The dominant warming season shifted from winter to spring in the arid region of Northwest China

Abstract

The arid region of Northwest China (ARNC) has experienced a significantly higher warming rate than the global average and exhibits pronounced seasonal asymmetry, which has important implications for the region’s water-dependent systems. To understand the spatiotemporal patterns and driving mechanisms of seasonal asymmetric warming in the ARNC, we investigated seasonal changes in temperature rise and their underlying causes based on station and reanalysis data. We found that the dominant season of temperature increase shifted from winter to spring. The contribution of spring warming to the total temperature increase rose from −5%–7% to 58%–59%, while the contribution of winter warming decreased from 60%–75% to −4%–9%. However, the mechanisms underlying spring warming and winter cooling differ. An increase in solar radiation caused by a decrease in cloud cover (R = −0.64) was the main reason for spring warming, while a strengthening Siberian High primarily drove winter cooling.

Introduction

The Sixth Assessment Report of the Intergovernmental Panel on Climate Change¹ unequivocally states that human activities have caused global warming. However, the extent of global warming has not been uniform globally. For example, the Arctic is warming at four times the global average rate², and mountainous regions are also warming rapidly³. Previous research has addressed these differences in warming rates at different latitudes and altitudes but has neglected seasonal differences in warming. Changes in temperature across different seasons not only blur the boundaries between seasons and affect human productivity and lifestyles⁴, but also change crop phenology⁵, disrupt animal reproduction and migration, and lead to various risks and disasters. For example, rising summer temperatures exacerbate heatwaves⁶ and wildfires⁷, whereas changes in winter temperatures can affect precipitation patterns⁸, snowmelt⁹, and seasonal runoff distributions¹⁰.

The arid region of Northwest China (ARNC) is located in the hinterland of the Eurasian continent, west of the Helan Mountains and north of the Qilian Mountains-Kunlun Mountains, including all of Xinjiang and the Hexi Corridor, with geographic coordinates between 73°–106°E and 35°–50°N, accounting for about a quarter of China’s land area. This region, one of the driest in the world, has a fragile ecosystem with a complex topography of mountains and basins. Water serves as a critical connection between the three major ecosystems: mountains, oases, and deserts. The climatic and hydrological elements of the ARNC are highly sensitive to climate change. Meteorological station data reveal that the overall temperature in the region has been on an upward trend over the past half-century (0.32 °C/decade)¹¹, which is significantly higher than the national average (0.25–0.29 °C/decade) and the global average (0.13 °C/decade)¹². In 1998, a “leapfrog” warming occurred, with the annual average temperature rising from 7.50 °C in 1960–1997 to 8.63 °C in 1998–2021. The temperature changes in the ARNC exhibit seasonal differences. Some studies have shown that a substantial increase in winter temperature may be the main reason for the increase in annual average temperature¹¹, with a contribution rate of up to 57.01%¹³. This phenomenon is consistent with the general background of strong winter warming in the middle and high latitudes of the Northern Hemisphere¹⁴. However, since the 1990s, a widespread cooling trend has appeared in the winters of the Eurasian continent^15,16, with some regions experiencing cold and snowy winters and frequent extreme cold wave events¹⁷. This is contrary to the expectations of strong winter warming. Some studies have also noted that spring warming has begun to occur and shows a strong upward trend¹⁸. Here, we hypothesize that winter may no longer be the fastest-warming season in the ARNC and that the season with the dominant temperature change may have shifted. To understand the asymmetric warming of seasons, it is necessary to examine the factors affecting temperature. Temperature change is a key indicator of climate change, and its driving factors are complex and diverse. In addition to traditional factors, human activity, altitude, latitude, cloud cover (CLD), aerosol optical depth (AOD), soil moisture (SM), and atmospheric circulation also have significant effects on temperature change. Clouds affect the surface energy balance by reflecting incoming solar radiation and capturing longwave radiation from the surface. During the day, thick clouds reflect more solar radiation, reducing surface absorption and having a cooling effect. At night, clouds capture longwave radiation from the surface, reducing surface heat loss and exerting a warming effect. Soil moisture affects surface energy fluxes, influencing temperature. This mechanism can lead to SM-temperature feedback, where increasing temperature leads to increased atmospheric moisture demand, resulting in soil drying and further enhancing the initial temperature increase¹⁹. Aerosols can affect Earth’s climate in two ways. When the sky is clear (no clouds), aerosols reflect incoming sunlight into space, blocking some of the energy that would otherwise reach the surface, thus having a cooling effect on the climate. Some key circulation systems also affect temperature changes in the ARNC. The Siberian High is a cold high-pressure system located in northeastern Asia. It usually occurs in winter and can affect weather patterns over much of the Northern Hemisphere. Previous studies have shown that an increase in the Siberian High could cause cooling over the Eurasian continent¹³. Understanding seasonal temperature changes in the ARNC requires in-depth research on the relationship between temperature and these factors.

