectively (see Equations (11) and (13)). The probability density function of the 4-parameter beta function is given by

$$\begin{array}{c}f(x;a,b,p,q)=\frac{1}{B(p,q){(b-a)}^{p+q+1}}{(x-a)}^{p-1}\cdot {(b-x)}^{q-1}\\ a\le x\le b;p,q>0\end{array}$$

(11)

The mixing function G(x) is

$$G(x)=(1-p)H(x)+pF(x)$$

(12)

where p is the proportion of days with precipitation to the total daily series; H(x) is a step function with precipitation taking the value of 1 and without precipitation taking the value of 0; F(x) is a 2-parameter gamma distribution function with a probability density function of

$$\begin{array}{c}f(x;k,\theta )=x^{k-1}\frac{e^{-x/\theta }}{{\theta }^k\mathrm{\Gamma }(k)}\\ x>0,k,\theta >0\end{array}$$

(13)

3. Results

3.1. Performance Evaluation of GCMs

3.1.1. Interannual Variability Skill

Figure 2 illustrates the IVS values for temperature and precipitation over the study area for 22 CMIP6 GCM simulations. Models with IVS values close to zero are considered optimal when simulating historical temperature and precipitation.

Figure 2a shows that the IVS values of simulating temperature for the 22 GCMs range from 0.001 to 0.535. Results indicate that the UKESM1-0-LL model is the worst performing model, in addition, the rest of the models have better performance with IVS scores below 0.300. Figure 2b shows that the IVS values of simulating precipitation for the 22 GCMs range from 0.058 to 1.218. It is noteworthy that 90% of the models have an IVS value of less than or equal to 1, indicating that they had a satisfactory performance in terms of IVS.

3.1.2. Taylor Diagram

Evaluating the performance of global climate models in simulating the spatial patterns of temperature and precipitation in the basin using Taylor diagrams (Figure 3). The results show that most of the models exhibited positive correlations with the observed data, indicating good spatial performance.

In terms of spatial simulation performance of temperature, the R values ranged from −0.544 to 0.918. The EC-Earth3 had the highest R value among the 22 GCMs. The closer the σ is to one, the better. As shown in our study, the σ ranged from 0.304 to 0.737, with ACCESS-CM2 showing high performance and ACCESS-ESM1-5 showing low performance in simulating spatial distributions. For most models, the E′ of temperature was between 0.40 and 1.50 °C. In terms of spatial simulation performance of precipitation, the R values ranged from 0.758 to 0.948. The NorESM2-LM had the highest R value among the 22 GCMs. Half of the models exhibited σ values between 0.60 and 1.00, indicating their ability to capture the precipitation cycle. The E′ of precipitation for most models was between 0.25 and 0.75 mm.

3.1.3. Ranking of GCMs

The ranking of IVS and S values for temperature and precipitation for the 22 GCMs is shown in

Table 2. Ranking of 22 CMIP6 models.

Model No	Model Name	Temperature				Precipitation
		IVS Value	Ranking	S Value	Ranking	IVS Value	Ranking	S Value	Ranking
M01	ACCESS-CM2	0.071	17	0.825	1	0.119	3	0.906	12
M02	ACCESS-ESM1-5	0.023	10	0.236	18	0.352	7	0.569	22
M03	BCC-CSM2-MR	0.08	18	0.473	15	0.058	1	0.747	18
M04	CanESM5	0.048	13	0.235	19	0.763	18	0.926	9
M05	CMCC-ESM2	0.004	3	0.642	8	0.392	10	0.935	7
M06	CNRM-CM6-1	0.001	1	0.611	10	0.934	20	0.688	20
M07	CNRM-ESM2-1	0.065	15	0.537	12	1.186	21	0.697	19
M08	EC-Earth3	0.021	9	0.825	2	0.367	9	0.989	2
M09	EC-Earth3-Veg	0.01	5	0.786	4	0.364	8	0.979	3
M10	EC-Earth3-Veg-LR	0.037	12	0.759	5	0.063	2	0.99	1
M11	HadGEM3-GC31-LL	0.014	8	0.588	11	0.605	15	0.939	6
M12	INM-CM4-8	0.049	14	0.18	22	0.482	12	0.928	8
M13	INM-CM5-0	0.013	7	0.197	20	0.59	14	0.977	4
M14	IPSL-CM6A-LR	0.001	2	0.653	7	0.29	6	0.765	17
M15	MIROC6	0.241	20	0.519	13	0.573	13	0.881	14
M16	MIROC-ES2L	0.031	11	0.195	21	0.461	11	0.861	15
M17	MPI-ESM1-2-HR	0.01	6	0.659	6	0.236	5	0.915	10
M18	MPI-ESM1-2-LR	0.006	4	0.282	16	0.194	4	0.889	13
M19	MRI-ESM2-0	0.157	19	0.81	3	0.714	17	0.911	11
M20	NorESM2-LM	0.066	16	0.264	17	0.607	16	0.629	21
M21	NorESM2-MM	0.292	21	0.474	14	1.218	22	0.834	16
M22	UKESM1-0-LL	0.535	22	0.625	9	0.788	19	0.972	5

