Region-dependent meteorological conditions for the winter cold hazards with and without precipitation in China

Cold hazard is one of the major meteorological disasters in winter. However, the meteorological conditions for the cold hazard events vary significantly with both the feature of the event and the region of occurrence. This study divides winter cold hazard events in China into three categories based on the daily gridded dataset of cold hazards from November 1980 to March 2020: events without wintry precipitation (hazardous low temperature, abrupt temperature drop, and/or freezing), with wintry precipitation only (hazardous sleet and/or snowstorm), and with both. The region-dependent multivariate meteorological conditions for each category of cold hazards are investigated using ERA5 reanalysis data. Results show that the surface air temperature (T2m) and its anomaly (T2m_anom) are lower than climatology during cold hazards. But the difference in T2m among provinces exceeds 30 °C, and even for the same province, the difference among different categories of cold hazards exceeds 10 °C. The region- and category-dependent differences of T2m_anom and daily temperature drop (∆T24) are also large, about 5 °C and 2 °C d−1, respectively. The Multivariate Empirical Orthogonal Function analysis has further been applied to not only the abovementioned temperature-related variables but also the precipitation-related variables (i.e., daily accumulated total precipitation, daily accumulated snowfall, and daily mean snow depth) in the middle and lower Yangtze River region, which reveals the event-mean state and spatial–temporal coupling evolution during the progression of the event for the selected key meteorological variables. The meteorological conditions for cold hazards put forward by this study could provide region-dependent and category-dependent reference for the prediction and warning of cold hazards.


Introduction
Cold hazard is one of the major meteorological disasters in China, exerting serious impacts on transportation, agriculture, industry, electricity, forestry, fishery, animal husbandry, and economic construction as well as people's daily life (Xiao 2009;Zheng et al. 2018). Under the background of global warming, the frequency of cold extremes and cold hazards does not decrease, and the spatial nonuniformity of occurrence is increasing (Gong and Ho 2004;Ou et al. 2015;Ding et al. 2020). Understanding the meteorological conditions of cold hazards and improving their forecast skills can provide a more accurate scientific reference for cold hazard mitigation planning, which is of great social and economic significance.
The cold hazards are not the same as the extreme cold events. Because the deterministic factors for the occurrence of cold hazards include not only the meteorological conditions but also the exposure and vulnerability of non-meteorological and climatic conditions (IPCC 2014). Despite numerous studies in the past on the meteorological conditions, temporal-spatial variation, and mechanism of extreme cold weather (Gong et al. 2012;Han et al. 2021;Yu et al. 2015a, b, c;Kenyon and Hegerl 2008;Yao et al. 2017), a few studies are focusing on the cold hazards and most of them are case studies.
Several studies have demonstrated that the meteorological conditions for cold hazards with wintry precipitation (i.e., hazardous sleet and snowstorm) are remarkably different among different regions (e.g., Ye et al. 2009;Wan et al. 2008;Peng et al. 2012). Particularly, the daily mean temperature is an important factor in the freezing rain strength under the precondition of certain precipitation in Hunan province (Ye et al. 2009). In Southwest China, sleet is more tightly related to the daily minimum temperature (Zhao et al. 2011;Peng et al. 2012). In Hubei province, both the process average and the process-minimum temperatures are important for the formation of sleet (Wan et al. 2008). The ranges of temperature conditions can vary significantly in provinces and even cities (Ye et al. 2009;Wan et al. 2008;Zhao et al. 2011;Peng et al. 2012;Zheng et al. 2018). Besides temperature, sufficient water vapor supply is also necessary for the occurrence of cold hazards with wintry precipitation Wang et al. 2019). The relative humidity in the ice growth process is generally up to 80-90%, but there are remarkable regional differences as well (Peng et al. 2012).
Cold hazards without wintry precipitation include three subgroups: hazardous low temperature, abrupt temperature drop, and freezing. Among them, extreme cold events are closest to hazardous low temperature, and cold air outbreaks are closest to abrupt temperature drop. But extreme cold events and cold air outbreaks do not necessarily cause disaster in record. There are variously defined temperature conditions for indicating the occurrence of a cold air outbreak event. The most widely used condition for cold air outbreaks is the process temperature drop exceeding 10 °C and temperature anomaly less than − 5 °C (Wang and Ding 2006), or the temperature drop exceeding 10 °C per one or two days (Qian and Zhang 2007). An extreme cold event at a single station is defined when the daily minimum temperature of a station reaches the extreme low-temperature threshold of the station (Han et al. 2021). Yu et al. (2015a) used the area percentage occupied by surface air temperature below climatology by at least half of the local standard deviations to detect the continental-scale cold events. The anomalous features of synoptic systems related to stronger East Asian winter monsoon and the strengthening of the equatorward cold air branch of isentropic meridional mass circulation are crucial to the cold air outbreaks in East Asia (Gong and Ho 2004;Ding 1990;Zhang et al. 1997;Yu et al. 2015a, b, c).

3
The main reason for the lack of cold hazard studies is the lack of detailed information and accessible formatted datasets of historical occurrences of cold hazards. Recently, Yu et al. (2022) made full use of the recorded text information on winter cold hazards in China, from the official disaster annual reports and literature. The information sources include China Meteorological Disaster Extension (Comprehensive Volume) , Document (Map) and Internet (2001-2003), China Meteorological Disaster Review (2004, and public information collected from network and national emergency management department ( -2020. With the complementary meteorological information derived from high-resolution reanalysis data, Yu et al. (2022) constructed a daily gridded dataset of each subgroup of cold hazards (i.e., low temperature, abrupt temperature drop, freezing, sleet, and snowstorm) covering the period from 1980 to 2020.