Based on station and multiple reanalysis datasets (CRU TS, ERA5, and CN-1 km), this study investigated the dominant season of temperature change in the ARNC from 1960 to 2020. It explores whether the dominant season changed during this period and attempts to determine, from a mechanistic perspective, which major climatic factors affect the dominant warming season in the region. Our results demonstrate a clear jump in the warming rates for winter and spring around the 1990s for the ARNC region, resulting in enhanced warming during spring and cooling during winter after this jump. The reduction of winter warming or even cooling after the 1990s is due to a strengthening of the Siberian High. An increase in solar radiation caused by a decrease in cloud cover was the main reason for recent spring warming. This study offers novel scientific insights into the influence of climate change on the asymmetric seasonal warming observed in this region.

Results

The observed shift in the dominant warming season from winter to spring

Based on a 30-year sliding window trend analysis of seasonal temperature using station observation data, the dominant season contributing to warming in the ARNC shifted from winter to spring. As shown in Fig. 1, during the early period, the winter warming rate was the fastest, far exceeding that of the other seasons, with minimal spring temperature change. However, in recent years, this pattern has reversed, with spring warming accelerating and exceeding changes in other seasons. During this period, winter temperatures exhibited a cooling trend.

Fig. 1: Seasonal temperature change trends based on stations.

a Seasonal warming rates during 1960–2020 with a 30-year moving window (the X-axis is the center year of the moving window). b The season with the fastest warming rate in the early period (1960–1990) and (c) the recent period (1991–2020). d, e Spring temperature change trends in the early and recent periods. f, g Winter temperature change trends in the early and recent periods.

To further illustrate this shift, we compared the seasons with the fastest warming rate at each station during the early (1960–1990) and recent (1991–2020) periods (Fig. 1b, c). In the early period, winter was the fastest-warming season at 86.5% of the stations, while the combined percentage of autumn and summer as the fastest-warming seasons was less than 15%. In the recent period, spring became the fastest-warming season at the vast majority of stations (91.9%), whereas no station had winter as the fastest-warming season.

The spatial distribution of temperature change trends in spring and winter at the stations further supports this shift. During 1960–1990, the warming rate in winter was 0.06 °C/year, with 97.3% of areas experiencing warming (35.1% significantly). Spring showed a cooling trend of 0.006 °C/year, with the majority of areas (67.6%) cooling. During 1991–2020, this pattern reversed, with winter showing a cooling trend of −0.008 °C/year, and 64.9% of areas cooling, while spring exhibited a significant warming trend of 0.07 °C/year at all stations.

Considering the sparse distribution of stations and lack of spatial representativeness, we repeated the analysis using the CRU TS, ERA5, and CN-1 km datasets (Figs. 2 and S1 and 2). All three datasets showed that winter was the dominant warming season in the early period, whereas spring warming surpassed it in the recent period, becoming the dominant season. Spatially, during 1960–1990, spring was dominated by cooling (CRU TS: −0.005 °C/year, ERA5: −0.02 °C/year, CN-1 km: −0.0033 °C/year), with the cooling area exceeding 70% in all three datasets. In the same period, all areas in winter showed an increasing trend (CRU TS: 0.061 °C/year, ERA5: 0.038 °C/year, CN-1 km: 0.062 °C/year). During 1991–2020, spring showed a strong warming trend, with the warming area reaching almost 100% in all three datasets. Winter exhibited a general cooling trend (CRU TS: −0.002 °C/year, ERA5: −0.015 °C/year, CN-1 km: −0.034 °C/year), with the cooling concentrated in the northern Xinjiang region.

Fig. 2: Temperature change trends based on CRU TS data.

a Spring and winter warming rates during 1960–2020 with a 30-year moving window, and the X-axis is the center year of the moving window. b Spring warming rate changes in 1960–1990 and (c) 1991–2020. d Winter warming rate changes in 1960–1990 and (e) 1991–2020. The black dots represent significance at the p < 0.05 level.

CRU TS exhibited closer agreement with station observations compared to ERA5, particularly during spring and winter (Fig. S3), with smaller errors (RMSE = 3.89 in spring, RMSE = 3.00 in winter) and higher correlations (R = 0.72 in spring, R = 0.82 in winter). The following mechanistic interpretation is based on the results of CRU TS and station data. Both datasets revealed a shift in seasonal temperature contributions in the ARNC. Spring’s contribution rose from −5% to −7% (1960–1990) to 58%–59% (1991–2020), while winter’s contribution fell from 60%–75% to −4% to −9%.