. Each model is assigned a number from 1 to 22, with 1 being the best and 22 being the worst. In terms of temperature simulation, CNRM-CM6-1 ranks 1st in IVS value and 10th in S value, while UKESM1-0-LL ranks 22nd in IVS value and 9th in S value. ACCESS-CM2 ranks 1st in S value and 14th in IVS value, while INM-CM4-8 ranks 22nd in S value and 17th in IVS value. For precipitation simulation, BCC-CSM2-MR and NorESM2-MM both rank 1st in IVS value, but differ in S value, with BCC-CSM2-MR ranking 18th and NorESM2-MM ranking 16th. EC-Earth3-Veg-LR and ACCESS-ESM1-5 both rank 1st in S value, but differ in IVS value, with EC-Earth3-Veg-LR ranking 2nd and ACCESS-ESM1-5 ranking 17th. In summary, GCMs ranking significantly varies for different statistical indices. Some models rank among the best in both IVS and S values for temperature, but low for precipitation. Other models rank high in IVS values but low in S values. Therefore, we need to select the appropriate model to simulate temperature and precipitation.

3.2. Climate Model Screening Results

Figure 4 shows the hierarchical clustering maps based on precipitation simulations for 22 CMIP6 GCMs. For instance, the EC-Earth-Consortium (M08~M10) of the European Union includes three models with very similar precipitation simulations. These models have a relatively large weight in the projected results of 22 CMIP6 GCMs. If all three models from this institution are selected, it could impact the credibility of future climate change projections. It’s important to note that simulation results from different models within the same institution may vary greatly.

According to Table 2, the specific process of climate model screening in this study is as follows. The screening results are shown in

Table 3. CMIP6 pattern screening results.

Model No	Model Name	Precipitation				Temperature
		IVS Value	Ranking	S Value	Ranking	IVS Value	Ranking	S Value	Ranking
M05	CMCC-ESM2	0.392	10	0.935	7	0.004	3	0.642	8
M10	EC-Earth3-Veg-LR	0.063	2	0.990	1	0.037	12	0.759	5
M14	IPSL-CM6A-LR	0.290	6	0.765	17	0.001	2	0.653	7
M16	MIROC-ES2L	0.461	11	0.861	15	0.031	11	0.195	21
M17	MPI-ESM1-2-HR	0.236	5	0.915	10	0.010	6	0.659	6

(1)

Based on the precipitation hierarchical clustering results, precipitation was divided into five groups using a distance value of 0.55 as the threshold.

(2)

Among the precipitation patterns I to V, M17 (MPI-ESM1-2-HR) was selected as the only pattern in group III, and the patterns selected in groups I, II, IV and V were to be determined.

(3)

In group I, M14 (IPSL-CM6A-LR) had better IVS and S simulations than M06 (CNRM-CM6-1) and M07 (CNRM-ESM2-1), so it was selected. In group II, M02 (ACCESS-ESM1-5) was the worst model among all 22 models, so M16 (MIROC-ES2L), which was the other model in the same group, was selected. In group IV, M10 (EC-Earth3-Veg-LR) had better IVS and S simulations than the other models in the same group, except for M03 (BCC-CSM2-MR), which had the second-best IVS and S simulations, so it was selected. According to the above model selection methods and considerations, five representative models, M05 (CMCC-ESM2, Italy), M10 (EC-Earth3-Veg-LR, EU), M14 (IPSL-CM6A-LR, France), M16 (MIROC-ES2L, Japan), and M17 (MPI-ESM1-2-HR, Germany), were finally selected for impact evaluation.

