Time-Lag Effects of Winter Arctic Sea Ice on Subsequent Spring Precipitation Variability over China and Its Possible Mechanisms

Abstract

Arctic sea ice variations exhibit relatively strong statistical associations with precipitation variability over northeastern and southern China. Using Arctic Ocean reanalysis data from the EU Copernicus Project, this study examines the time-lagged statistical relationships between winter Arctic sea ice conditions and subsequent spring precipitation variability over China through wavelet analysis and Granger causality tests. Singular value decomposition (SVD) identifies the Barents, Kara, East Siberian, and Chukchi Seas as key regions exhibiting strong associations with spring precipitation anomalies. Increased winter sea ice in the East Siberian and Chukchi Seas generates positive geopotential height anomalies over the Arctic and negative anomalies over Northeast Asia, adjusting upper-level jet streams and influencing precipitation patterns in Northeast China. Conversely, increased sea ice in the Barents–Kara Seas leads to persistent negative geopotential height anomalies simultaneously occurring over both the Arctic and South China regions, enhancing southern jet stream activity and intensifying warm-moist airflow at the 850 hPa level, thus favoring precipitation in southern China. Compared to considering only climate factors such as the Pacific Decadal Oscillation (PDO), El Niño–Southern Oscillation (ENSO), and Arctic Oscillation (AO), the inclusion of Arctic sea ice significantly enhances the influence of multiple climate factors on precipitation variability in China.

1. Introduction

Arctic sea ice is a “regulator” of the Arctic climate and an “indicator” of global change [1]. With the intensification of global warming, the Arctic region has experienced the most significant temperature rise in the world, exceeding twice the global average, which is also known as “Arctic amplification” [2,3]. From the late 20th century to the 21st century, the coverage, thickness, and surface snow depth of Arctic sea ice have decreased, and the speed of sea ice’s movement and the sea temperature have increased, with longer melting seasons [4,5,6]. These trends are expected to continue, and seasonal ice-free conditions in the Arctic Ocean could occur by the middle of this century or even earlier [7].

Variations in Arctic sea ice are linked to changes in China’s climate and extreme weather events, influencing temperature [8] and precipitation patterns in various regions. There have been extensive studies on the correlation between Arctic sea ice and precipitation in Northeast China. Guo et al. [9] found that the total Arctic sea ice extent in spring (February–April) is positively correlated with summer precipitation in the Yangtze River Basin but negatively correlated with summer precipitation in Northeast China, the Indochina Peninsula, and the South China Sea. Zhang et al. [10] identified a significant negative correlation between spring precipitation in Northeast China and the preceding winter’s Barents Sea ice concentration. Similarly, Li et al. [11] reported a negative correlation between the intensity of Barents Sea ice and precipitation in Northeast China during May. Furthermore, Han et al. [12], through observational analysis and atmospheric simulation experiments, found that an increase in Barents Sea ice in early spring leads to increased rainfall in northern Northeast China while causing an abnormal decrease in precipitation in the southern part of the region.

Changes in Arctic sea ice also influence precipitation in southern China. Zhao et al. [13] found that a reduction in spring sea ice extent in the Bering Sea and the Sea of Okhotsk leads to an increase in summer monsoonal precipitation in southeastern China. Similarly, Wu et al. [14] demonstrated that a decline in the spring sea ice concentration in the Arctic Ocean and Greenland Sea is associated with a reduction in summer precipitation in southern China. Additionally, Wu et al. [15] discovered that the variability of Arctic sea ice in the Norwegian Sea and the Barents Sea during the preceding winter is closely linked to the interannual variation in East Asian spring precipitation, showing a significant negative correlation.

Arctic sea ice variability is closely linked to atmospheric circulation changes over the Arctic, with the Arctic Oscillation (AO) and North Atlantic Oscillation (NAO) acting as key intermediaries connecting sea ice conditions to the Northern Hemisphere’s climate patterns [16]. He et al. [17] found that the decline in Arctic sea ice in June triggers an anomalous Rossby wave train, which propagates downstream and impacts the circulation over East Asia. This leads to the formation of a tripole precipitation pattern in August, characterized by negative precipitation anomalies in northern and southern China, while Japan, Korea, and central China experience positive anomalies. Wu et al. [15] demonstrated that Arctic sea ice anomalies persist from winter into the following spring, continuously generating Rossby wave trains across the Eurasian continent and inducing downstream circulation anomalies. This wave train pattern is often associated with an anomalous low-pressure system over the Mongolian Plateau, which enhances the East Asian subtropical westerly jet stream. The strengthening of this jet stream, accompanied by low-level convergence and upper-level divergence, enhances local convection, thereby favoring increased spring precipitation in East Asia. Physical analyses based on observations and numerical simulations [11] indicate that a decrease in the Barents Sea ice concentration may trigger a Rossby wave train that propagates southeastward, crossing the Mongolia–Baikal region and reaching Northeast China. This process establishes a zonal anticyclone–cyclone–anticyclone teleconnection pattern over the mid-to-high latitudes of Eurasia. Under the influence of this pattern, southerly winds prevail over Northeast China, transporting warm and moist air into the region. Additionally, increased local atmospheric instability and upward motion further enhance favorable moisture and dynamic conditions, leading to increased precipitation in Northeast China during May.