Possible mechanisms of seasonal shifts in climate warming

Previous studies have indicated that warming in the ARNC is mainly driven by an increase in the minimum temperature^13,20. However, some studies have suggested that the warming rate of maximum temperatures has recently begun to exceed that of minimum temperatures²¹. Therefore, we re-evaluated the role of diurnal temperature changes in the leading shift during warming seasons. Figure 3 shows the trend changes in average, minimum, and maximum temperatures in winter and spring during 1960–1990 and 1991–2020. Warming in the early winter period was mainly driven by an increase in minimum temperatures, whereas cooling in the recent winter period was mainly influenced by a decrease in maximum temperatures. In spring, cooling in the early period was mainly driven by a decrease in the maximum temperature, while recent warming was also mainly related to an increase in the maximum temperature. These analyses suggest that changes in spring temperatures are closely related to maximum temperatures, whereas a shift from minimum to maximum temperatures occurs due to factors that influence changes in winter temperatures. This may also imply that the mechanisms affecting winter and spring variability in the ARNC differ.

Fig. 3: Changes in average, minimum, and maximum temperatures in the ARNC during winter and spring.

Based on CRU TS data, trends in winter and spring (a) average, (b) minimum, and (c) maximum temperatures in 1960–1990 and 1991–2020. Asterisk represents significant changes. Based on CRU TS data, warming rates of (d) spring and (e) winter average, minimum, and maximum temperatures during 1960–2020 with a 30-year moving window (the X-axis is the center year of the moving window).

Previous studies have shown that in addition to greenhouse gases, temperature changes are largely affected by changes in CLD, AOD, and SM²¹. These drivers are highly correlated, with AOD affecting cloud albedo and lifetime, and reduced CLD leading to decreased precipitation, which can cause soil drought and exacerbate land-atmosphere feedback. To address the multicollinearity problem and accurately identify the correlations between temperature and CLD, AOD, and SM, we conducted a partial correlation analysis based on CRU TS cloud data, MERRA-2 AOD, and Global Land Evaporation Amsterdam Model SM data for the ARNC from 1980 to 2020.

Figure 4 shows the spatial distributions of the partial correlation coefficients between CLD, AOD, SM, and spring and winter temperatures. Spring temperature was negatively correlated with CLD and SM and positively correlated with AOD. The correlation coefficients were −0.64, −0.10, and 0.18, respectively. A significant negative correlation between temperature and CLD was observed in 91.8% of the study area, whereas only 16.3% and 15.4% of the study area exhibited significant correlations with AOD and SM, respectively. Winter temperature was positively correlated with CLD and SM and negatively correlated with AOD. The correlation coefficients were 0.05, 0.20, and −0.21, respectively, and none were significant. Spatially, temperature showed a heterogeneous relationship with the three drivers, with positive correlations in the northern Xinjiang and Hami regions and negative correlations in the Kashi region. Areas with significant correlations between temperature and CLD, AOD, and SM were also smaller, at 24.9%, 20.1%, and 6.9%, respectively. This analysis suggests that CLD is likely the main factor affecting spring temperatures, whereas changes in winter temperatures may not be significantly related to these three factors.

Fig. 4: Relationship between spring and winter temperatures and CLD, AOD, and SM.

Spatial distribution of partial correlation coefficients between spring temperature and (a) CLD, (b) AOD, and (c) SM. Spatial distribution of partial correlation coefficients between winter temperature and (d) CLD, (e) AOD, and (f) SM. Black dots represent p < 0.05. g The partial correlation coefficients between CLD, AOD, and SM and spring temperature. h The partial correlation coefficients between CLD, AOD, and SM and winter temperature. Two asterisks represent p < 0.01. i Standardized time series of spring temperature and CLD. j Standardized time series of winter temperature and CLD. The original time series was conducted as a 10-year moving average.

Reduced CLD can increase solar radiation during the day, thereby exacerbating temperature increases. Figure 5 compares the spatial distributions of CLD and radiation changes from ERA5 during 1960–1990 and 1991–2020 to further reveal whether CLD changes affected the spring and winter temperature changes. In the early period, CLD in spring and winter increased at rates of 0.07/a and 0.11/a, respectively. In the recent period, CLD in spring and winter both showed a decreasing trend, with rates of −0.11/a and −0.09/a, respectively. Spatially, during the early period, areas with increasing CLD in spring and winter accounted for 72.2% and 78.7%, respectively, and were concentrated in southern Xinjiang and the Hexi Corridor regions. In the recent period, areas with decreasing CLD in spring and winter accounted for 94.3% and 91.1%, respectively, and were concentrated in northern Xinjiang. The CLD changes based on CRU TS data also showed similar trends. Correspondingly, the radiation in spring and winter exhibited a decreasing trend in the early period and an increasing trend in the recent period. During 1991–2020, areas with increasing radiation levels in spring and winter accounted for 88.9% and 93.7% of the total area, respectively. This analysis further illustrates that CLD reduction is the main reason for recent spring warming but is not the driving mechanism for recent winter cooling.