3.3. Climate Change Projection Results

3.3.1. Temperature Changes

All five models under the SSPs scenarios projected an increase in temperature from 2021 to 2050 and showed a good agreement (

Table 4. Temperature change/°C under the SSPs scenarios in 2021–2050 (Base period: 1956–2016).

Scenario	Model	Year	Spring	Summer	Autumn	Winter
SSP1-2.6	CMCC-ESM2	1.3	1.2	1.5	1.0	1.5
	EC-Earth3-Veg-LR	1.1	0.5	1.2	1.5	1.3
	IPSL-CM6A-LR	2.1	1.9	1.7	1.9	2.7
	MIROC-ES2L	2.4	2.6	1.9	2.0	2.9
	MPI-ESM1-2-HR	0.8	0.4	0.8	0.7	1.3
SSP2-4.5	CMCC-ESM2	1.3	0.8	1.2	1.3	1.8
	EC-Earth3-Veg-LR	1.6	1.7	1.5	1.9	1.3
	IPSL-CM6A-LR	2.1	1.7	2.0	2.7	2.1
	MIROC-ES2L	2.9	2.7	2.6	3.0	3.2
	MPI-ESM1-2-HR	1.9	1.7	1.8	1.9	2.2
SSP5-8.5	CMCC-ESM2	1.5	1.3	1.5	1.1	2
	EC-Earth3-Veg-LR	1.6	1.3	1.2	2.0	1.8
	IPSL-CM6A-LR	2.5	2.1	2.1	2.6	3.2
	MIROC-ES2L	2.1	2.6	1.9	1.9	1.9
	MPI-ESM1-2-HR	1.3	1.1	1.2	1.1	1.6

). The annual mean temperature increased by 1.6 °C, 1.3 °C, and 1.6 °C, respectively. For the seasonal mean temperature, the increase was most pronounced in winter, about 0.28 °C higher than the annual mean, while the spring and summer temperatures showed a decreasing trend, about 0.20 °C and 0.16 °C lower than the annual mean, respectively. The projected temperature showed an increasing trend under the selected models, with MIROC-ES2L having the largest increase and MPI-ESM1-2-HR having the smallest increase in temperature under the SSP1-2.6 and SSP5-8.5 scenarios, respectively. In contrast, CMCC-ESM2 had the smallest increase under the SSP2-4.5 scenario.

3.3.2. Precipitation Changes

Table 5. Precipitation changes/% under the SSPs scenarios in 2021–2050 (Base period: 1956–2016).

Scenario	Model	Year	Spring	Summer	Autumn	Winter
SSP1-2.6	CMCC-ESM2	4.4	18.2	0.9	1.6	17.5
	EC-Earth3-Veg-LR	10.5	6.1	11.9	12.1	−1.1
	IPSL-CM6A-LR	5.7	12.9	6.2	−1.0	−2.5
	MIROC-ES2L	5.1	20.2	−0.3	4.0	40.6
	MPI-ESM1-2-HR	11.5	8.9	12.4	7.5	28.1
SSP2-4.5	CMCC-ESM2	−0.2	15.9	−4.6	−1.8	12.4
	EC-Earth3-Veg-LR	13.8	13.4	15.2	7.1	23.2
	IPSL-CM6A-LR	5.0	13.0	6.7	−9.4	7.8
	MIROC-ES2L	7.7	16.5	5.2	5.1	26.4
	MPI-ESM1-2-HR	9.4	−0.8	4.0	39.6	13.4
SSP5-8.5	CMCC-ESM2	4.2	19.3	0.7	−0.8	21.7
	EC-Earth3-Veg-LR	9.7	15.1	10.7	4.6	−8.2
	IPSL-CM6A-LR	6.6	11.1	8.2	1.7	−16.6
	MIROC-ES2L	2.6	17.1	−1.8	3.4	10.6
	MPI-ESM1-2-HR	10.6	7.9	10.5	12.0	19.1

shows the changes in precipitation under the SSPs scenarios from 2021 to 2050. The annual average precipitation is projected to have an increasing trend of 5.7%, 7.7%, and 6.6% under the SSPs scenarios, respectively. Seasonal average precipitation is projected to increase more in spring and winter than in other seasons, with increases ranging from 12.9% to 15.1% in spring and from 10.6% to 17.5% in winter. Autumn precipitation is projected to increase by a smaller amount than in other seasons, with increases ranging from 3.4% to 5.1%. Under the SSP1-2.6 and SSP2-4.5 scenarios, MIROC-ES2L projected the largest increase in spring and winter precipitation, while IPSL-CM6A-LR projected the smallest increase in winter precipitation under all three scenarios. Under the SSP5-8.5 scenario, CMCC-ESM2 had the smallest projected increase in spring and winter precipitation.