Using the daily gridded cold hazard dataset in the China domain derived by Yu et al. (2022), this study aims at revealing the region-dependent meteorological conditions for each subgroup of cold hazards in China. The remainder of this paper is organized as follows. In Sect. 2, data and methods are introduced and we divide the cold hazard events into three categories based on the occurrence of the subgroup of cold hazards. Results are shown in Sect. 3 and Sect. 4. Section 3 demonstrates the region-dependent temperature conditions during three categories of cold hazards. In Sect. 4, the middle and lower Yangtze River region is selected as a typical region, and the Multivariate Empirical Orthogonal Function (MV-EOF) analysis is conducted to take other critical meteorological conditions into the synthesized consideration for the multivariable coupled meteorological conditions of cold hazards. Discussion and conclusions are given in Sect. 5 and 6, respectively.

Data
The cold hazard data used in this study are the daily gridded cold hazard occurrence dataset in the China domain derived by Yu et al. (2022). The spatial resolution of the data fields is on 0.5° latitude × 0.66° longitude grids covering the period from November 1980 to March 2020. For the occurrence of each subgroup of cold hazards (i.e., low temperature, abrupt temperature drop, freezing, sleet, and snowstorm), the corresponding variable value is set as 1 when the corresponding cold hazards occur at this grid point, otherwise is set as 0. The values of occurrence are the same at all the grid points belonging to the same province, because of the limited level of detail in the official records (see details in Yu et al. 2022).
The meteorological variables of interest include daily mean 2-m temperature (T 2m ), anomalies of daily mean 2-m temperature (T 2m _ anom ), 24-h 2-m temperature change (∆T 24 ), daily accumulated total precipitation (PREP), daily accumulated snowfall (SNOW), and daily mean snow depth (SNODP). The nationwide observation data of meteorological variable fields can be obtained from the automatic weather stations that are available since 1995. The other data source is the reanalysis dataset, which has a large variety of variables, higher spatial and temporal resolutions. We compare the reanalysis datasets including Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA2) from the Goddard Earth Observing System Model, Version 5 (Gong et al. 2020), ERA5 from European Centre for Medium-Range Weather Forecasts (ECMWF, Hersbach et al. 2020), and JRA55 (Japanese 55-year Reanalysis, Hou et al. 2017) with the observation 1 3 data. The root-mean square error (RMSE) of the three reanalysis datasets with the observations (see Fig. A1 in Supplementary Information) shows that the temperature-related variables in MERRA2 have relatively large deviations, while the precipitation-related variables in JRA55 have large deviations from observational data. In addition, the variables such as PREP in ERA5 are closest to the observations in the middle and lower Yangtze River Region during the cold hazard events that we investigated (Figs. A2-3 in Supplementary Information). Moreover, the ERA5 reanalysis data use a more recent model and data assimilation system (Albergel et al. 2018), which consequently provide products with higher accuracy and good applicability in most of China (Liu et al. 2021;Jiao et al. 2021). Therefore, we use in this study the daily fields of T 2m , PREP, SNOW, SNODP derived from the hourly ERA5 reanalysis dataset in the time period from 1980 to 2020. The ∆T 24 at a given day is derived by subtracting the surface air temperature at 00Z GMT the next day from that at 00Z GMT today. The daily climatological-mean T 2m fields are obtained by averaging the data across the 40 years for each calendar day from 1 November to 31 March. T 2m _ anom fields are obtained by removing the daily climatology from the total field of T 2m .

Categorization of cold hazard events
According to the cold hazard records, in some places (such as Yunnan Province, etc.), cold hazards could occur without extreme cold temperature or temperature drop due to the existence of wintry precipitations. Because of the highly possibly different disaster-caused meteorological conditions between cold hazards with and without wintry precipitation, the cold hazards are divided into three categories depending on whether it is with wintry precipitation or not in this study. Specifically, a cold hazard is defined as cold hazards without wintry precipitation (denoted as "DRY-C") if it is accompanied by hazardous low temperature and/or abrupt temperature drop and/or freezing occur but no hazardous sleet and snowstorm. A cold hazard is defined as cold hazards with wintry precipitation only (denoted as "WET-C") if hazardous sleet and/or snowstorm occur without accompany by hazardous low temperature and abrupt temperature drop and freezing. The remaining cold hazard events are denoted as "BOTH-C," during which hazardous low temperature and/ or abrupt temperature drop and/or freezing occur, in accompany by hazardous sleet and/or snowstorms.
The spatial distribution of the total number of occurrence days of the three categories of cold hazards in China from 1980 to 2020 is shown in Fig. 1. The occurrence days of the DRY-C events are generally less than those of WET-C and BOTH-C events, inferring that the possibility of the low temperature, abrupt temperature drop, and/or freezing alone to inducing hazards is relatively low. In other words, the disaster-caused capability of cold hazards with wintry precipitation (sleet and/or snowstorm) is much stronger. We can also see from Fig. 1 the remarkable regional differences among the occurrence of cold hazards belonging to different categories. Specifically, the larger occurrence of DRY-C events is found mainly in Northern China. WET-C events prefer to occur in pastoral areas such as Western China and Inner Mongolia. Large values of occurrence days of BOTH-C events center in South China. The occurrence days of the three categories of cold hazards are roughly equal in North and Northeast China, while for the rest regions, one or two categories of cold hazards are dominant.