Fig. 5: Changes in cloud cover (CLD) and solar radiation (spring and winter) 1960–1990 vs. 1991–2020.

a–d show cloud cover change for spring and winter in two time periods. e–h show solar radiation change for spring and winter in two time periods. CLD units are (%/a), and radiation units are (J m⁻²). Black dots represent p < 0.05.

Previous studies have indicated that winter cooling over Eurasia may be related to large-scale circulation factors²². In Fig. 6, we investigate the potential influence of large-scale circulation on winter temperature trends in the ARNC. The results of the correlation analysis between the five circulation factors that widely affect the ARNC and winter temperature show a significant negative correlation between the SHI and winter temperature (detrended winter temperature) (R = −0.57 without detrending; R = −0.63 with detrending, Fig. S4). The enhanced SHI may be the main reason for winter cooling in the region. The negative SHI anomaly is consistent with the winter temperature anomaly fluctuation. The period of the 1980s and 1990s was the weakest period of the Siberian High over the past 100 years. During this period, winter temperatures in the ARNC showed an increasing trend. With an increase in the SHI after the 1990s, winter temperatures began to decrease. Spatially, the SHI and winter temperature showed a significant negative correlation in almost all regions, with the correlation coefficient increased from southwest to northeast.

Fig. 6: Driving mechanisms of winter temperature cooling.

a Standardized time series of winter temperature and SHI. Note that the standardized time series of SHI is multiplied by −1. b Spatial distribution of the correlation coefficient between winter temperature and SHI. Black areas represent p < 0.05. c Composite analysis of surface temperature for strong SHI years, weak SHI years, and the difference between the two. d Composite analysis of surface pressure and wind field for strong and weak SHI years and the difference between the two. Black dots represent significant differences with p < 0.05.

To unveil how the SHI influences winter temperatures in the ARNC, we employed a composite analysis method to investigate the differences in temperature, surface pressure, and wind field between strong and weak SHI years. Strong SHI years correlated with colder winters in the region, driven by multiple factors. North of 40°N, a widespread area exhibited colder-than-average temperatures (negative temperature anomalies). Furthermore, a high-pressure system dominated the West Siberian Plain, generating a large anticyclone. This system directed prevailing winds from the north, transporting cold air southward and inducing significant cooling in the ARNC. Conversely, weak SHI years were characterized by warmer winters. Across Eurasia, temperatures were generally higher than average (positive temperature anomaly), and the West Siberian Plain experienced low pressure, forming a cyclone. This low-pressure system facilitated southerly winds, which obstructed the southward movement of cold air, resulting in warmer temperatures during weak SHI years. This substantial contrast in temperature anomalies between strong and weak SHI years was widespread across Central Asia, the Central Siberian Plateau, and northern China.

Cloud formation is significantly influenced by variations in sea-level pressure. High-pressure systems typically impede cloud formation as they induce downward air movement. To further explore the relationship between sea-level pressure and the decline in spring cloud cover, we analyzed changes in surface and tropospheric sea-level pressure between 1960 and 2020 (Fig. S5). Composite analysis revealed that from 1960 to 1990, sea-level pressure exhibited a positive anomaly (higher pressure) north of 50°N and a negative anomaly (lower pressure) south of 50°N. However, this pressure anomaly pattern reversed between 1961 and 2020, with positive anomalies observed in the northwest arid region relative to surrounding areas. Similar patterns were observed for the troposphere, suggesting that in recent years, subsidence prevailed across the entire northwest region, from the upper troposphere to the surface, favoring reduced cloud cover and increased solar radiation.

Discussion

Based on station observations and three sets of reanalysis data, this study reveals the phenomenon of asymmetric warming in the ARNC and demonstrates that the season dominating temperature rise shifted from winter to spring. Previous studies in regions such as western North America²³ and China²⁴ generally considered winter to be the season with the fastest warming rate, but this dominant season is not constant. This study demonstrates that this seasonal shift extends beyond the ARNC but also exists on the wider Eurasian continent. Situated at the periphery of the ARNC in Central Asia, the latest research also confirms that spring has the fastest warming rate¹⁸. It is worth noting that the dominant position of winter warming has not only been replaced by spring warming but has also shown a downward trend in recent years. When studying the extratropical land of the Northern Hemisphere, the winter trend in the north significantly differs from that in the other three seasons. Since 1987, there has been no statistically significant warming trend in winter in the Northern Hemisphere. Winter temperature is close to neutral or shows a downward trend. In contrast, the other three seasons showed significant warming trends²⁵. This indicates that this type of dominant seasonal change may occur widely in regions other than the ARNC.