Although previous studies have explored the potential link between Arctic sea ice melting and East Asian precipitation, their explanations regarding the key regions and underlying mechanisms remain inconsistent. The objective of this study is to conduct a comprehensive analysis of the spatiotemporal characteristics of Arctic sea ice variations based on the sea ice reanalysis dataset from the European Union’s Copernicus Program. By examining different sea ice regions, this study explores the spatial and temporal relationships between Arctic sea ice anomalies and the precipitation variability over northeastern and southern China. Furthermore, from the perspective of atmospheric circulation, we investigate the possible mechanisms through which Arctic sea ice anomalies are statistically linked to spring precipitation patterns in these selected regions. The remainder of this paper is organized as follows: Section 2 introduces the datasets and methods used in this study. Section 3 presents the results of the spatiotemporal characteristics of Arctic sea ice and its statistical associations with precipitation variability. Section 4 discusses the possible physical mechanisms and further analyzes the influences of multiple climate factors, including ENSO, PDO, AO, and Arctic sea ice, on precipitation variability in China. Section 5 concludes the paper with a summary of the key findings and suggestions for future research.

2. Datasets and Methods

2.1. List of Abbreviations

Throughout this paper, a number of abbreviations and acronyms are employed. To enhance clarity and facilitate comprehension, the definitions of key terms and their corresponding abbreviations are presented in

Table 1. List of abbreviations used in this paper.

Term	Definition
ENSO	El Niño-Southern Oscillation
PDO	Pacific Decadal Oscillation
AO	Arctic Oscillation
SIC	Sea Ice Concentration
SIT	Sea Ice Thickness
EOF	Empirical Orthogonal Function
SVD	Singular Value Decomposition
MWC	Multiple Wavelet Coherence
AWC	Average Wavelet Coherence
PASC	Percentage of Area with Significant Coherence
Z200	200 hPa Geopotential Height
Z500	500 hPa Geopotential Height
SLP	Sea Level Pressure
U200	200 hPa Zonal Wind
UV850	850 hPa Horizontal Wind

. This table encompasses all relevant abbreviations that are frequently referenced throughout the study.

2.2. Study Area

The study area is confined to the ocean above the latitude of 66.5° N on the Atlantic side, with Fury and Hecla Strait as the southern boundary of the Canadian Archipelago. We divided the area into the following 5 sub-regions, as shown in Figure 1: Barents–Kara Seas, Arctic Marginal Seas (Beaufort–Chukchi–East Siberian–Laptev Seas), Baffin–Archipelago, Central Arctic, and Greenland–Norwegian Seas.

2.3. Datasets

The Arctic Ocean Physics Reanalysis dataset “ARCTIC_MULTIYEAR_PHY_002_003” is a reanalysis product provided by the Arctic Monitoring and Forecasting Centre (ARC MFC), operated by the Nansen Environmental and Remote Sensing Center (NERSC) in Bergen, Norway, through the TOPAZ4b system, and is available from the Copernicus Marine Service (http://marine.copernicus.eu/) (accessed on 21 March 2024). It employs advanced sequential data assimilation methods and mixed coordinate ocean models, integrating remote sensing data for enhanced accuracy and comprehensiveness. Observations that are assimilated by TOPAZ4 include sea surface temperature from the Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA), sea ice concentration from OSI-SAF, and CS2SMOS sea-ice thickness data from AWI. The ice thickness product is the weekly merged CS2SMOS dataset from AWI [18], combining thin sea ice measurements from the ESA SMOS mission to the thick sea ice retrievals from another ESA mission CryoSAT2. The spatial resolution of the reanalysis datasets is 12.5 km, and the temporal resolution is monthly.

The precipitation dataset used in this study is the CHM_PRE dataset provided by the National Tibetan Plateau Data Center [19], operated in Beijing, China, and is available at https://data.tpdc.ac.cn/ (accessed on 11 December 2024). This dataset is constructed based on daily observational data collected from 2839 stations, including 2419 stations across mainland China and 420 stations in surrounding regions. The dataset covers the period from 1961 to 2022, with a spatial resolution of 0.1° and a geographical range of 18° N–54° N and 72° E–136° E.

The atmospheric circulation factors used in this study are derived from ERA5 (https://cds.climate.copernicus.eu/), a widely used atmospheric reanalysis dataset produced by the European Centre for Medium-Range Weather Forecasts (ECMWF), located in Reading, United Kingdom. ERA5 is generated using 4D-Var data assimilation and model forecasts from cycle CY41R2 of the ECMWF Integrated Forecast System (IFS) (accessed on 2 December 2024) [20]. The ERA5 dataset contains more than 200 parameters and variables with global coverage, spanning from 1940 to the present date in near real time. ERA5 provides a large number of atmospheric, land, and oceanic climate variables. In this study, we selected the 200 hPa geopotential height (Z200), 500 hPa geopotential height (Z500), sea level pressure (SLP), 200 hPa zonal wind (U200), and 850 hPa horizontal wind (UV850).

The Niño 3.4 index, PDO index, and AO index were obtained from the Earth System Research Laboratory of the National Oceanic and Atmospheric Administration (NOAA), located in Boulder, Colorado, United States, and available at http://www.esrl.noaa.gov/ (accessed on 2 December 2024).

2.4. Methods

This study adopts a multi-method, multi-scale, and integrated analytical framework to reveal the spatiotemporal coupling mechanisms and dominant variability patterns between precipitation in China and Arctic sea ice. First, empirical orthogonal function (EOF) analysis is employed to reduce the dimensionality of high-dimensional spatiotemporal precipitation data, extracting the dominant variability structures and their evolution patterns within the region. Secondly, singular value decomposition (SVD) is used to construct a cross-covariance matrix between the precipitation and Arctic sea ice fields, extracting paired spatial modes with the highest covariance to analyze their teleconnection and co-evolution patterns. For dynamical mechanisms, both regression analysis and composite analysis are applied. Regression analysis quantifies the statistical relationships between precipitation and circulation fields, identifying key large-scale circulation anomalies. Composite analysis further verifies and attributes these atmospheric circulation and moisture transport patterns by comparing climate field characteristics during different sea-ice anomaly years, ensuring the robustness of the identified mechanisms. Additionally, the T–N three-dimensional wave-activity flux theory is introduced to quantitatively diagnose the energy transmission process of quasi-stationary Rossby waves, exploring the physical link between precipitation anomalies and Arctic sea ice variations. Furthermore, bivariate wavelet coherence (WTC) and partial wavelet coherence (PWC) analyses are used to assess the influence of multiple climate factors on precipitation in China across different timescales. Statistical significance is determined using the Student’s t-test, with the degrees of freedom set to n − 2, where n represents the number of rows in the variable dataset. The technical roadmap of this study is shown in Figure 2.