MV-EOF analysis
To reveal the dominant multi-variates spatial-temporal evolution of coupling features during cold hazards, we consider the six key meteorological variables from perspectives of temperature (T 2m , T 2m _ anom, and ∆T 24 ) and precipitation (PREP, SNOW, and SNODP). An MV-EOF analysis is conducted based on both spatial and inter-variable coherence (Wang 1992). In specific, we use symbols x 1 , x 2 , ..., x 6 to indicate six key variables, i.e., T 2m , T 2m _ anom , ∆T 24 , PREP, SNOW, and SNODP. We first perform a temporal average during all the days of selected events and get x 1 , x 2 , ..., x 6 , where n is the occurrence number of days that are shown in Fig. 1. The event-mean fields of six key variables are the function of latitude and longitude and are denoted as T 2m , T 2m _ anom , ∆ T 24 , PREP , SNOW and SNODP . They together make up the matrix MEAN , which represents the average meteorological conditions for a certain category of cold hazards.
We subtract the event-mean fields from the total fields of six key variables to get the deviations x di , which is the function of time t , latitude , and longitude .
Because of the different magnitudes of different variables, we normalize the deviation field of each variable by dividing x di by the corresponding local standard deviation s i according to x ′ i constitute the meteorological characteristic matrices of a certain category of cold hazards ( X ′ ). Similarly, x i and s i constitute matrices X and S , respectively. These three matrices vary with time, where the time includes all days when cold hazard events occurred. The schematic of the input data matrices is illustrated in Fig. 2.
The outputs of the six key meteorological variables coupling domain MV-EOF analysis are pairs of eigenvectors and principal components M 1 , p 1 , M 2 , p 2 , … , M r , p r , where r is the number of modes. It is noted that the eigenvectors (spatial patterns) are non-dimensional and lost their physical significance. To investigate the features and magnitudes of spatial mode more directly, we divide the time series by standard deviation and get p � 1 , p � 2 , … , p � r , and the spatial anomaly physical field for a specific MV-EOF (1) mode is formed by regressing the variable deviation field x ′ i onto the corresponding principal components (PC).
Thereby, for the tth day of cold hazard occurrence, the spatial distribution of six key meteorological variables can be expressed by: where X are the functions of variables, latitude, and longitude, p ′ j is only related to time. Among them, j = 1, 2, … , r represents the number of MV-EOF modes. The part of the X except MEAN represents the spatial-temporal evolution of coupling features of six key variables within events, which are represented by T 2m ′, T 2m _ anom ′, ∆T 24 ′, PREP′, SNOW′, and SNODP′, respectively.

Spatial distribution of means of surface air temperature-related variables during cold hazard events
To begin with, we investigate the spatial distribution of the means of the three temperaturerelated variables (i.e., T 2m , T 2m _ anom, and ∆T 24 ) during the three categories of cold hazard events in winters from 1980 to 2020 that are displayed in Fig. 3. We can observe several common features of the composite mean of T 2m and T 2m _ anom among all categories of cold hazards. Seen from Fig. 3a-c, the means of T 2m are mostly lower than 0 °C, except for that to the south of the Yangtze River. In addition, the means of T 2m decrease with increasing latitude, with the lowest values in Northeast China (below − 24 °C) but the highest values in southern part of Guangxi province, which is far above 0 °C (around 10-15 °C). This is consistent with the regional distribution of extreme cold events (Qi et al. 2019). The means of T 2m _ anom , however, are negative during all categories of cold hazards in most regions of China ( Fig. 3d-f), indicating less regional difference compared with means of T 2m . Exceptions are found in Northeast and Northwest China. The means of T 2m _ anom are close to or even above 0 °C in all categories of cold hazards in Northwest China and reach 1-4 °C during WET-C events in Northeast China. This indicates that significantly below-normal temperature is not a necessary condition for cold hazards to occur there. The daily temperature drop that is manifested by negative values of ∆T 24 is found in DRY-C events in most of China but looks to be an unnecessary condition during the remaining categories of cold hazards (particularly in Northeast and Northwest China, Fig. 3g-i). It is more interesting to see the remarkable differences in the means of the three temperature-related variables among different categories of cold hazards, which will be referred to as category-dependent differences hereafter and displayed in Fig. 4. The categorydependent differences in the means of T 2m and ∆T 24 tend to have larger magnitudes in the eastern part of China, which are in the range of − 8 ~ 8 °C and − 1.5 ~ 1.5 °C d −1 , respectively. Particularly, in the middle and lower Yangtze River region (particularly over Hunan, Hubei, Jiangxi, and Anhui provinces, etc.), the means of T 2m during DRY-C events are much higher than those during WET-C and BOTH-C events. This is probably because, to cause cold hazards with wintry precipitation in this region, where the climate is relatively warmer than the Northern part, T 2m has to be low enough to form sleet or snow and allow snow depth growth. In Northeast China, however, the means of T 2m tend to be lowest during BOTH-C events. This feature is most evident in Heilongjiang and Jilin provinces. This is probably because DRY-C events that occurred in this region are always in the early winter attributed to the disastrous temperature drop (Table A1 in Supplementary Information), and WET-C events there tend to involve warm moist air associated with cold fronts and Northeast China cyclones (Ren et al. 2016). In Western China, the mean values of T 2m increase from DRY-C to BOTH-C, and to WET-C. The warmest condition for WET-C events might be due to the fact that during the severe snow and freezing-rain event in the southwestern regions of China such as Sichuan and Yunnan provinces, the Western Pacific subtropical high and the southern branch trough always transport a large amount of warm and humid air into this region (Zhao et al. 2011). While in Northwestern China, such as Gansu and Xinjiang provinces, the strong cold air intrusion is always guided by the northwesterly wind in front of the ridge over the Ural Mountains during the cold wave process , consistent with the coldest condition during DRY-C events. The features of the spatial distribution of category-dependent differences of means of T 2m _ anom (Fig. 4d-f) are basically consistent with those of T 2m . As to the mean values of ∆T 24 , DRY-C events experience the largest negative ∆T 24 in most regions of China, while BOTH-C events experience the smallest negative ∆T 24, particularly in Eastern China (Fig. 4g-i). This indicates that the condition of abrupt temperature drop is important for causing DRY-C events but not necessary for causing BOTH-C events.