Our study also indicated that the mechanisms of recent spring warming and winter cooling differ. Strong spring warming was related to the increase in radiation caused by the decrease in CLD, whereas winter cooling was mainly controlled by the change in the intensity of the SHI. The influence of CLD on spring temperatures has also been found in the Eurasian continent and China²⁶. The increase in radiation intensity caused by a decrease in CLD is consistent with the recent change from dimming to brightening after the 1980s²⁷. The decrease in CLD also explains why the global diurnal temperature range has recently changed from increasing to decreasing²¹. An increase in radiation caused the maximum temperature to increase faster than the minimum temperature. Similarly, we found that spring warming in the ARNC was mainly caused by an increase in the maximum temperature.

Our study further highlights that subsidence prevailing across the entire ARNC from the upper troposphere to the surface is a key circulation condition contributing to the decline in spring cloud cover. Additionally, the recent resurgence of spring wind speeds, changes in water vapor content, and variations in condensation nuclei could also influence spring cloud formation processes. Li et al.²⁸ found that due to the asymmetric warming between high and low latitudes, the declining trend of wind speed in the ARNC has reversed to an upward trend in 1992, creating conditions for more clear skies. Cloud cover is directly proportional to humidity, and relative humidity is lowest in the northwest arid region during spring and has shown a decreasing trend in recent years, in line with rising temperatures²⁹. Moreover, the improved air quality also reduces the availability of condensation nuclei essential for cloud formation³⁰.

The impact of cloud cover on spring temperature is complex and can be influenced by factors such as cloud thickness, type, altitude, and diurnal cycle. The first is the Reduction in Cloud Cover. A decrease in cloud cover, particularly during spring, leads to increased absorption of solar radiation by the Earth’s surface, contributing to atmospheric warming. This effect is more pronounced in spring compared to winter due to the larger seasonal variation in incoming solar radiation, with significantly higher solar irradiance during spring. The second is the differential Impacts of Cloud Height and Type. Different cloud types and heights exhibit varying radiative properties, influencing their impact on surface temperature. High-level clouds, such as cirrus, cirrocumulus, and cirrostratus, are typically thin and have low reflectivity, allowing more solar radiation to reach the surface and contribute to warming. Conversely, low-level clouds, such as cumulus, stratocumulus, and nimbostratus, are often thicker and more reflective, scattering incoming solar radiation back into space and promoting cooling. Analysis of NCEP data reveals a more pronounced decline in low-level clouds compared to high-level clouds over the ARNC (Fig. S6). This trend is particularly evident in the high-altitude Tienshan Mountains, further amplifying the elevation dependence of warming. The reduction in low-level clouds allows a greater proportion of solar radiation to penetrate the atmosphere and be absorbed by the Earth’s surface, exacerbating spring warming. The third is the Diurnal Regulation by Cloud Cover: Clouds play a crucial role in regulating energy balance throughout the day. During the daytime, clouds cool the Earth by reflecting solar radiation back into space. Conversely, at night, clouds act as an insulating blanket, trapping infrared radiation and preventing heat loss from the Earth’s surface. The extent of absorption and emission by clouds depends on various factors, including cloud thickness, water content, saturated vapor density, cloud location, and wavelength.

Previous studies have provided explanations for the winter cooling phenomenon; however, the mechanisms are relatively complex. The recent decline in winter temperatures is primarily attributed to the strengthening of the Siberian High, which has induced high pressure and northerly winds, bringing in large amounts of cold air. While the high-pressure system generates subsiding air currents that hinder cloud formation and lead to increased solar radiation (winter also has the weakest solar radiation throughout the year), this cannot offset the strong cold wave brought by the Siberian High, resulting in an overall decrease in winter temperatures. Some studies have pointed out that in recent decades, the Arctic has been warming rapidly, whereas winter in the Northern Hemisphere has been abnormally cold. It is speculated that this “Arctic warming, continental cooling” trend pattern is caused by sea ice loss³¹. However, other model-based studies have shown that the loss of Arctic sea ice and the increase in high-latitude sea surface temperature have not affected large-scale atmospheric circulation features (such as the Arctic Oscillation, the North American Pattern, and Eurasian winter blocking frequency) in the Northern Hemisphere over the past few decades. The observed increase in the Eurasian winter blocking frequency over 1998–2012 appears to be a manifestation of atmospheric internal variability³². Although this study explored the mechanisms of winter cooling in strong and weak SHI years, the impact of Arctic sea ice on the climate of this region requires further exploration.

