Evaluation and Projection of the Influence of the August Asian–Pacific Oscillation on Precipitation in Northern Xinjiang Based on CMIP6 Simulations

Abstract

Based on CMIP6 model data and reanalysis data, two multi-model ensemble means—the “best” model ensemble (BMME) and the negative correlation ensemble (NCE)—were derived from 30 models to simulate the August Asian–Pacific Oscillation (APO) and the influence of the August APO on September precipitation over northern Xinjiang (NXPI). The results show that BMME performs better than individual models in simulating the eddy temperature in August. Overall, the BMME-simulated APO intensity shows a general decreasing trend from 2015 to 2100. Based on NCE, regressions of the precipitation and 850-hPa wind fields onto the APOI reproduce spatial patterns similar to the observations under the historical scenario. Furthermore, the NCE-simulated correlation between APO Index (APOI) and NXPI remains steadily negative during 2021–2040 under both SSP2-4.5 and SSP5-8.5 scenarios, but the negative correlation weakens significantly over the subsequent 60 years. This may be related to the southeastward shift of the negative geopotential height anomaly center over East Asia.

1. Introduction

The Asian–Pacific Oscillation (APO) is distinguished by a zonal dipole pattern in the mid-to-upper tropospheric eddy temperature spanning the Asia-Pacific region [1]. The APO not only exerts a substantial influence on precipitation patterns across East Asia [2,3,4,5] but also plays a pivotal role in modulating key features of Northern Hemisphere atmospheric circulation, such as tropical cyclone activity in the western North Pacific [6], sea surface temperature patterns in the North Pacific [7] and North Atlantic [8], precipitation variability over North America [9], and its teleconnection with the North Atlantic Oscillation [10] and the Pacific Decadal Oscillation [11]. While the APO attains its peak intensity during summer, its presence is also discernible in other seasons [12]. Recent research endeavors have delved into the cross-seasonal impacts of the APO on precipitation in China.

The Coupled Model Intercomparison Project Phase 6 (CMIP6) provides a comprehensive framework for evaluating and comparing climate models from different modeling centers worldwide [13]. In CMIP6, the participating models reproduce the principal Earth system components, including atmospheric, oceanic, terrestrial, and sea-ice processes, driven by historical and future forcing conditions. These models are widely used to investigate climate variability, project future climate conditions, and assess regional impacts, such as precipitation and temperature [14,15,16]. However, studies on the APO and its relationship with precipitation in CMIP6 models are relatively limited and mainly focus on the summer season. Based on previous studies [17], we evaluate 30 CMIP6 models to obtain multi-model ensemble means that reproduce the observations. Using the ensembles, we further calculate the projected changes in August APO intensity and the relationship between August APO and September precipitation in northern Xinjiang, with the aim of providing projections for precipitation in this region.

The structure of this paper is organized into three sections. Section 2 details the data and methods, followed by the results presented in Section 3. Finally, Section 4 provides the conclusions and discussion.

2. Data and Methods

2.1. Data

Monthly data from the r1i1p1f1 member (the standard physical configuration and first initialization ensemble member of each CMIP6 model) of CMIP6 models under the historical, SSP2-4.5, and SSP5-8.5 scenarios were used in this study. The variables analyzed include precipitation, air temperature, geopotential height, and horizontal wind components [18,19].

Table 1. Details of the 30 CMIP6 models used in this study.

Model Number	Model Name	Institution/Country	Horizontal Resolution (lat × lon)
1	ACCESS-CM2	CSIRO-ARCCSS/Australia	144 × 192
2	ACCESS-ESM1-5	CSIRO/Australia	144 × 192
3	AWI-CM-1-1-MR	AWI/Germany	192 × 384
4	BCC-CSM2-MR	BCC/China	160 × 320
5	CanESM5	CCCma/Canada	64 × 128
6	CAS-ESM2-0	CAS/China	128 × 256
7	CESM2-WACCM	NCAR/USA	192 × 288
8	CIESM	THU/China	192 × 288
9	CMCC-CM2-SR5	CMCC/Italy	192 × 288
10	CMCC-ESM2	CMCC/Italy	192 × 288
11	EC-Earth3-CC	EC-Earth-Consortium/Europe	256 × 512
12	EC-Earth3-Veg-LR	EC-Earth-Consortium/Europe	256 × 512
13	EC-Earth3-Veg	EC-Earth-Consortium/Europe	256 × 512
14	EC-Earth3	EC-Earth-Consortium/Europe	160 × 320
15	FGOALS-f3-L	CAS/China	180 × 288
16	FGOALS-g3	CAS/China	80 × 180
17	FIO-ESM2-0	FIO-QJNM/China	192 × 288
18	CDFL-ESM4	NOAA-GFDL/USA	180 × 288
19	INM-CM4-8	INM/Russia	120 × 180
20	IPSL-CM6A-LR	IPSL/France	143 × 144
21	KACE-1-0-G	NIMS-KMA/Republic of Korea	144 × 192
22	KIOST-ESM	KIOST/Republic of Korea	96 × 192
23	MIROC6	MIROC/Japan	128 × 256
24	MPI-ESM1-2-HR	MPI-M/Germany	192 × 384
25	MPI-ESM1-2-LR	MPI-M/Germany	96 × 192
26	MRI-ESM2-0	MRI/Japan	160 × 320
27	NESM3	NUIST/China	96 × 192
28	NorESM2-LM	NCC/Norway	96 × 144
29	NorESM2-MM	NCC/Norway	192 × 288
30	TaiESM1	AS-RCEC/China	192 × 288

