Long-term trends of precipitation and erosivity over Northeast China during 1961-2020

Wenting Wng , Shuiqing Yin , Jun Yu , Zeng He , Yun Xie

a College of Education for the Future, Beijing Normal University at Zhuhai, Zhuhai, 519087, China

b State Key Laboratory of Earth Surface Processes and Resource Ecology, Faculty of Geographical Science, Beijing Normal University,Beijing,100875,China

c College of Arts and Sciences, Beijing Normal University at Zhuhai, Zhuhai, 519087, China

d Zhixing College, Beijing Normal University at Zhuhai, Zhuhai, 519087, China

Keywords:Rainfall erosivity Climate change Storm characteristic Hourly precipitation Northeast China

A B S T R A C T Northeast China (NEC) is one of the vital commercial grain bases in China and it has suffered from soil erosion due to prolonged cultivation and lack of protection.To determine long-term trends of precipitation and rainfall erosivity over NEC during the latest decades, daily precipitation for the entire year during 1961-2020 and hourly precipitation for the warm season(May to September)during 1971-2020 were collected for 192 and 126 stations, respectively.Three seasons, including the cold season (October to April), early warm season (May to June), and late warm season (July to September) were divided according to the combination of precipitation and vegetation.Results demonstrate: (1) Daily precipitation reveals total precipitation and rainfall erosivity in the cold season and early warm season increase significantly at relative rates of 3.1% ～6.1% compared with the average during 1961-2020, and those in the late warm season decrease insignificantly.(2) Hourly precipitation reveals storms occurring in the early and late warm seasons have undergone significant increasing changes, which shift towards longer storm duration, larger amount, peak intensity, kinetic energy, and rainfall erosivity during 1971-2020.Moreover, the frequency of extreme storms increased.(3) Rainfall erosivities estimated from daily precipitation during 1971-2020 increase insignificantly for the early and late warm season, whereas those from hourly precipitation increase significantly (6.1% and 5.5%, respectively), which indicates daily precipitation may not be able to capture the trend fully under the warming background, and precipitation at higher resolutions than the daily scale is necessary to detect trends of rainfall erosivity more accurately.

1.Introduction

The global surface temperature has undergone a significant warming process over the past century(IPCC,2021).Along with the changing temperature, temporal and spatial variations of precipitation may be altered (Chen et al., 2021).Theoretically, the atmospheric water-holding capacity increases by 7% per degree of warming under the condition of constant relative humidity according to the Clapeyron-Clausius equation (Trenberth, 1998).Based on satellite observations,Wentz(2007)reported that global total precipitation increased at the same rate of 7% per degree of warming as the total atmospheric water.Meanwhile, increase in the frequency and magnitude of heavy and extreme precipitation events have been reported in many regions around the world(Brown et al., 2020; Groisman et al., 2005; Lenderink & van Meijgaard, 2010; Shaw et al., 2011).

The warming climate has impacted the soil erosion process through many aspects, of which the most direct impact is the change in rainfall erosivity (Nearing, 2001).Rainfall erosivity represents the potential ability of rainfall and runoff to cause water erosion, which is one of the main factors in the widely used empirical soil erosion models USLE, RUSLE, and CSLE (Liu et al.,2002; Renard et al., 1997; Wischmeier & Smith, 1978).Annual average rainfall erosivity (the R-factor) is defined as the average annual sum of the event EI30,which is calculated as the product of the total kinetic energy,E,and the maximum 30-min intensity,I30,of an event(Wischmeier & Smith,1958).Heavy to extreme events tend to generate comparatively higher erosivity due to greater kinetic energy and higher peak precipitation intensity than light and moderate events (Johannsen et al., 2020; Nunes et al., 2014).As a consequence of the change in precipitation, particularly the increase in frequency and magnitude of heavy events, rainfall erosivity may increase and further raise the risk of regional soil erosion.Analysis of long-term trends of rainfall erosivity based on historical observations have shown diverse results, including increasing, decreasing, or insignificant changes depending on research areas and analysis periods (Qin et al., 2016; Wang, Yang,et al., 2022; Oliveira et al., 2013; Fiener et al., 2013).Most studies on future projections of precipitation generated by GCMs (Global Climate Models) have consistently indicated that the rainfall erosivity will increase under the scenarios of warming climate (Li et al., 2021; Panagos et al., 2022; Riquetti et al., 2020; Zhang et al., 2010).A recent study on global rainfall erosivity and soil erosion has shown globally averaged rainfall erosivity will increase by 27-34.3% and soil erosion will increase by 30-66% for 2070 under different emission scenarios,compared with the 2010 global rainfall erosivity and soil erosion baselines (Panagos et al., 2022).