While the paper delves into the primary drivers of seasonal temperature asymmetry in the ARNC, several additional factors (such as snow/ice albedo feedback, water vapor, aerosol, and land-use) and feedback mechanisms play a crucial role in shaping these seasonal variations³. Intensified spring warming triggers accelerated snowmelt, exposing the darker underlying land surface. This increased absorption of solar radiation by the darker land enhances warming, reinforcing a positive snow albedo feedback (Li et al. 2022)³³. Conversely, winter cooling promotes the expansion of snow cover. Snow’s high albedo reflects more solar radiation back into space, amplifying the winter cooling effect. Dust storms and sandstorms prevalent in arid regions in the spring affect solar radiation and hence atmospheric temperatures. Dust deposition on glaciers and snow cover further accelerates their melting through a phenomenon known as snow darkening³⁴. Additionally, land-use changes also alter surface albedo and heat fluxes, potentially influencing seasonal temperature variations³⁵. Deforestation, for instance, replaces vegetation with lower albedo surfaces, increasing solar radiation absorption and contributing to warming.

ARNC relies heavily on snowmelt for water, but climate change disrupts this delicate system³⁶. Asymmetric seasonal temperature variations alter seasonal snowmelt patterns, impacting both water resources and ecosystems. Ecological impacts: while beneficial for snowpack accumulation, prolonged winter cooling delays spring vegetation growth and shortens the growing season, ultimately reducing plant productivity³⁷. In comparison, an extended growing season due to spring warming comes at a cost—increased risk of frost damage and earlier water uptake by plants, leading to summer soil drying and ecosystem stress^38,39. Water resource impacts: the colder climate paradoxically boosts snowfall but also induces soil freezing, impeding water infiltration and groundwater replenishment. Rapid spring warming leads to excessive snowmelt, surpassing the ground’s absorption capacity, thus increases surface runoff at the expense of groundwater recharge, intensifying water scarcity in this arid region. Additionally, earlier spring snowmelt may result in reduced summer snowmelt and therefore pose challenges for water management and irrigated agriculture⁴⁰.

This study provides new insights and perspectives on climate change in the ARNC over recent decades. Under the influence of greenhouse gas emissions, temperatures in this region will continue to rise in the future. The study highlights the complexity of climate change at the regional scale, revealing that temperature variations not only exhibit altitudinal and spatial differences but also demonstrate seasonal asymmetry. Investigating the mechanisms behind these disparities in the fastest-warming seasons is crucial for improving predictions of the intensity and distribution of future climate warming in this region. The ARNC’s water resource system, which relies on mountain precipitation and snowmelt recharge, is particularly sensitive to temperature changes in different seasons, especially the potential increase in spring warming. This may increase the difficulty of future water resources management, necessitating the development of appropriate engineering measures in advance to deal with reasonable runoff allocation, flood, and drought prevention. In the future, we will evaluate the ability of models to simulate this seasonal warming asymmetry and continue to monitor whether the seasons that dominate temperature changes in the ARNCs will continue to shift. Meanwhile, we consider extending the findings of this study to the global scale, analyzing the seasons that dominate temperature changes in different regions, and understanding whether this shift constitutes a general pattern globally.

Methods

Study area

The ARNC is situated in the mid-latitudes of Eurasia (Fig. 7), characterized by a continental climate predominantly shaped by mountainous terrain and deserts. Approximately 80% of the region’s total area comprises arid and semi-arid landscapes, encompassing deserts and the Gobi, contributing to its dry climate. Deep within the continental hinterland, the ARNC experiences a multiyear average annual precipitation of 156.36 mm, with evaporation rates 8–10 times higher than precipitation¹³. Inland rivers, such as the Tarim River, Ili River, and Hei River, dominate the ARNC landscape, originating primarily from mountainous regions. These rivers draw water from diverse sources, including glacial snowmelt in high mountainous areas, precipitation in mid-mountain forest belts, and groundwater from bedrock fissures in low mountainous regions. Climate change significantly impacts water resources in the ARNC.

Fig. 7: Location of the arid region of northwest China (ARNC).

The red triangles show the stations used in this study. The figure in the upper right corner shows the global location of the ARNC.

Dataset

Daily surface air temperature data were acquired from the China Meteorological Data Service Center (http://data.cma.cn/en). This dataset underwent strict quality control, including extreme value and time-consistency tests, before release. The quality control process employed the methods of Feng et al.⁴¹ with a more stringent data screening methodology. Meteorological stations with missing temperature data exceeding 15 days or with more than 1% of the annual data unrecorded were excluded from the analysis. Missing data were interpolated using linear interpolation in the remaining cases. After the screening, 37 stations from 1960 to 2020 were selected, with missing values for all stations less than 0.1%. Detailed information on the meteorological stations is presented in Table S1.