provides a concise overview of the CMIP6 models included in this study. More comprehensive information for each model can be found at https://esgf-node.llnl.gov/search/cmip6/ (accessed on 25 September 2025).

The atmospheric parameters, taken from the NCEP/NCAR reanalysis archive [20], are provided at a spatial resolution of 2.5° × 2.5°. The precipitation fields come from the CN05.1 daily product created by Wu and Gao [21], with a grid resolution of 0.25° × 0.25°.

To facilitate comparison with observations, all CMIP6 variables except precipitation (pr) were interpolated onto a 2.5° × 2.5° grid using bilinear interpolation to match the resolution of the reanalysis data. Subsequently, both the CN05.1 and 30 CMIP6 precipitation datasets were interpolated to a 1.5° × 1.5° grid.

2.2. Methods

The Taylor diagram evaluates the relationship between model outputs and observations based on three metrics: correlation coefficient in space (CC), centered root-mean-square error (RMSE), and the standard deviation ratio (Ratio) [22]. Models exhibiting CC and Ratio near 1, along with RMSE approaching 0, show better consistency with observations. In this study, the Taylor diagram is employed to evaluate how well CMIP6 models capture the spatial patterns.

This study uses the interannual variability skill score (IVS) introduced by Chen et al. to evaluate the capability of each CMIP6 model in simulating interannual variability [23]. The IVS is defined as:

$$IVS={\left({\displaystyle \frac{ST{D}_M}{ST{D}_O}}-{\displaystyle \frac{ST{D}_O}{ST{D}_M}}\right)}^2$$

(1)

Here, $ST{D}_M$ and $ST{D}_O$ represent the interannual standard deviations of the model and observational variables, respectively. The closer the IVS value is to 0, the better the model’s ability to simulate interannual variability.

The Comprehensive rating metrics (MR) is an indicator that reflects the overall ranking of a model’s ability to reproduce various aspects of the observations [24]. It is defined as follows:

$$MR=1-{\displaystyle \frac{1}{nm}}{\sum }_{i=1}^{n}ran{k}_i$$

(2)

Here, $n$ represents the number of indicators included in the comprehensive ranking, $m$ denotes the number of models evaluated, and $ran{k}_i$ refers to the ranking of each indicator. Models with MR values approaching 1 exhibit better simulation skill.

According to Zhao [1], eddy temperature is calculated as $T^{\prime }=T-\overline{T}$, where $T$ represents the vertical mean of air temperature between the 500 and 200 hPa, and the $\overline{T}$ represents its zonal mean. Following previous work [17], the APOI is calculated as:

$$\mathrm{APOI} =T_{{85}^{\circ }–{135}^{\circ }\mathrm{E},{35}^{\circ }–{55}^{\circ }\mathrm{N}}^{\prime }-T_{{160}^{\circ }\mathrm{E}–{130}^{\circ }\mathrm{W},{20}^{\circ }–{45}^{\circ }\mathrm{N}}^{\prime }$$

(3)

The northern Xinjiang precipitation index (NXPI) is defined as the area-averaged precipitation over (43–50° N, 82–94° E) in September.

Except for the Taylor diagram analysis, which used the climatological mean of the original data, linear trends were removed from the data in all other analyses. The statistical significance of the correlation coefficients was evaluated through Student’s t-test.

The reanalysis data and the historical scenario cover the period from 1961 to 2014, while the SSP2-4.5 and SSP5-8.5 scenarios span from 2015 to 2100.