2.4.1. Empirical Orthogonal Function (EOF)

EOF is a dimensionality reduction method for climate fields based on the orthogonal decomposition of the covariance matrix [21]. Its core idea is to decouple high-dimensional spatiotemporal datasets into a linear combination of spatial modes and temporal coefficients, thereby extracting dominant spatiotemporal variability signals. Given the observational data of a climate variable field in the matrix form:

$$\begin{array}{c}X=\left[\begin{array}{ccc}{x}_{11}& \dots & x_{1n}\\ ⋮& \ddots & ⋮\\ x_{m1}& \dots & x_{mn}\end{array}\right]\end{array}$$

(1)

where m represents spatial points, typically observation stations or grid points, and n denotes temporal points, i.e., the number of observations; x_mn represents the observation value at the m-th station or grid point during the n-th observation. The EOF decomposition expresses X as:

$$\begin{array}{c}X=VZ\#\end{array}$$

(2)

where

$$\begin{array}{c}V=\left[\begin{array}{ccc}{v}_{11}& \dots & v_{1m}\\ ⋮& \ddots & ⋮\\ v_{m1}& \dots & v_{mm}\end{array}\right],Z=\left[\begin{array}{ccc}{z}_{11}& \dots & z_{1n}\\ ⋮& \ddots & ⋮\\ z_{m1}& \dots & z_{mn}\end{array}\right]\end{array}$$

(3)

These are referred to as the spatial function matrix and the temporal coefficient matrix, respectively. Given the orthogonality properties of V and Z, the spatial function matrix, V, is obtained from the eigenvectors of $X{X}^T$, while the temporal coefficient matrix, Z, is derived as $V^{T}X$. The variance contribution of each eigenvector is calculated as follows:

$$\begin{array}{c}{R}_k={\displaystyle \frac{{\lambda }_k}{\sum _{i=1}^m {\lambda }_i}}\left(k=\mathrm{1,2},\dots ,p\left(p<m\right)\right)\end{array}$$

(4)

where ${\lambda }_k$ is the k-th eigenvalue. The cumulative variance contribution of the first p eigenvectors is given by the following:

$$\begin{array}{c}G={\displaystyle \frac{\sum _{i=1}^p {\lambda }_i}{\sum _{i=1}^m {\lambda }_i\left(p<m\right)}}\end{array}$$

(5)

The leading eigenvectors obtained from EOF decomposition represent the dominant variability structure of the climate variable field across the entire region. Given the advantages of EOF in extracting the primary variability patterns of climate fields, this study applies it to analyze the spatiotemporal evolution of precipitation in China. The first three modes are used to reveal the dominant spatial patterns of precipitation variability and their temporal evolution.

2.4.2. Singular Value Decomposition (SVD)

SVD [22] is an orthogonal decomposition method based on the cross-covariance matrix of two variable fields. By applying SVD, the original high-dimensional spatiotemporal correlation problem can be transformed into a limited number of time-varying coupling modes, effectively extracting the coupled coherent patterns between two meteorological fields. Mathematically, SVD performs spectral decomposition on the cross-covariance matrix to isolate paired spatial modes with the highest covariance, thereby revealing the spatiotemporal teleconnection characteristics between the two fields. In this study, SVD is used to analyze the cross-regional teleconnection mechanism between precipitation in China and Arctic sea ice, focusing on the first two modes with significant covariance contributions to uncover key co-evolution patterns.

2.4.3. T–N Three-Dimensional Wave-Activity Flux

Takaya and Nakamura [23], building upon Plumb’s theoretical foundation, developed the formulation for the three-dimensional propagation of quasi-stationary waves, known as the T–N flux (also referred to as wave-activity flux), under conditions where the basic flow is variable. This formulation allows for the diagnosis of Rossby wave perturbation energy propagation and the significance of nonlinear interactions, providing a means to describe the energy propagation characteristics of quasi-stationary Rossby waves. Under the Wentzel–Kramers–Brillouin (WKB) approximation, this flux is independent of the wave phase and aligns with the local group velocity of stationary Rossby wave trains in the horizontal direction. In the logarithmic pressure coordinate system, it can be expressed as follows:

$$\begin{array}{c}W={\displaystyle \frac{\mathrm{pcos}\phi }{2\left|\mathit{U}\right|}}\left(\begin{array}{c}{\displaystyle \frac{U}{a^2{\cos }^2\phi }}\left[{\left({\displaystyle \frac{\partial {\psi }^{\prime }}{\partial \lambda }}\right)}^2-{\psi }^{\prime }{\displaystyle \frac{{\partial }^2{\psi }^{\prime }}{\partial {\lambda }^2}}\right]+{\displaystyle \frac{V}{a^2\cos \phi }}\left[{\displaystyle \frac{\partial {\psi }^{\prime }}{\partial \lambda }}{\displaystyle \frac{\partial {\psi }^{\prime }}{\partial \phi }}-{\psi }^{\prime }{\displaystyle \frac{{\partial }^2{\psi }^{\prime }}{\partial \lambda \partial \phi }}\right]\\ {\displaystyle \frac{U}{a^2\cos \phi }}\left[{\displaystyle \frac{\partial {\psi }^{\prime }}{\partial \lambda }}{\displaystyle \frac{\partial {\psi }^{\prime }}{\partial \phi }}-{\psi }^{\prime }{\displaystyle \frac{{\partial }^2{\psi }^{\prime }}{\partial \lambda \partial \phi }}\right]+{\displaystyle \frac{V}{a^2}}\left[{\left({\displaystyle \frac{\partial {\psi }^{\prime }}{\partial \phi }}\right)}^2-{\psi }^{\prime }{\displaystyle \frac{{\partial }^2{\psi }^{\prime }}{\partial {\phi }^2}}\right]\\ {\displaystyle \frac{f_0^2}{N^2}}\left\{{\displaystyle \frac{U}{\mathrm{acos}\phi }}\left[{\displaystyle \frac{\partial {\psi }^{\prime }}{\partial \lambda }}{\displaystyle \frac{\partial {\psi }^{\prime }}{\partial z}}-{\psi }^{\prime }{\displaystyle \frac{{\partial }^2{\psi }^{\prime }}{\partial \lambda \partial z}}\right]+{\displaystyle \frac{V}{a}}\left[{\displaystyle \frac{\partial {\psi }^{\prime }}{\partial \phi }}{\displaystyle \frac{\partial {\psi }^{\prime }}{\partial z}}-{\psi }^{\prime }{\displaystyle \frac{{\partial }^2{\psi }^{\prime }}{\partial \phi \partial z}}\right]\right\}\end{array}\right)\end{array}$$

(6)

In Equation (6), p denotes the pressure normalized by 1000 hPa; φ and λ represent the latitude and longitude, respectively; f is the Coriolis parameter; a is the Earth’s radius; U and V are the zonal and meridional components of the basic-state flow field; ψ′ is the quasi-geostrophic perturbation stream function; and N² represents the buoyancy frequency (Brunt–Väisälä frequency). This formulation couples dynamical and thermodynamic perturbation terms to describe the three-dimensional propagation of Rossby wave energy, where the horizontal component reflects barotropic energy transmission. In this study, this method is employed to analyze the dynamical linkage between Arctic sea ice anomalies and precipitation variability in China.

2.4.4. Multi-Wavelet Coherence (MWC)

MWC is an effective method for investigating the scale-dependent relationships between two variables [24]. MWC can unravel a series of multivariate relationships, determine the multivariate relationships of variables in space or time scales, and also identify the proportion of variance related to predictive variables and response variables [25]. In wavelet analysis, a 95% confidence level is obtained based on the first-order autocorrelation coefficient through Monte Carlo simulations (1000 repetitions) [26]. The resonance period refers to a strong coupling relationship that occurs on a specific time scale. The quantitative assessment of the predictive variables’ explanatory power for the response variable is conducted by calculating the percentage area of significant coherence outside the cone of influence (percent area of significant coherence, PASC), as well as the average wavelet coherence (AWC) [24,27]. The AWC was calculated by averaging the wavelet coherence produced over all scales according to the coherence values produced in the PWC computation [26]. The PASC was obtained by calculating the ratio of the number of significant values of power over the total number of values of the power. The larger the PASC, the higher the AWC, indicating that the specific predictor variable can explain more of the sea ice variation.

2.4.5. Mann–Kendall (MK) Trend Test

Mann–Kendall (MK) trend test, which is a non-parametric test method recommended by the World Meteorological Organization [28], is used to test the temporal changes in sea ice. The specific principles of the MK are as follows:

Assuming that the samples in the sequence {x₁,…,x_N} are randomly independent, the statistic S is constructed as follows:

$$\begin{array}{c}S={\displaystyle \sum _{i=1}^{N-1}} {\displaystyle \sum _{j=i+1}^N} sgn\left(x_j-x_i\right)\end{array}$$

(7)

among

$$\begin{array}{c}sgn\left(x_j-x_i\right)=\left\{\begin{array}{cc}+1,& x_j-x_i>0\\ 0,& x_j-x_i=0\\ -1,& x_j-x_i<0\end{array}\right.\end{array}$$

(8)

where S > 0 indicates that the sequence has an increasing trend, and S < 0 indicates that the sequence has a decreasing trend.

The statistic τ measures the strength of a continuous trend, as follows:

$$\begin{array}{c}\tau ={\displaystyle \frac{2S}{N\left(N-1\right)}}\end{array}$$

(9)

Mann and Kendall proved that when n ≥ 8, the statistic S roughly follows a normal distribution, and its variance is as follows:

$$\begin{array}{c}{\sigma }_s^2={\displaystyle \frac{1}{18}}\left[N\left(N-1\right)\left(2N+5\right)-\sum _{i=1}^m t_i\left(t_i-1\right)\left(2t_i+5\right)\right]\end{array}$$

(10)

where m is the number of groups, Var(S) is equal to x in the dataset, and $t_i$ is the number of data points in group i equal group.

The hypothesis test Z is defined below and approximates the standard normal distribution:

$$\begin{array}{c}z=\left\{\begin{array}{c}(S-1)/{\sigma }_s \mathrm{i}\mathrm{f} S>0\\ 0 \mathrm{i}\mathrm{f} S=0\\ (S+1)/{\sigma }_s \mathrm{i}\mathrm{f} S<0\end{array}\right.\end{array}$$

(11)

When Z > 0, there is an upward trend; when Z < 0, there is a decreasing trend.