To effectively elucidate spatial temperature variability and ensure the robustness of our findings, we utilized three reanalysis products: the Climatic Research Unit Time-Series version 4.07 (CRU TS), the fifth generation of the European Centre for Medium-Range Weather Forecasts product ERA5, and a 1 km monthly temperature and precipitation dataset tailored for China (CN-1 km).

The CRU TS temperature dataset integrates data from diverse sources, encompassing weather station records, ship logs, and modern satellite observations⁴². Stringent calibration procedures have been employed to mitigate biases and inconsistencies arising from disparate measurement techniques, establishing it as a cornerstone resource in climate research for tracking long-term temperature patterns and variations. This dataset provides monthly spatiotemporal data at a resolution of 0.5° × 0.5°, covering the period from 1901 to 2022. (https://crudata.uea.ac.uk/cru/data/hrg/).

ERA5, the fifth-generation reanalysis product developed by the European Centre for Medium-Range Weather Forecasts, is specifically designed to serve as meteorological forcing data for land surface and hydrological models⁴³. Reanalysis seamlessly merges model data with observational records from across the globe, utilizing the principles of physics to generate a globally comprehensive and cohesive dataset. The temperature dataset was obtained from ERA5 monthly averaged data. This dataset offers a monthly spatiotemporal resolution of 0.25° × 0.25°, spanning 1940 to the present. (https://cds.climate.copernicus.eu/cdsapp#;/home).

The CN-1 km dataset boasts a spatial resolution of approximately 1 km, covering the period from 1901 to 2021⁴⁴. It was meticulously crafted by downscaling the global 0.5° climate dataset released by CRU and the high-resolution global climate dataset published by WorldClim, using a delta spatial downscaling method optimized specifically for the China region. Additionally, it underwent validation using data from 496 independent meteorological observation points to ensure result reliability (https://poles.tpdc.ac.cn/en/data/71ab4677-b66c-4fd1-a004-b2a541c4d5bf/).

To further explore the potential mechanisms underlying temperature variations, this study utilized data on radiation, CLD, SM, and AOD. Monthly incident shortwave radiation data were sourced from the ERA5 dataset, provided on a 0.25° grid, representing all-sky conditions. Monthly total CLD data were acquired from the CRU TS dataset, provided on a 0.5° grid. Monthly surface SM data were obtained from the Global Land Evaporation Amsterdam Model version 3.5 dataset, featuring a spatial resolution of 0.25°. This global dataset spans the period from 1981 to 2020 and is derived from satellite and reanalysis data (https://www.gleam.eu/#dataset). Monthly AOD data were sourced from the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) dataset, starting in 1980, with a spatial resolution of 0.625° × 0.5° (https://disc.gsfc.nasa.gov/datasets?project=MERRA-2).

Additionally, temperature, surface air pressure, tropopause air pressure, and wind speed (U and V components) were selected to explore the effects of large-scale circulation factors on temperature changes in the ARNC using monthly global atmospheric reanalysis data. These data were jointly developed by the National Centers for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR), featuring a spatial resolution of 2.5° and temporal coverage spanning from 1948 to the present (https://psl.noaa.gov/data/gridded/data.ncep.reanalysis.html).

Previous studies have confirmed that large-scale circulation factors have important impacts on temperature in the ARNC^13,42. This study selected five major circulation factors that affecting temperature in the region: the Siberian High index (SHI), Arctic Oscillation index (AOI), North Atlantic Oscillation Index (NAOI), Pacific-North American Pattern Index (PNAI), and the Southern Oscillation Index (SOI). Except for SHI, which was calculated, data for the other four factors were obtained from the Climate Prediction Center of the National Weather Service, U.S. (http://www.cpc.ncep.noaa.gov).

Seasonal contribution analysis

To analyze seasonal temperature variations within the ARNC, we defined spring as including March, April, and May; summer as June, July, and August; autumn as September, October, and November; and winter as December, January, and February. Correspondingly, the annual mean value of a variable for a given year was computed as the average of the 12 months from December of the preceding year to November of that year. The Mann–Kendall statistical test^45,46 assessed the statistical significance of trends in the annual mean temperatures across the entire study area. To identify whether the dominant season contributing to warming in the ARNC has changed, a 30-year moving window was used to calculate the temperature change trend for each season. For example, for the central year 1990, the time period would be 1975–2005

To quantify the importance of each season’s temperature change to the annual change, the study period was divided into two parts: 1960–1990 (early period) and 1991–2020 (recent period). Taking the early period as an example, the calculation process for the importance of a certain seasonal change in the annual change¹³ was as follows:

$${I}_{s,i}=\frac{\left|{T}_{s,i}-{\mathop{T}\limits^{\leftharpoonup}}_{{S},1960-1990}\right|}{\sum \left|{T}_{s,i}-{\mathop{T}\limits^{\leftharpoonup}}_{S,1960-1990}\right|}\times 100 \%$$

(1)

$${\rm{s}}\in \left\{{\hbox{``}} {\rm{spring}}{\hbox{''}} ,{\hbox{''}} {\rm{summer}},{\rm{autumn}},{\rm{winter}}{\hbox{''}} \right\}$$

(2)

where ${I}_{s,i}$ is the importance of the winter temperature variation in year i to the overall temperature variation in the same year.