Probability distribution of surface air temperature-related variables during cold hazards in each province of China
Now, we have known that the means of the three temperature-related variables vary significantly with the category and location of cold hazards. It is necessary to take a closer look at the province-dependent temperature condition by examining the probability distribution functions (PDFs) of T 2m , T 2m _ anom, and ∆T 24 for each province or municipality during days when each category of cold hazards occur. The kurtosis coefficients of PDFs of T 2m and ∆T 24 (right panels in Fig. 5a-c and g-i) are positive almost in all the provinces or municipalities. This indicates that the distribution of T 2m and ∆T 24 during each category of cold hazards tends to be more concentrated than the normal distribution. The kurtoses of T 2m _ anom (right panels in Fig. 5d-f) are smaller, which indicates that the probability distributions of T 2m _ anom are more scattered when the background temperature with the clear meridional gradient has been removed. The kurtoses of T 2m _ anom for provinces such as Inner Mongolia are negative in all categories of cold hazards, making it harder to find the certain range of T 2m _ anom indicative of the occurrence of cold hazards in these regions. Besides, the kurtosis coefficients of PDFs of T 2m , T 2m _ anom, and ∆T 24 are all largest in WET-C events, indicating that the ranges of the three temperature variables tend to be narrower and thus more certain during WET-C events than the other two categories of cold hazards.
The fact that most of the provinces or municipalities have a more concentrated probability distribution of T 2m allows us to further sort the PDFs of provinces or municipalities based on the T 2m range with the maximum values of PDF. The sorting rule is that the higher the T 2m range of peak PDF is, the higher the ranking will be. After the sorting process, we plot the sorted PDFs of T 2m by shadings in Fig. 5a-c and the maps of ranking numbers are shown in Fig. 6a-c. It is seen that the province-dependent changes in the T 2m range with the peak PDF can be as large as 30 °C during DRY-C and BOTH-C events but 10-15 °C during WET-C events. This is consistent with the results of previous case studies, stating that there exists a large difference in T 2m in different regions during low temperature and abrupt temperature drop events (Zhang et al. 2019;Li et al. 2021;Sheng et al. 2021) but a small difference in T 2m during sleet hazards (Mao and Li, 2015;He and Shao 2011;Zhao et al. 2011;Peng et al. 2012). The T 2m PDF ranking numbers of all categories of cold hazard events increase northward (Fig. 6a-c), indicating that the T 2m with the maximum occurrence probability during cold hazards tends to be lower in the northern part of China than the southern part.
Considering that the province-dependent T 2m ranges with the peak PDF during cold hazards might be strongly dominated by the climatic state, we further remove the climatological PDF and obtain the PDF anomaly during cold hazards (contours in Fig. 5a-c). The PDF anomaly during all the three categories of cold hazards exhibits positive values in the colder T 2m ranges but negative values in the warmer T 2m ranges. In other words, the anomalous occurrence probability shifts toward colder temperature ranges, which might be responsible for inducing hazards. An exception can be found in Northeast China Table 1 Main features of six key meteorological variables from the perspective of the event-mean fields and the spatial-temporal evolution within events for DRY-C, WET-C, and BOTH-C events in the middle and lower Yangtze River region Meteorological vari- (particularly Heilongjiang and Jilin provinces), where the T 2m PDF anomaly is close to zero, and these provinces tend to rank at the back during all three categories of cold hazards. This is consistent with Figs. 3c, 4b, c, suggesting that the cold climatological condition plays a dominant role in the cold hazards occurring. Even positive T 2m PDF anomalies in the warmer ranges can be found during DRY-C events in regions such as Anhui, Jiangxi, and Hunan provinces, thus confirming that T 2m is generally higher during DRY-C events in the middle and lower Yangtze River region. Past studies also show that T 2m was in the range of 5-10 °C during low-temperature hazards (Wang et al. 2008a, b), but below 0 °C during sleet and snowstorm hazards (Mao and Li, 2015) in this region. Distinct from T 2m , the sorted province-dependent PDFs of T 2m _ anom (Fig. 5d-f) exhibit little difference among different provinces, manifested by the almost vertical structure of the T 2m _ anom ranges of large PDF values. The inter-province changes in the T 2m _ anom range with maximum PDF are only about 5 °C for WET-C and BOTH-C events and 10 °C for DRY-C events, which is also impressive though smaller compared to T 2m . The ranking number of T 2m _ anom is shown in Fig. 6d-f does not show the latitude-dependent features as can be seen from that of T 2m . This indicates that besides the latitudinal changes of temperature determined by the solar radiation and solar zenith angle, the difference in the T 2m _ anom during cold hazards among provinces may be related to other factors, such as  Richardson et al. 2018). The T 2m _ anom PDF anomaly during cold hazards (contours in Fig. 5d-f) shows consistent characteristics with that of T 2m (contours in Fig. 5a-c), namely positive/negative values of PDF anomaly tend to be in the negative/ positive T 2m _ anom ranges. Comparing the results of T 2m _ anom with those of T 2m , we see that for different regions, the importance of temperature conditions based on T 2m and T 2m _ anom can be different (Fig. 6a-f). For instance, in the middle and lower Yangtze River region, the ranking of the T 2m _ anom is further back than that of T 2m , which indicates that the anomalously cold condition is more closely related to cold hazards in this region. Contrarily in Northeast China, the ranking of T 2m _ anom is in the front than that of T 2m , confirming the much closer relationship of the climatological cold condition than the anomalous cold condition with the cold hazards occurring in the northeast.