For a given significance level α, if$\left|\mathrm{Z}\right| > {\mathrm{Z}}_{1-\mathsf{\alpha }/2}$, this indicates a significant upward or downward trend in the time series. In this paper, the p-value is used to represent Z. When the p-value is less than a pre-defined significance level (typically 0.05), the data are considered to exhibit a significant trend. In this study, the p-values are categorized into three levels, as follows: a trend with a p-value between 0.05 and 0.1 is classified as slightly significant (indicated by +1 or −1 for positive or negative trends, respectively); a trend with a p-value between 0.01 and 0.05 is considered moderately significant (indicated by +2 or −2); and a trend with a p-value less than 0.01 is regarded as highly significant (indicated by +3 or −3).

Computational resources: All data processing and statistical analyses in this study were conducted using MATLAB R2022a and Python 3.9 on a personal workstation equipped with an AMD Ryzen 7 6800H CPU (8 cores, 3.20 GHz), 16 GB RAM, and a 477 GB SSD. The datasets processed in this study (including precipitation observations, sea ice reanalysis, and atmospheric circulation data) totaled approximately 35 GB. The computational requirements for wavelet analysis, singular value decomposition (SVD), and Granger causality testing were moderate, and all analytical tasks could be completed within several hours. The available computational resources were sufficient to ensure efficient data processing and analysis without compromising the accuracy and reliability of the results.

3. Results

3.1. Spatiotemporal Variations in Arctic Sea Ice

Sea ice concentration (SIC) refers to the percentage of sea ice area in each grid, the sea ice extent (SIE) refers to the total area of all grid cells where the SIC is not less than a certain threshold, usually 15%. In this study, a grid cell is considered ice-free if its SIC is less than 15%, following the threshold commonly used in previous studies by Parkinson and Cavalieri [29,30]. The spatial resolution of the grid is 12.5 × 12.5 km. The average SIC within the study area is calculated using a weighted method. Overall, SIC reached its annual maximum in March and minimum in September. In March, the Central Arctic, Arctic Marginal Seas, and Baffin–Archipelago were mostly covered by sea ice (Figure 3a). In September, the SIC was the highest in the Central Arctic, and decreased toward the marginal areas of the Arctic Ocean, with almost no sea ice coverage in the Barents–Kara Seas and Greenland–Norwegian Seas (Figure 3b).

The analysis of the annual SIC variations in Arctic sub-regions (Figure 4a) reveals a significant Arctic sea ice decline. The most pronounced ice loss occurred in the Barents Sea, Chukchi Sea, Beaufort Sea, Labrador Sea, and Greenland Sea, consistent with findings from previous studies [4,31]. Regional decline rates exhibit spatial heterogeneity, as follows: the Central Arctic (0.0017%/yr), Arctic marginal seas (0.0047%/yr), and Barents–Kara Sea (0.0055%/yr) form a distinct descending gradient in annual ice loss. The grid-scale Mann–Kendall trend tests (Figure 4b) confirm significant seasonal disparities in ice retreat. Summer–autumn (June–November) SIC fluctuations intensified by 42% compared to winter–spring (December–May). The record minimum SIC occurred in September 2012 due to extreme storm activity [32], with secondary lows observed in September 2016 and 2020.

Spatiotemporal analysis of sea ice thickness (SIT) (Figure 5) demonstrates that the “perennial multiyear ice zone” along the Canada–Greenland coast maintains maximum thickness, while the Barents–Kara Seas and Greenland–Norwegian Seas, influenced by the warm North Atlantic Current, remain persistently thin. The pan-Arctic annual mean SIT decreased by 0.13 m overall (1991–2023), with the Arctic marginal seas exhibiting the fastest thinning rate (0.0238 m/yr), followed by the Baffin–Archipelago region (0.0096 m/yr). Notably, localized thickening occurred in northern Greenland Sea and the Arctic Basin–Greenland Sea junction (Figure 5b). SIT seasonal cycles exhibited phase delay characteristics relative to SIC, peaking in May and reaching the minima in October. Although thinning trends persist across all seasons, winter–spring periods show relative stability, while summer–autumn (June–November) account for 76% of annual thickness loss.

The seasonal cycle of SIT lagged behind that of SIC, with a maximum value in May and a minimum value in October. SIT showed a general declining trend across all four seasons. However, it remained relatively stable during winter and spring, while experiencing significant declines in summer and autumn.

3.2. Spatiotemporal Variations in Precipitation in China

Figure 6 presents the spatial distribution of total precipitation and the interannual variation trends in China from 1991 to 2022. The precipitation trend is obtained by performing linear regression between the grid-based precipitation data and time. The total precipitation distribution (Figure 6a) exhibited a typical monsoon climate pattern; over the past 32 years, the Yangtze River Basin and areas to the south have received an average annual precipitation exceeding 1500 mm, with some regions experiencing over 2000 mm due to orographic uplift. In contrast, the North China Plain and the Loess Plateau received less than 400 mm, while the arid inland regions of Northwest China recorded even lower values, below 250 mm. The interannual variation trend (Figure 6b) highlights the significant wetting in the Northeast China Plain, Sichuan Basin, Hainan Island, and areas surrounding the Yangtze River Delta. Conversely, drying trends were observed in the eastern Yunnan–Guizhou Plateau and along the Fujian coast.