The spatial variation of precipitation from 2021 to 2050 under the SSPs scenarios is shown in Figure 5. It can be seen that the annual precipitation in the basin is increasing, except for some areas where precipitation decreases in autumn. The annual precipitation change in the basin is more than 2%, and below the Wu Dao Gou station the precipitation increases by more than 5%. On a seasonal scale, except for a decrease in precipitation in autumn, there are significant increases in other seasons, especially in spring and winter, when precipitation increases by more than 10%. Under the SSP1-2.6 scenario, precipitation below Feng Man Station increases significantly by more than 20% in winter. Under the SSP2-4.5 scenario, precipitation above Gaoli Chengzi Station exceeds 10% in spring and winter. Under the SSP5-8.5 scenario, precipitation at Gaoli Chengzi Station decreases by 2% in autumn.

The projected precipitation changes under the SSP1-2.6 scenario for 2021–2050 are shown in Figure 6. Annual precipitation will trend upwards in all GCMs under the SSP1-2.6 scenario, with EC-Earth3-Veg-LR and MPI-ESM1-2-HR projecting the most significant increases. Overall, all models projected an increase in spring precipitation. The CMCC-ESM2 projected a comparative decrease in summer precipitation above the Wu Daogou station and a decrease of more than 2% in autumn precipitation below the Han Yangtun station. The EC-Earth3-Veg-LR projected a significant decrease in winter precipitation in the basin, with some areas experiencing a decrease of more than 20%. The IPSL-CM6A-LR projected a decrease in precipitation above the Guan Madian station. The MIROC-ES2L projected a decrease in most areas above De Hui Station, with a significant increase of more than 20% in winter precipitation. The MPI-ESM1-2-HR projected an increase in precipitation on all time scales.

The projected precipitation changes under the SSP2-4.5 scenario for 2021–2050 are shown in Figure 7. The annual precipitation for the GCMs trended upwards under the SSP2-4.5 scenario, except for a decrease in the CMCC-ESM2 projection, with the EC-Earth3-Veg-LR projection showing the largest increase. CMCC-ESM2 projected a decrease of more than 5% above the Wu Daogou station in summer and more than 10% at the Feng Man station in autumn. In contrast, EC-Earth3-Veg-LR and MIROC-ES2L projected an increase in all seasons. IPSL-CM6A-LR projected a decrease in the basin in autumn, with some areas experiencing a 10–20% decrease. MPI-ESM1-2-HR showed an overall decrease in the basin in spring.

The projected precipitation changes for the 2021–2050 under the SSP5-8.5 scenario are shown in Figure 8. The annual precipitation for the GCMs trended upwards under the SSP5-8.5 scenario, with the most significant increase projected by MPI-ESM1-2-HR. CMCC-ESM2 projected a decrease in summer precipitation above the Wu Daogou station and a decrease in autumn above the Guan Madian station. EC-Earth3-Veg-LR projected a significant decrease in winter, with a decrease of more than 10% above the Gaoli Chengzi station. Similarly, IPSL-CM6A-LR projected a more significant reduction in winter precipitation, with a decrease of 20% above Hai Long station. MIROC-ES2L projected a slight decrease in summer precipitation. MPI-ESM1-2-HR showed an increase in precipitation in all seasons of the basin.

4. Discussion

In this study, the temporal and spatial simulation capabilities of the models were evaluated using IVS and Taylor diagrams. The five best performing GCMs were selected based on a hierarchical clustering approach, and projected future changes in temperature and precipitation in the basin under the SSPs.