3. Results

3.1. The Impact of August APO on September Precipitation and Atmospheric Circulation Based on Observations

The leading EOF mode of upper-tropospheric $T^{\prime }$ reveals a zonal dipole pattern (Figure 1a), with positive $T^{\prime }$ anomalies over East Asia and negative $T^{\prime }$ anomalies over the North Pacific. Figure 1b shows the negative correlation between the precipitation field and the APOI for 1961–2014, based on observations. The anomalous cyclonic circulation caused by negative geopotential height anomalies over the Arabian Sea and East Asia (Figure 1c) can generate northerly wind anomalies along the moisture pathway from the Indian Ocean to northern Xinjiang. These northerly wind anomalies are unfavorable for precipitation in northern Xinjiang (Figure 1d). The APO signal in August can persist into September and trigger anomalous atmospheric circulation. Zhu and Hua further discussed that these thermal anomalies can induce the above-mentioned northerly wind response in September using the Linear Baroclinic Model, which suppresses precipitation in northern Xinjiang [17].

3.2. Evaluation and Projection of the August APO

To evaluate the performance of 30 models in simulating the August APO, we conducted Taylor diagram analysis of the horizontal spatial modes of the 500–200 hPa $T^{\prime }$ climatology (Figure 2a). Among the 30 models, the CC relative to observations ranged from 0.70 (KIOST-ESM) to 0.98 (CESM2-WACCM), the Ratio from 0.67 (KIOST-ESM) to 1.38 (ACCESS-ESM1-5), and the RMSE from 0.29 (AWI-CM-1-1-MR) to 0.72 (KIOST-ESM). Overall, CMIP6 models show good skill in simulating the spatial patterns of the August APO, capturing its main spatial features.

In addition, we used the IVS to assess the capability of each CMIP6 model in representing interannual fluctuations (Figure 2b). The results indicate substantial differences among models, with MPI-ESM1-2-LR performing best (IVS = 0.007). Based on the four metrics—CC, Ratio, RMSE, and IVS—we conducted MR of the 30 models (Figure 2c). The top five models (AWI-CM-1-1-MR, EC-Earth3-Veg-LR, GFDL-ESM4, EC-Earth3-Veg, and EC-Earth3-CC) were selected to generate the BMME. Taylor diagram analysis (Figure 2a) shows that BMME simulates the August ${\mathrm{T}}^{\prime }$ more accurately than any individual model, with CC = 0.97, Ratio = 1.04, and RMSE = 0.25.

We further calculated the 21-year moving average of the BMME-simulated APOI from 1961 to 2100 under different scenarios (Figure 2d). Under both scenarios, the APO intensity exhibits a general decreasing trend from 2015 to 2100. Notably, the difference reaches its maximum in 2071, when the APOI under the SSP2-4.5 scenario is 0.403 higher than that under the SSP5-8.5 scenario. After 2089, the APOI under the SSP5-8.5 scenario surpasses that under the SSP2-4.5 scenario. Moreover, considering the ±1 standard deviation of the BMME, the projected range under the SSP5-8.5 scenario is larger than that under the SSP2-4.5 scenario.

3.3. Evaluation and Projection of the Impact of August APO on September Precipitation over Northern Xinjiang

Figure 2a shows that most of the CMIP6 models can simulate the August APO well. Similarly, we conducted a Taylor diagram analysis of the precipitation over northern Xinjiang in September from 30 models (Figure 3a). Among the 30 models, the CC relative to observations ranged from −0.13 (NorESM2-LM) to 0.88 (MRI-ESM2-0), the Ratio from 0.0007 (CIESM) to 4.59 (CAS-ESM2-0), and the RMSE from 0.22 (MPI-ESM1-2-HR) to 1.18 (CAS-ESM2-0). Overall, most models do not perform well in simulating September precipitation over northern Xinjiang.

To evaluate the influence of the August APO on September precipitation over northern Xinjiang in CMIP6 models, we calculated the correlation between the APOI and NXPI for 1961–2014 under the historical scenario (Figure 2b). Most models reasonably captured the negative correlation between APOI and NXPI. The three models that passed the 90% confidence level (BCC-CSM2-MR = −0.27, MIROC6 = −0.27, and MPI-ESM1-2-HR = −0.38) were averaged to form the negative correlation ensemble (NCE). The correlations between APOI and NXPI in the observations and the NCE are −0.27 and −0.31, respectively.

To evaluate the performance of the NCE, we compared it with the All-Model Mean Ensemble (AMME) under the historical scenario during 1961–2014. In the regressions of September precipitation anomalies over northern Xinjiang onto the APOI, both the NCE (Figure 4d) and AMME (Figure 4a) exhibited a clear negative correlation, with the NCE showing a stronger negative correlation. For the AMME, the APOI exhibited a significant positive correlation with the 500-hPa geopotential height over East Asia (Figure 4b), which is inconsistent with the observed negative correlation. In contrast, the NCE reproduced a negative correlation (Figure 4e), although it did not pass the significance test. In the regressions of 850-hPa wind fields onto the APOI, both the AMME (Figure 4c) and NCE (Figure 4f) exhibited northerly wind anomalies along the moisture pathway from the Indian Ocean to northern Xinjiang, consistent with observations and effectively suppressing precipitation. However, over northern Xinjiang, the NCE reproduced northerly wind anomalies similar to that observed, whereas the AMME did not. This may explain why the NCE provides a more reasonable assessment than the AMME, offering a basis for predicting the relationship between the APO and precipitation over northern Xinjiang in future projections.