Figure 7 presents the EOF analysis results of the precipitation in China from 1991 to 2022, providing insight into its spatial distribution and temporal variations. The percentage of variance explained by each mode is indicated in the upper right corner of each subplot. EOF1 (Figure 6a) explains 24.2% of the total variance, representing the primary mode of precipitation variability, particularly the north–south contrast. The spatial pattern exhibits a “− + − +” distribution from north to south, with positive anomalies centered in northern Northeast China and the region south of the Yangtze River, while negative anomalies extend across the Xinjiang–Inner Mongolia–Jilin belt and the Tibetan Plateau to the Loess Plateau. This mode reflects the typical out-of-phase relationship between northern and southern China’s precipitation. The corresponding PC1 time series (Figure 7b) shows complex fluctuations, with multiple alternating positive and negative values. EOF2 (Figure 7c) explains 17.4% of the variance and exhibits a uniform negative phase across most of China, indicating synchronous precipitation variations (either increasing or decreasing). However, Hainan Island and its surrounding coastal areas display a significant positive anomaly, forming a local inverse-phase relationship. EOF3 (Figure 7e) accounts for 9.7% of the variance, highlighting a positive phase over the Yunnan–Guizhou Plateau and southern Tibet, while Northeast China exhibits a distinct negative phase. The corresponding PC3 time series (Figure 7f) remains mostly in a positive phase from 1991 to 2010 but undergoes a phase shift to negative after 2010.

3.3. Time-Lag Effects of Arctic Sea Ice on Precipitation in China and Key Sea Area

To identify the key time periods with strong correlation between Arctic sea ice and precipitation in China, this study employed multiple statistical methods, including wavelet analysis, lagged correlation analysis, and Granger causality testing, for cross-validation. Given the seasonal fluctuations in the monthly data, all variables were deseasonalized before analysis. In Figure 8, cross-wavelet transform analysis is used to examine the phase relationship and significance of the two variables across different timescales. The bold contours indicate the 95% confidence level against yellow noise. The scale is measured in months, and the shaded region represents the cone of influence (COI), where edge effects may distort the results. Rightward arrows indicate a positive phase relationship, while leftward arrows indicate a negative phase relationship. The results (Figure 8) indicate a significant correlation between sea ice and precipitation within the 4–20-month timescale, with an estimated 1–4-month lag of precipitation behind SIC based on the phase relationships. Lagged correlation analysis and Granger causality testing show peak values when precipitation lags behind sea ice by one month, with correlations gradually weakening in the subsequent months. These statistical analyses suggest that the impact of sea ice on precipitation is most pronounced on a seasonal timescale. Therefore, further correlation analysis is conducted between spring sea ice and summer precipitation, summer sea ice and autumn precipitation, autumn sea ice and winter precipitation, and winter sea ice and spring precipitation of the following year. The results indicate that winter sea ice has the strongest correlation with spring precipitation in the following year. Thus, the subsequent analyses primarily focus on these two seasons.

To further identify the key Arctic sea ice regions influencing precipitation distribution in China, this study applies SVD. For clarity, we define September–November (SON), December–February (DJF), March–May (MAM), and June–August (JJA) as autumn, winter, spring, and summer, respectively. The study analyzes the co-variability patterns between winter Arctic SIC anomalies and spring precipitation in China (CHM) from 1991 to 2022. Heterogeneous correlation coefficients are calculated to characterize the spatial relationships between the two fields. The key response regions across fields are identified by computing the correlation between raw data and the opposite-field mode time series, with dotted areas indicating statistically significant regions.

The first SVD mode (n = 0) (Figure 9a,b) and the second SVD mode (n = 1) (Figure 9c,d) explain 31.9% and 18.8% of the covariance, respectively, collectively accounting for 50.6% of the total variance. This suggests that the primary Arctic sea ice–China precipitation coupling relationships can be well represented by the first two modes, indicating a strong teleconnection signal between Arctic sea ice and East Asian precipitation. In the first mode, the Barents Sea, Kara Sea, and Bering Strait exhibit significant negative correlations in the left field (Figure 9a), corresponding to a spatially coherent negative correlation pattern in southern China in the right field (Figure 9b). The high mode correlation coefficients pass the significance test, indicating that when winter sea ice concentration in the Barents Sea, Kara Sea, and Bering Strait decreases (increases), spring precipitation in the lower-latitude regions of China also decreases (increases). In the second mode, the East Siberian Sea, Chukchi Sea, and parts of the Arctic Basin near the Bering Strait exhibit negative values in the left field, while northern Northeast China, parts of southern Tibet, and the Sichuan Basin show negative precipitation anomalies in the right field. This indicates that when winter sea ice concentration in the East Siberian Sea and Chukchi Sea decreases (increases), precipitation in northern Northeast China, southern Tibet, and the Sichuan Basin also decreases (increases). A linear regression analysis of the SVD mode time coefficients shows a significant downward trend in the first mode (slope: −0.0847/year), suggesting that the impact of Arctic sea ice anomalies on East Asian precipitation is gradually intensifying. The second mode also shows a decreasing trend (slope: −0.0464/year), though with a relatively smaller magnitude of change.

3.4. Response of Spring Atmospheric Circulation Anomalies to Winter Arctic Sea Ice Reduction

Previous studies have shown that sea ice melting significantly reduces surface albedo, leading to an increase in surface heat flux anomalies over the Arctic [33,34,35]. This diabatic heating effect causes substantial polar warming, weakens the meridional temperature gradient, and subsequently reduces the strength of the westerly jet stream while enhancing meridional atmospheric activity [36]. The thermal anomalies induced by sea ice loss can also trigger Rossby wave source anomalies, which propagate eastward toward East Asia in the form of wave trains, leading to significant adjustments in the mid-to high-latitude circulation system over Eurasia [37,38]. Meanwhile, Arctic warming decreases atmospheric static stability, promoting the development of baroclinic instability, which further amplifies circulation anomalies through enhanced dynamic instability [39].