The sorted province-dependent PDFs of ∆T 24 (Fig. 5g-i) also exhibit anomalous probability shifts toward negative values during DRY-C events in most regions of China except Gansu, Xinjiang, and Shaanxi provinces, etc. The provinces in the latitude band 28-36°N (Central China) ranks at the back (Fig. 6g). This indicates the condition of large daily temperature drop is particularly important for DRY-C events in these regions. During WET-C and BOTH-C events, however, the province-dependent PDFs of ∆T 24 concentrate around 0, and no anomalous probability shifts during WET-C and BOTH-C events. Fig. 6 The maps of the ranking number of T 2m , T 2m_anom and ∆T 24 for each province during days of a, d, g DRY-C, b, e, h WET-C, and c, f, i BOTH-C events

Regional multivariate spatiotemporal coupling features
Besides temperature conditions, precipitation, snowfall, and snow depth are also typical meteorological variables closely related to the occurrence of cold hazards. It is necessary to investigate the multivariate spatiotemporal coupling features during cold hazards. MV-EOF analysis has been conducted for cold hazards that occurred in a selected region, namely the middle and lower Yangtze River Region composed of Hunan, Hubei, Jiangxi, Jiangsu, and Anhui provinces. This region is selected because these provinces often experienced cold hazards at the same time according to the hazard information list (Table A1 in Supplementary Information) in the winters from 1980 to 2019, thus applicable for the MV-EOF analysis. More importantly, the region exhibits large differences in temperature conditions among different hazard categories, but in the meanwhile, the kurtosis coefficients of T 2m and T 2m_anom in these provinces are relatively large, inferring that meteorological conditions such as temperature can be one of the indicators for cold hazards despite other social and economic factors. Therefore, we expect that the potential of predicting cold hazards based on meteorological conditions corresponding to a specific category of cold hazard is relatively high for the middle and lower Yangtze River region.
According to Table A1, in the middle and lower Yangtze River region, we find only one DRY-C event, two WET-C events, and 11 BOTH-C events during the past 40 years. This suggests that BOTH-C events are the most frequently occurring category of cold hazards in the middle and lower Yangtze River region, with many related case studies (Zhang et al. 2019;Yao et al. 2012;Zuo et al. 2017;Mao and Li 2015;Ding et al. 2008). The DRY-C event lasted 16 days, and the main subgroup of hazards includes disastrous abrupt temperature drop and low temperature. The two WET-C events lasted 5 days and occurred in mid-to-late January. The main subgroup of hazards was a snowstorm. The 11 BOTH-C events covered a total of 67 days with an average duration of 6 days. It is worth noting that the 4 BOTH-C events that occurred in December 2010 were written as one piece of record from the official disaster annual report and literature but composed of several rounds of cold surges. Next, we will investigate both the event-mean state and spatial-temporal coupled evolution in the progression of the cold hazard event of three temperature-related variables (T 2m , T 2m _ anom , ∆T 24 ) and three precipitation-related variables (PREP, SNOW, SNODP) for DRY-C event, WET-C events, and BOTH-C events, respectively.

DRY-C
The event-mean fields of the six key meteorological variables (i.e., T 2m , T 2m _ anom , ∆T 24 , PREP, SNOW, and SNODP) during the DRY-C event are shown in Fig. 7a-f. It is seen from Fig. 7b that the generally negative T 2m_anom is in the range of − 5 ~ − 2 °C and tends to have larger amplitudes toward the northwest. In the meanwhile, the entire region Fig. 7 Spatial patterns of the event-mean-field of the six key variables during a-f DRY-C, g-l WET-C, and m-r BOTH-C events in the middle and lower Yangtze River region. The six key variables are the daily mean temperature and temperature anomaly (units: °C), 24-h temperature change (units: °C d −1 ), accumulated daily precipitation (units: mm d −1 ), accumulated daily snowfall (units: mm d −1 ) and daily mean snow depth (units: mm hr −1 ). (The white dots indicate the stations passing the 95% significance level by student's t-test) experienced uniform ΔT 24 at about − 1 °C/day (Fig. 7c). Nevertheless, the T 2m was mostly higher than 8 °C due to the relatively warm climate in this region (Fig. 7a). Accompanied by the anomalous cold and significant temperature drop was plenty of rainfall, with maximums in the southeast of the region (Fig. 7d). Snowfall, however, was small and sporadic, and the snow depth was close to 0, consistent with the nature of DRY-C events.