We further calculated the correlation coefficients between APOI and NXPI for NCE and individual models under the SSP2-4.5 and SSP5-8.5 scenarios (Figure 5a–d). During 2021–2040 (Figure 5a), NCE exhibited higher correlations than other individual models, with r = −0.33 under SSP2-4.5 and r = −0.43 under SSP5-8.5. Under the SSP5-8.5 scenario for 2041–2100, the negative correlations weakened across three stages, remaining at −0.13, −0.14, and −0.26, respectively. Under the SSP2-4.5 scenario, the negative correlation during 2041–2060 decreased to −0.19, and in 2061–2080 and 2081–2100, it turned positive, with correlations of 0.27 and 0.19, respectively (

Table 2. Correlations between APOI and NXPI simulated by NCE.

	2021–2040	2041–2060	2061–2080	2081–2100
SSP2-4.5	−0.33	−0.19	0.27	0.19
SSP5-8.5	−0.43	−0.13	−0.14	−0.26

).

Previous experiments have shown that the negative correlation between August APO and September precipitation in northern Xinjiang weakens to varying degrees in the future. We further examined the lagged impact of APO on precipitation and atmospheric circulation under the SSP2-4.5 (Figure 6a–f) and SSP5-8.5 (Figure 6g–l) scenarios. Under both scenarios, the regression results of NCE outperform those of AMME. Compared with the SSP2-4.5 scenario (Figure 6d), NCE under the SSP5-8.5 scenario maintained a stronger negative correlation between APOI and the precipitation field (Figure 6j), consistent with the results shown in Figure 5. Comparing Figure 6e and Figure 6k, relative to the observational data, the center of the negative geopotential height anomalies in NCE shifts southeastward in the future, which weakens the control of the anomalous cyclone on precipitation over northern Xinjiang. This may explain the reduction in negative correlation under both scenarios. Under both the SSP2-4.5 (Figure 6f) and SSP5-8.5 (Figure 6l) scenarios, the NCE similarly simulated a weaker anomalous northerly wind from the Arabian Sea to northern Xinjiang, which is consistent with our expectations.

4. Conclusions

Based on CMIP6 model data and reanalysis data, two multi-model ensemble means—BMME and NCE—were derived from 30 models to simulate the August APO and the influence of the August APO on September precipitation over northern Xinjiang. The variations in APO intensity under the historical, SSP2-4.5, and SSP5-8.5 scenarios were analyzed using BMME. The correlations between the APOI and NXPI under the SSP2-4.5 and SSP5-8.5 scenarios were projected using NCE, and the possible causes of the changes were examined from the perspective of atmospheric circulation.

All models successfully reproduce the horizontal distribution of the August eddy temperature. The BMME derived from five models (AWI-CM-1-1-MR, EC-Earth3-Veg-LR, GFDL-ESM4, EC-Earth3-Veg, and EC-Earth3-CC) shows better performance in simulating the August eddy temperature than any individual model, with CC = 0.97, Ratio = 1.04, and RMSE = 0.25. Under both SSP2-4.5 and SSP5-8.5 scenarios, the APO intensity calculated by BMME exhibits a general decreasing trend from 2015 to 2100.

However, most models do not perform well in simulating September precipitation over northern Xinjiang. Subsequently, we calculated the correlation between APOI and NXPI, and most models exhibited a reasonably good negative correlation. Based on the three models that passed the 90% significance test (BCC-CSM2-MR = −0.27, MIROC6 = −0.27, and MPI-ESM1-2-HR = −0.38), the NCE (r = −0.31) was constructed to investigate the influence of the August APO on September precipitation over northern Xinjiang. In the historical period, the regression of precipitation onto the APOI based on NCE reproduces a negative correlation pattern similar to observations, performing better than AMME. Moreover, NCE more accurately simulates the northeasterly wind anomalies over northern Xinjiang at 850-hPa. Under the SSP2-4.5 and SSP5-8.5 scenarios, the negative correlation between APOI and NXPI remains stable during 2021–2040, with correlation coefficients of −0.33 under SSP2-4.5 and −0.43 under SSP5-8.5. Over the following three 20-year periods, the correlation weakens to −0.19, 0.27, and 0.19 under SSP2-4.5, and to −0.13, −0.14, and −0.26 under SSP5-8.5. This may be related to the southeastward shift of the negative geopotential height anomaly center over East Asia.