To further analyze whether the impact of winter Arctic sea ice on spring precipitation in China aligns with the mechanisms mentioned above, key regions with high correlations in Figure 9a,c, namely, the Barents–Kara Seas and Arctic Marginal Seas—are identified as critical areas where sea ice influences precipitation. The area-weighted SIC indices for these key regions are defined as BSIC and ASIC, respectively. When investigating the possible mechanisms of sea ice’s influence on precipitation, it is essential to examine multiple atmospheric levels and key circulation factors. In this study, the selected circulation indices include: Z200 (200 hPa geopotential height) and Z500 (500 hPa geopotential height) to represent the upper and mid-tropospheric geopotential height distributions, which help identify wave train structures and ridge-trough locations; SLP (sea level pressure) to reflect the intensity and spatial distribution of cyclonic and anticyclonic activities near the surface; U200 (200 hPa zonal wind) to determine the position and strength of the upper-level westerly jet stream; UV850 (850 hPa horizontal wind) to analyze low-level moisture transport pathways and wind convergence/divergence characteristics.

Regression analysis is performed between ASIC and the selected circulation indices. Additionally, ASIC is ranked, and high-value years and low-value years are identified for composite difference analysis using the same circulation indices. For ASIC, the high-value years are 1995, 1999, 2000, 1991, and 2003, while the low-value years are 2008, 2018, 2020, 2017, and 2007. The results show that the spatial patterns and signs in the key regions (such as the mid-to-high latitudes and East Asia) are consistent between the regression analysis maps and composite analysis maps, indicating that the detected signals are stable and reliable.

The left panel of Figure 10 illustrates that ASIC sea ice anomalies primarily influence precipitation in northern China, particularly in Northeast China, through a “Polar–Siberia–Northeast Asia” wave train propagation pattern. At the 200 hPa geopotential height level (Figure 10a), when winter ASIC is high, a significant positive geopotential height anomaly appears over the Arctic, extending toward the high latitudes of Eurasia, with an opposite-sign anomaly center over Northeast Asia. This pattern resembles a wave train (Rossby wave) propagating from the polar region to mid-latitudes. The position and intensity of the upper-level jet stream over the North Pacific and Northeast Asia also change, leading to north–south jet stream oscillations, which in turn influence frontal activity and interactions between warm and cold air masses over East Asia. The composite analysis (Figure 10b) aligns well with the regression results, showing that in high-ASIC years, positive height anomalies are observed near the Arctic and northern Eurasia, while negative anomalies appear over the North Pacific and parts of East Asia. The opposite pattern occurs in low-ASIC years. At the 500 hPa level (Figure 10c), a series of alternating positive and negative anomaly centers extends from the polar region to East Asia, reflecting mid-tropospheric wave train structures that further regulate convective conditions in Northeast China, influencing regional precipitation processes. Meanwhile, SLP anomaly distributions (Figure 10e) suggest that increased sea ice stabilizes the polar low-pressure system. The regression and composite analysis of U200 (Figure 10g,h) indicate that when winter ASIC is high, the upper-level westerlies in this region are significantly strengthened, potentially enhancing upper-level divergence. This, in turn, promotes lower-level convergence and ascending motion, creating favorable conditions for moisture convergence and precipitation.

A similar analysis is conducted for BSIC, including regression analysis with circulation indices and ranking BSIC values to identify high-value and low-value years for composite difference analysis. The high-value years for BSIC are 1994, 1998, 1999, 2003, and 2004, while the low-value years are 2012, 2013, 2016, 2017, and 2018. The 200 hPa geopotential height regression map (Figure 11a) shows that when winter Barents–Kara Seas ice increases (BSIC is high), negative geopotential height anomalies tend to appear over high-latitude regions (Arctic–North Atlantic) and mid- to high-latitude Eurasia, forming a wave train pattern propagating from the polar region to Eurasia. At the same time, the jet stream near South China strengthens, improving upper-level divergence conditions, which favors convection initiation or intensified frontal activity. At the 500 hPa level (Figure 11c,d), a wave train signal can also be observed extending from the Arctic–Eurasia region to South Asia. When the BSIC increases, negative height anomalies appear over the Barents Sea and parts of Siberia, while a negative anomaly center forms near South China. The U200 regression and composite maps (Figure 11g,h) show positive anomalies over South China, indicating that the negative geopotential height anomalies are coupled with the subtropical jet stream, which intensifies over South China, further enhancing upper-level divergence and atmospheric instability. The 850 hPa wind field maps (Figure 11i,j) reveal a clear northward transport of warm and moist air over South China, providing ample moisture supply for local convective activity, thereby facilitating convection development and precipitation formation.

Overall, an increase in sea ice induces multi-level circulation adjustments from the upper atmosphere (Z200, Z500, and U200) to the near-surface (SLP, UV850), generating different teleconnection wave trains. The ASIC anomalies primarily led to positive geopotential height anomalies over the Arctic, which propagate toward Northeast Asia, forming negative geopotential height anomalies in this region. This process is accompanied by upper-level jet adjustments, influencing the interaction of cold and warm air masses and frontal activity in Northeast China, thereby regulating precipitation in northern regions. In contrast, the BSIC anomalies exhibit a continuous negative geopotential height anomaly pattern extending from the Barents Sea to South China, which is associated with jet stream intensification over South China and enhanced warm-moist air transport at the 850 hPa level, creating more favorable conditions for precipitation in southern regions. Both the ASIC and BSIC influence precipitation by altering large-scale circulation patterns, but their effects are observed in different regions.

To further verify the reasonableness of the mechanism analysis, this study conducts composite analysis of the 300 hPa stream function and the corresponding wave-activity flux in response to winter sea ice anomalies. This flux intuitively reflects the propagation path and intensity of quasi-stationary Rossby wave energy in the atmosphere. The distribution of the stream function in Figure 11 largely coincides with the anomaly centers in the regression and composite analyses. Both subplots in Figure 12 reveal a clear propagation path, illustrating a strong wave perturbation originating from the Barents–Kara Seas, which propagates southeastward, eventually reaching East Asia, forming a stable “Barents–Kara Seas–East Asia” teleconnection mode.