The initial selection and detailed process of models are based on the simulation results of the projected changes in temperature and precipitation in this study [34]. To accomplish this, skill scores on temporal and spatial scales are considered. However, this method may have two main drawbacks. First, it may miss some good models that can simulate one of these elements well, but not the other. For instance, the BCC-CSM2-MR was found to perform best in precipitation simulations, but poorly in the assessment of temperature. Second, it may ignore some models that perform well in other subdomains. Such as, CMCC-ESM2 had moderate performance in simulations of both temperature and precipitation, and in general could simulate both elements better. The application of these models to temperature and precipitation has been documented in other literature. Zhu et al. [35] found that CMCC-ESM2 and MPI-ESM1-2-HR were among the best performers of CMIP6 GCMs for temperature and precipitation in China, respectively. Luo et al. [36] compared the simulation of CMIP6 and CMIP5 GCMs for precipitation in China and found that the IPSL-CM6A-LR was better for precipitation simulation. Gopinadh and Naresh [37] evaluated the simulation capability of the CMIP6 model for precipitation in India and found that MIROC-ES2L could simulate seasonal average precipitation better. Another study [38] examined the simulation capability of CMIP6 GCMs on the interannual and interdecadal time scales of SST in the Indian Ocean and found that the EC-Earth3-Veg-LR has better simulation capability for temperature by skill scoring results. There is no GCM can show good performance for all regions [39].

Temperature was projected to increase under the three scenarios, with greater warming in autumn and winter. This is consistent with the findings of Lu et al. [40] who projected a continued increase in future temperatures from 2015–2060, particularly under the SSP5-8.5 scenario. And an increase in precipitation of 5.7%, 7.7% and 6.6% under the three scenarios, with larger increases in spring and winter. Similar results were obtained by Tian et al. [40] who projected an increasing trend in future precipitation as well, however, this is different from our results, and the reason for this is also mentioned in his study: it is possible that precipitation increases significantly in all scenarios after 2060, especially in the SSP5-8.5 scenario. Our study reveals that higher emission scenarios do not mean higher temperatures and more precipitation [41], but also that future precipitation changes are not only heavily dependent on temperature [42], but that temperature-related factors may be influenced by atmospheric circulation, storms, and other factors [43]. It can also be seen that different study periods may also lead to different conclusions, for example our study period is 2021–2050, while Krishna et al. [44] divides the study period into near (2015–2044), mid-future (2045–2074) and distant-future (2075–2100), so that we have different results. The reasons for these differences may be derived from different emission scenarios, feedback mechanisms and internal variability of GCMs.

There are differences between the various models in CMIP6, and not all models are suitable for every study. This necessitates careful model selection and revision. The differences in GCMs may arise from factors such as model structure, regional topography, and atmospheric circulation. Future climate change could impact water resources and agricultural production in different regions. Changes in temperature and precipitation under the SSPs may pose threats to food security and human survival. For instance, an increase in autumn temperatures could lead to droughts, while an increase in summer precipitation could result in floods. These potential impacts should be considered in future studies.

Topographic features have a significant impact on climate change [45,46]. Firstly, topographic features affect air movement and water vapor condensation, leading to differences in precipitation and temperature in different regions. Secondly, they affect the distribution and variability of temperature by influencing the surface energy balance and altering surface temperature, humidity and albedo. In addition it enhances monsoon precipitation, the impact of which increases with global warming. Further consideration of orographic factors for mapping is important for regional flood control, etc.

5. Conclusions

In this study, we quantitatively evaluated the performance of 22 CMIP6 GCMs in modeling historical temperature and precipitation in the Second Songhua River Basin. We also projected future temperature and precipitation in the basin from 2021 to 2050. We used a Taylor diagram and three error indicators (R, σ, and E′), as well as hierarchical clustering for screening. Considering the similarity of climate simulation results and the simulation effects of different models on temperature and precipitation, we selected five models: CMCC-ESM2, EC-Earth3-Veg-LR, IPSL-CM6A-LR, MIROC-ES2L, and MPI-ESM1-2-HR. These GCMs were used for three SSPs to make more reliable projections of temperature and precipitation for the period 2021–2050. The results show that temperature and precipitation increase significantly under the three scenarios. Precipitation shows an increasing trend at the annual scale from 2021 to 2050, while at the seasonal scale it shows a mixed state of increase and decrease among the different models. This study focused only on the Second Songhua River Basin; further studies can be conducted for the whole of China.