SHI calculation

To investigate the dynamics between the winter temperature trend and the SHI, we calculated the SHI using the definition of Gong and Wang⁴⁷, with the following equation:

$${{{SHI}}}=\frac{{\sum }_{i=1}^{n}{P}_{i}\lambda \cos {\varphi }_{i}}{{\sum }_{i=1}^{n}\lambda \cos {\varphi }_{i}}$$

(3)

where SHI represents Siberian High index value for the entire region, ${P}_{i}$ is the sea level pressure value at a given location i; N is the total number of locations for which sea level pressure values are available (in this study N = 144); ${\varphi }_{i}$ is the latitude of the location; and $\lambda$ is a factor that limits the application of SHI calculations to only those locations where the sea level pressure value is greater than a certain threshold value. The value of $\lambda$ was set as follows:

$$\lambda =\left\{\begin{array}{cc}1 & {\text{if}}\,{P}_{i}\ge 1028\,{\rm{hPa}}\\ 0 & {\text{if}}\,{P}_{i} < 1028\,{\rm{hPa}}\end{array}\right.$$

(4)

The monthly SHI was calculated as the standardized 500 hPa geopotential height anomaly averaged over the region (50°–70°N, 80°–100°E). Geopotential height data were obtained from the NCEP/NCAR reanalysis dataset spanning the period from 1850 to 2020 on a 5° grid. Using these data, we calculated the winter SHI for the period 1960–2020.

Statistical analysis

In this study, Pearson’s correlation and partial correlation analyses were utilized to examine the correlation between temperature and other climatic factors in the ARNC. Pearson’s correlation analysis was employed to ascertain the linear correlation between two time series, denoted as M and N. The correlation coefficient ranges between ±1 and is determined by the following formula:

$$r=\frac{{\sum }_{i=1}^{n}\left({M}_{i}-{\rm{M}}\right)\left({N}_{i}-\mathop{N}\limits^{\leftharpoonup}\right)}{\sqrt{{\sum }_{i=1}^{n}{\left({M}_{i}-\mathop{M}\limits^{\leftharpoonup}\right)}^{2}}\sqrt{{\sum }_{i=1}^{n}{\left({N}_{i}-\mathop{N}\limits^{\leftharpoonup}\right)}^{2}}}$$

(5)

where r is the Pearson’s correlation coefficient, $\mathop{M}\limits^{\leftharpoonup}$ is the mean value of M, $\mathop{N}\limits^{\leftharpoonup}$ is the mean value of N, and n is the length of M and N.

Partial correlation analysis is a statistical technique used to the association between two variables while controlling for the influence of other variables. This method proves beneficial when multiple variables are correlated, and the isolation of the unique relationship between two specific variables is desired. In this study, we employed partial correlation analysis to investigate the relationship between temperature and three other variables: CLD, SM, and AOD. We computed the second-order partial correlation coefficient between temperature and each individual variable, while controlling for the effects of the remaining two. The calculation of the second-order partial correlation coefficient proceeded as follows:

$${r}_{{ij},{mn}}=\frac{{r}_{{ij},n}-{r}_{{im},n}{r}_{{jm},n}}{\sqrt{\left({1-{r}_{{im},n}}^{2}\right)\left({1-{r}_{{jm},n}}^{2}\right)}}$$

(6)

where$\,{r}_{{ij}}$ is the correlation coefficient between variables i and j; ${r}_{{im}}$ is the correlation coefficient between variables i and m, and ${r}_{{jm}}$ is the correlation coefficient between variables j and m, respectively.

Composite analysis

To analyze the impact of the SHI on winter temperatures in the ARNC, we employed a composite analysis method to explore variations in relevant meteorological variables (surface temperature, surface pressure, and wind field) between strong and weak SHI years. The SHI was standardized, with years having an index greater than 1 (or less than −1) times the standard deviation classified as strong (or weak) years. Strong SHI years included 1964, 1965, 1966, 1969, 1984, 2005, 2011, 2012, and 2018, while weak SHI years encompassed 1972, 1973, 1989, 1992, 1997, 2003, 2007, and 2015. To derive temperature, pressure, and wind speed anomalies, we removed the mean winter temperature (1960–2020). The composite analysis involved calculating the mean values for all strong and weak SHI years, followed by the utilization of Student’s t test to evaluate the significance of the differences between them.