To investigate the temporal-spatial coupled evolutions of six key meteorological variables within the event, we remove the event-mean values from the total fields at each day during the cold hazard events, and then, we conduct MV-EOF analysis on the standardized meteorological variable fields, as introduced in Sect. 2. The first leading MV-EOF mode (MV-EOF1) accounts for 47.9% of the total variance, while the total variance explained by the second leading MV-EOF mode decreases to 18%. Therefore, we only focus on the MV-EOF1 mode that represents the dominant covariations of six key meteorological variables. The spatial patterns of the leading MV-EOF (Fig. 8a-f) are characterized by the prevailing negative anomalies of temperature-related variables (i.e., T 2m ′, T 2m _ anom ′ and ∆T 24 ′) and positive PREP′ and SNOW′, while SNODP′ is close to zero. Larger magnitudes of negative T 2m ′ and T 2m_anom ′ are mainly in Hunan, Hubei, and Anhui provinces. The magnitudes of negative ∆T 24 ′ decreased northeastward, reflecting the larger daily temperature drop or increase in the southwest part of the region. The spatial distributions of the three precipitation-related variables in MV-EOF1 were in general agreement with their event-mean fields. The magnitudes of spatial-temporal evolution of the six variables were comparable to or even larger than the event-mean fields. Therefore, it is necessary to consider the spatial-temporal evolution of coupling features of six key variables within the event.
The event can be divided into two stages according to the corresponding PC1 (Fig. 9a): (i) negative PC1 in the early stage (day 1-10) with negative peaks on day 8 ~ 9 and (ii) positive PC1 in the later stage (day 11-16) with a positive peak on day 16. In the early stage, T 2m ′, T 2m _ anom ′, and ∆T 24 ′ were positive, while PREP′, SNOW′, and SNODP′ were negative. This indicates that accompanied by the relatively warmer temperature condition, the precipitation was also smaller in the early stage of the DRY-C event. In the later stage, T 2m ′, T 2m _ anom ′, and ∆T 24 ′ were negative, while PREP' and SNOW' were positive. This reflects the abrupt drop in temperature and relatively colder conditions, accompanied by the increase in rainfall in the eastern part and a slight increase in snowfall in the northeastern part of the middle and lower Yangtze River region in the later stage of the DRY-C event.
In brief, this DRY-C event that occurred in the middle and lower Yangtze River region was accompanied by a relatively warm but below-normal temperature condition associated with rain. During the progression of this event, T 2m ′ and T 2m _ anom ′ had abrupt drops accompanied by the nonuniform increase in rain and snow.

WET-C
Seen from the event-mean fields of meteorological variables shown in Fig. 7g-l, a much colder event-mean condition is accompanied by WET-C events than DRY-C events (Fig. 7g-h vs. a-b). In contrast to the above-0 °C T 2m averaged over the DRY-C event, ‾T 2m fell below 0 °C in the northern part (Jiangsu, Anhui, and Hubei). T 2m_anom reached − 5 °C in the western rims of Hubei and Hunan provinces. However, the negative values of ΔT 24 (Fig. 7i) were smaller than those during the DRY-C event (Fig. 7c). The three precipitation-related variables had similar spatial distribution, namely minimums in the south of Hunan and Jiangxi but maximums in Hubei and Anhui (Fig. 7j-l). SNOW and SNODP 1 3 Fig. 8 Same as Fig. 7, but for the spatial patterns of the first MV-EOF mode of the six key variables during a-f DRY-C, g-l WET-C, and m-r BOTH-C events. (The variance contribution is: 47.9, 42.1 and 42.1%, respectively) during WET-C events were significantly greater than those during DRY-C events, indicating the dominant contributions to the occurrence of WET-C hazards from a large amount of snowfall and its accumulation at the surface.
The first MV-EOF mode accounts for 42.1% of the total variance within the progression of WET-C events (Fig. 8g-l), but the magnitudes of T 2m ′ and T 2m _ anom ′ in MV-EOF1 mode are evidently smaller than the event-mean fields. Besides, PC1 exhibited distinct evolution between the two WET-C events (Fig. 9b), which indicates less consistency in the changes of the six meteorological variables within the process. These results suggest that there are no remarkable changes in the progression during the WET-C events, thus not discussed in detail here. Therefore, the WET-C events that occurred in the middle and lower Yangtze River region were accompanied by colder event-mean temperature conditions with a weaker temperature drop and significantly greater SNOW and SNODP than the DRY-C events, while the changes within the WET-C events are small and neglectable.

BOTH-C
According to the event-mean fields of meteorological variables associated with the 11 BOTH-C events (Fig. 7m-n), T 2m was several degrees above 0 °C, while T 2m_anom was negative. Both of them basically passed the 95% significance level by Student's t-test in the entire region. The magnitudes of negative T 2m_anom were relatively smaller than those during DRY-C and WET-C events. Region-wide temperature drop, manifested by the event-mean negative values ∆T 24 in the entire region, is also found (Fig. 7o). But this is Fig. 9 The PC1 of the first MV-EOF mode of the six key variables during a DRY-C, b WET-C, and c BOTH-C events, where the black vertical line separates different events. Red/blue bars indicating positive/ negative values. The black numbers on the abscissa represent the year of the occurrence of cold hazard events (e.g., "91" denotes year 1991 and "10" denotes year 2010), and the brown numbers represent the month and day of occurrence (e.g., "1226" means December 26) statistically insignificant, probably due to the fast nature of the temperature drop process. SNOW and SNODP were significantly larger, in the range of 0.5 ~ 2 mm/day and 1 ~ 3 mm/ hour, covering almost the entire region ( Fig. 7q-r). PREP was statistically significant only in the southeast corner, where liquid precipitation was dominant corresponding to the higher temperature (Fig. 7p). Despite that, the three precipitation-related variables had smaller amplitudes during the BOTH-C events than the WET-C events, but large values were more widely distributed.