For ASIC (Figure 12a), winter sea ice anomalies generate significant positive geopotential height anomalies over the Arctic, forming an anticyclonic region. This anomaly triggers planetary wave energy perturbations, which extend from Siberia to Northeast Asia. As energy propagates downstream, the disturbance transitions into negative geopotential height anomalies, forming an upper-level trough, which adjusts the position and strength of the upper-level jet stream, modifies the interaction of cold and warm air masses, and influences frontal activity, ultimately regulating precipitation distribution and intensity in northern China. For BSIC (Figure 12b), winter sea ice anomalies over the Barents–Kara Seas generate a pronounced cyclonic negative geopotential height anomaly at high latitudes, which serves as a source region for planetary wave energy. The wave-activity flux propagates southeastward from Siberia to East Asia, while an additional propagation pathway extends from the North Pacific, traveling from high-latitude Pacific regions to southern China, thereby modulating precipitation in South China. Overall, the two T–N wave-activity flux diagrams provide a clear visualization of how Arctic sea ice anomalies remotely modulate East Asian atmospheric circulation by exciting and adjusting the planetary wave energy propagation paths. These mechanisms exert distinct influences on precipitation patterns in northern and southern China through different pathways.

4. Discussion

The interdecadal variability of precipitation in China is influenced by multiple factors, including El Niño-Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), Arctic Oscillation (AO), Antarctic Oscillation (AAO), and Arctic sea ice (SIC) [40,41,42]. In this section, the Niño 3.4 index is selected to represent the impact of ENSO, combined with the PDO index, AO index, and SIC for analysis. The multi-wavelet coherence (MWC) method is employed to reveal the scale-dependent effects and synergistic relationships of different climate factors on precipitation in China.

Figure 13 presents the results of the MWC analysis between precipitation in China and major climate factors at the monthly timescale. Specifically, Figure 13a illustrates the time–frequency relationships between precipitation and ENSO, PDO, and AO. Significant resonance regions are observed within the 4–16-month band, although they are relatively scattered and do not fully cover all time periods. After 2011, coherence gradually emerges in the 32–64-month mid-range scale, suggesting that the synergistic influence of ENSO, PDO, and AO on China’s precipitation may have strengthened in recent years. Overall, the average wavelet coherence between precipitation and these three climate factors is 0.54, with significant coherence regions accounting for 18.61% of the total area.

Based on this, Figure 13b further incorporates the influence of SIC. The results show a notable enhancement of coherence within the 4–20-month low- to mid-frequency ranges, covering a broader time span. The proportion of significant coherence area increases to 38.58%, and the average coherence value rises to 0.64. This indicates that the inclusion of Arctic sea ice variability further strengthens the synergistic effects among ENSO, PDO, AO, and precipitation, demonstrating a more pronounced modulation of precipitation variability across a wider range of time–frequency scales.

Previous studies, such as by Hussain et al. [43], showed that climate models, like ENSO, PDO, and AO, intermittently influence regional precipitation across multiple time scales. Consistent with their findings, our study reveals significant associations between China’s precipitation variability and major climate modes at both short (4–16 months) and long (32–64 months) periods. Nalley et al. [44] emphasized the importance of jointly considering multiple climate drivers to better explain hydroclimatic variability. In comparison, our study also takes into account Arctic sea ice variability (SIC), suggesting that the inclusion of polar processes may further enhance the coherence and extent of climate influences on regional precipitation patterns.

5. Conclusions

This study investigated the time-lagged statistical relationships between winter Arctic sea ice anomalies and spring precipitation variability in China, based on precipitation data from the National Tibetan Plateau Data Center and Arctic Ocean reanalysis data from the EU Copernicus Program for 1991–2022. Based on wavelet analysis and Granger causality testing, a strong correlation is found between winter sea ice variations and precipitation in the following spring. SVD identifies the Barents Sea, Kara Sea, East Siberian Sea, and Chukchi Sea as key regions influencing precipitation in China. A decrease in sea ice within these key areas is associated with reduced spring precipitation in coastal South China and parts of Northeast China.

The mechanism analysis indicates that winter sea ice changes in key Arctic regions influence turbulent heat fluxes, inducing atmospheric energy fluctuations that propagate toward East Asia through atmospheric wave trains. Increased sea ice in the East Siberian and Chukchi Seas generates positive geopotential height anomalies over the Arctic, accompanied by negative anomalies over Northeast Asia. This pattern adjusts the upper-level jet stream, influencing frontal activity and precipitation in northern China. Conversely, increased sea ice in the Barents–Kara Seas produces persistent negative geopotential height anomalies extending from the Barents Sea toward South China, strengthening jet stream activity over southern China and enhancing warm–moist air transport at 850 hPa, thereby promoting precipitation in southern China.

Further analysis based on multi-wavelet coherence reveals that the inclusion of Arctic sea ice variability further enhances the synergistic effects between precipitation and ENSO, PDO, and AO, exhibiting a more pronounced modulation of precipitation variability across a broader range of time–frequency scales. Although this study identifies significant statistical associations between winter Arctic sea ice anomalies and spring precipitation variability in China, several limitations should be acknowledged. First, the analyses relied on reanalysis and observational datasets, which may introduce inherent uncertainties. Second, the identified relationships are statistical in nature, and the underlying physical mechanisms remain complex and are not fully established. Therefore, future research combining high-resolution dynamical modeling and multi-source observational validation is needed to further explore and verify potential causal pathways linking Arctic sea ice changes to regional climate variability.