There is remarkable coupled spatial-temporal evolution of the six meteorological variables during the progression of BOTH-C events, which can be represented by the spatial pattern of MV-EOF1 mode and its corresponding PC1. It is seen from Fig. 8m that the T 2m ′ in MV-EOF1 is uniformly negative in the entire region, in the range from − 6 to − 4 °C, which was comparable to the T 2m (Fig. 7m). The T 2m _ anom ′ and ∆T 24 ′ were also negative with magnitudes more than twice of the event-mean fields (Fig. 8n-o), indicating large changes during the progression of the events. The positive values of SNOW' and SNODP' covered the entire region, as their event means (Fig. 8q-r vs. Figures 7q-r). The largest variations of SNOW' during the progression of the event occur in the middle of the region, while the region of largest variations of SNODP' exhibits a slight shift toward the north, probably due to the lower temperature in the north. The PREP' in MV-EOF1 (Fig. 8p), however, shows close-to-zero values in most areas of this region, with relatively large magnitudes in the southeast corner only, indicating smaller variations in precipitation during the progression of events.
It is interesting to note that, most BOTH-C events in the middle and lower Yangtze River region occurred after 2000, especially after 2010 (Fig. 9c), which may be related to global warming (Ding et al. 2008;Wang et al. 2008a, b;Dai et al. 2022). The PC1 (Fig. 9c) was always negative in the early stage but positive in the later stage during most of the 11 BOTH-C events, despite the slight differences in the duration of the two stages from case to case. This feature is further confirmed by Fig. 10, which displays the rearranged time series of PC1 during each event and their composite means. In the early stage with negative values of PC1, T 2m ′, T 2m_anom ′ and ∆T 24 ′ were positive, while SNOW' and SNODP' were negative. This indicates that accompanied by the warm temperatures and less snowfall was less snow depth. In the later stage with positive values of PC1, opposite patterns of T 2m ′, T 2m_anom ′, ∆T 24 ′, SNOW′, and SNODP′ occurred, indicating that the arrival of cold air in the entire region brings about a significant increase in snowfall and the accumulation of snow depth. Further, we divided these 11 events into two categories, one dominated by positive PC1 values (including the first 3 events and the 2014 event) and the other one dominated by negative values (like the cases in 2010 and 2020). Their composite means (curves in Fig. 10b) show that the increasing trend of PC1 can be observed in both categories, despite that the latter category is preceded by warmer condition, while the former category is preceded by colder condition. Combining the MV-EOF1 fields with the eventmean fields, we find that in the later stage, T 2m was basically below 0 °C and T 2m_anom could exceed − 6 °C, while ∆T 24 was only about − 3 °C. PREP was larger in the southeast corner at about 8 mm. SNOW can reach 3 mm in the entire region, exceeding the level of moderate snow. The SNODP could increase to about 4 mm, slightly smaller than that in WET-C events. With the weak negative T 2m _ anom , three precipitation-related variables had a regionwide and uniform distribution during BOTH-C events. Above all, though T 2m_anom , ΔT 24 , PREP , and SNOW were not the largest among three categories of cold hazards during BOTH-C events, but SNOW covers the entire region and the SNODP was region-wide large due to the lowerT 2m . During the progression of this event, T 2m ′ and T 2m_anom ′ were significantly decreasing associated with an increasingly significant temperature drop per 24 h. With the gradual increase in SNOW', SNODP' was effectively accumulated.
In summary, in the middle and lower Yangtze River region, both the event-mean fields and the spatial-temporal evolution of coupling within the period of cold hazard events are quite different among the three categories. The major differences are listed in Table 1.

Discussion
This study demonstrates the large region-and category-dependent differences in temperature conditions in winter for DRY-C, WET-C, and BOTH-C cold hazard events in China. The multivariate meteorological conditions in the middle and lower Yangtze River region put forward by this study further reveal both the event-mean state and spatial-temporal coupling evolution during the progression of events for the temperature-related variables and precipitation-related variables. One way to utilize these meteorological conditions for detecting a high probability of occurrence of cold hazard is as follows. One can first standardize the daily meteorological variable fields following Eq. (3) and Eq. (5) using the x i ( , ) and s i ( , ) derived from days when cold hazard of specific category occurred in history. Then, a day of high probability of occurrence of cold hazard is detected if the following two criteria are both satisfied: (i) the daily standardized temperature-related variables are all negative and the precipitation-related variables are all positive; and (ii) the spatial correlation coefficients between the daily fields of the six key meteorological variables and the corresponding spatial patterns of MV-EOF1 pass the 95% significance level. Our primary validation in the middle and lower Yangtze River region yields that 63% of BOTH-C events, which are the most frequent in this region, can be successfully detected.
It should be admitted that this study only focuses on the meteorological conditions that are possibly responsible for the occurrence of cold hazards. Whether the cold events can cause disasters or not is also determined by social, economic development, and other non-meteorological factors (Liu et al.2020;Hu et al. 2021;Xu et al. 2016;Gao et al. 2021), which is necessary to consider in future research. Therefore, only using the meteorological condition could lead to relatively large false alarm rate in detecting the occurrence of cold hazard. Previous studies have shown that extreme cold days show a decreasing trend associated with global warming (Zuo et al. 2022;Kong 2020;Huang and Chen 2014;Kanno and Iwasaki 2020;Heo et al. 2018). But there are also plenty of studies showing that extreme cold days have increased locally linked with warm Arctic episodes and the amplification of atmospheric planetary waves (Screen and Simmonds 2014;Cohen et al. 2018;Johnson et al. 2018;Yu et al. 2022). So how will these three categories of cold hazard events change in the future? And will its characteristics change? It is of great value to use the future projection of meteorological variables in the multi-model CMIP6 dataset and the region-and category-dependent cold hazard detection method introduced in this study to evaluate the possibility of cold hazards occurring under the background of global warming, from the meteorological perspective.

Conclusions
Based on the daily gridded dataset of winter cold hazards in China from November 1980 to March 2020 derived by Yu et al. (2022), this study divides cold hazard events into three categories: DRY-C events (cold hazards without wintry precipitation including hazardous low temperature, abrupt temperature drop and freezing), WET-C events (cold hazards with wintry precipitation only including hazardous sleet and snowstorm), and BOTH-C events (both cold hazards without and with precipitation occur). And the disaster-caused capability of DRY-C events is relatively low than those of WET-C and BOTH-C events. Using the hourly ERA5 reanalysis dataset, we investigate the regiondependent meteorological conditions during each category of cold hazards in China.
The main features of temperature conditions include daily mean 2-m temperature (T 2m ), anomalies of daily mean 2-m temperature (T 2m _ anom ), and 24-h temperature change (∆T 24 ) for different provinces of China, and different categories of cold hazards have been the first found.
(1) The T 2m during cold hazards decrease with increasing latitude with the difference among provinces exceeding 30 °C. Below 0 °C temperature is not a necessary condition for the occurrence of cold hazards. The province-dependent changes in the T 2m can be as large as 30 °C during DRY-C and BOTH-C events but only 10-15 °C during WET-C events. The differences in T 2m for the same province among different categories 1 3 of cold hazards can exceed 10 °C. The kurtosis coefficients of T 2m are almost positive in the province, which indicates that a specific range of T 2m is a good indicator of cold hazards. Meanwhile, the anomalous occurrence probability shifts toward colder temperature ranges. (2) The T 2m _ anom tends to be negative during cold hazards. It exhibits less regional difference, which is up to 5 °C. The T 2m _ anom is mainly in the range of − 7 ~ − 2 °C, − 6 ~ 4 °C, − 4 ~ 2 °C, respectively, for DRY-C, WET-C, and BOTH-C events and the category-dependent differences of T 2m _ anom can be as large as 6 °C. The kurtosis of T 2m _ anom is generally smaller, indicating that the T 2m _ anom is more scattered than the T 2m during cold hazards. (3) The regional and category-dependent differences of means of ∆T 24 are about 2 °C d −1 . DRY-C events experienced the largest negative ∆T 24 in most regions of China, indicating the condition of large daily temperature drop is particularly important for DRY-C events in these regions. But during WET-C and BOTH-C events, no significant temperature drop can be observed.
Further combining the precipitation-related variables (i.e., daily accumulated total precipitation (PREP), daily accumulated snowfall (SNOW), and daily mean snow depth (SNODP)) and using MV-EOF analysis, we investigated the spatial-temporal coupled evolution of the key meteorological variables in the middle and lower Yangtze River region. BOTH-C events are the most frequently occurring category of cold hazards in this region, while there was only 1 DRY-C event and 2 WET-C events. Event-mean and spatial-temporal evolution in the progressing of events of the meteorological variables is separately examined, and details are summarized in Table 1. Briefly speaking, T 2m and T 2m _ anom had the highest event-mean values and largest spatial-temporal changes during the progression of DRY-C events but had the lowest event-mean values with little change during WET-C events. The daily temperature drop was also the largest and region-wide during the DRY-C event. During BOTH-C events, the event-mean temperature drop was relatively weak compared to the DRY-C event, but the changes during the progression of events are the largest. During WET-C events, however, the temperature drop was only in a small part of the region. The event-mean of precipitation-related variables was the largest during WET-C events, but their spatial-temporal evolution during events shared no common features. Both event-mean and spatial-temporal evolution of PREP was large during the DRY-C event, but event-mean and spatial-temporal evolution of SNOW and SNODP were large and region-wide during BOTH-C events.
In order to alleviate or prevent the adverse effects of cold hazards, it is of great use to enhance the resilience of infrastructure, increase public awareness, and financial support, supply of coal, electricity and oil, traffic diversion, and commodity supply by administrative departments such as departments of energy, transportation, and agriculture (Wei et al. 2009;Huang 2022;Wang 2022). Some of these initiatives need ample time to make plans and implement. This requires that the cold hazards should be forecasted in advance. Our results showing the region-and category-dependent meteorological conditions during three categories of cold hazards in China can be considered as quantitative standards for detecting a high probability of occurrence of cold hazards from the perspective of meteorological conditions. This would help improve the current cold hazard prevention system, which is of great significance to the economic and social society in China.