Northeast China (NEC) is one of the most important food production and commercial grain bases in China,which occupies 8%of the total land areas in China and produces 20%of grain production(National Bureau of Statistics of China, 2013).The cropland of NEC has been suffering from water erosion since the mid-1650s due to human migration and cultivation.The soil erosion rate increased slowly before 1950 and accelerated much faster in recent several decades (Xie et al., 2019).At the same time, the forest cover declined (Mackenzie et al., 2018), which further strengthened the regional erosion risk.By comparing soil erosion induced by water in China based on national surveys during 1995-1996 and 2010-2012, Yue et al.(2016) reported that soil erosion rates in Northeast China and South China experienced the fastest increases among all regions of China.Wang et al., 2022 utilized the137Cs technique to detect the soil erosion rate in NEC and found that the mean erosion rate over the region is 22.2 mm (10a)-1, which implied that the black soil in NEC could be entirely eroded in just 113 years.However, soil erosion in NEC hasn't received much attention due to its relatively lower sediment discharge compared to other regions in China, such as the Yellow River basin.Soil erosion has led to the thinning of black topsoil and the reduction of soil organic matter content and the accelerated soil erosion may become a potential threat to China's food security.

Analysis based on observations during 1961-2000 revealed that NEC has experienced an increasing trend of extreme precipitation,and the increasing trend of extreme precipitation amount at the hourly scale was more significant than that at the daily scale(Zhang&Zhai,2011).It can be inferred that regional rainfall erosivity may experience an increasing process as well.However,Qin et al.(2016)reported a significant decreasing trend(p<0.05)of regional rainfall erosivity over NEC based on long-term daily precipitation through 1951-2010, while projections of future rainfall erosivity of NEC generated from six GCMs models suggested a general increasing trend by mid-21st century (Zhang et al., 2010).The difference in trends may result from different analysis periods and data resolution used among studies.For example,based on daily precipitation series during 1951-2010, Qin et al.(2016) reported a significant decreasing trend (p < 0.01) of rainfall erosivity over the Loess Plateau, however, Wang, Yang, et al.(2022) detected a significant increasing trend with both hourly precipitation and daily precipitation during 1971-2020.Sun et al.(2021) also pointed out that when the study period updated from the ending in 2000-2009 to 2018 with additional observations for recent years, the increased signal of extreme precipitation has been strengthened and became more detectable.Updating the data series to the latest decade is necessary for the trend detection of precipitation and rainfall erosivity occurring in NEC, as additional warming has occurred during this period (IPCC, 2021), which is important for making proper soil conservation policy in NEC.

Soil erosion is a complex process, which is influenced by multiple factors, including rainfall erosivity, soil, landform, land cover,and conservation measures.Changes in precipitation and rainfall erosivity in different seasons may have a different impact on soil erosion due to the combined effect of precipitation and vegetation coverage for relatively flat cropland in NEC(Liu et al.,2002;Renard et al.,1997;Wischmeier&Smith,1978).The combination of intraannual precipitation and vegetation coverage of NEC can be divided into three periods within a year: (1) snow and low vegetation coverage in the cold season (October to April); (2) relatively more precipitation amount than cold season while low vegetation coverage in the early warm season (May to June); (3) the most abundant precipitation amount and the highest vegetation coverage in the late warm season (July to September).Monthly,seasonal,and annual rainfall erosivity was accumulated from storm rainfall erosivity.Therefore, in addition to the trend of annual rainfall erosivity over NEC,changes in rainfall erosivity at seasonal,monthly,and storm event scales are worthy of attention,however,this has not been reported in previous studies.

To determine and quantify the changes of precipitation and rainfall erosivity over NEC at annual,seasonal,monthly,and storm event scales during the latest decades, long-term daily precipitation data during 1961-2020 and hourly precipitation data during 1971-2020 were collected to derive precipitation and erosivity indices, and their long-term trends were analyzed by using the combination of the MK test and field significance test.Results achieved in this study may provide a reference for scientific planning on regional soil and water conservation measures in NEC.

2.Data and method

2.1.Study area

The study area, NEC, locates in the northeastern part of China and covers 145 million ha of area,which spans about 15 degrees of latitude and 20 degrees of longitude (Fig.1; Liu et al., 2021).In terms of terrain, NEC comprises Songliao Plain in the central area and mountains on three sides,which are Great Khingan Mountains,Lesser Khingan Mountains,and Changbai Mountains,from the west to the east respectively.

Fig.1.The location of Northeast China and the distribution of stations used in this study: circles (triangles) represent stations with daily (hourly) rainfall data.

The climate of NEC is a temperate continental monsoon climate with long winters and short summers.Annual average daily temperature varies within stations among -4.1-11.2°C.Annual average precipitation ranges from 238 to 1078 mm,with an average of 557.6 mm and a standard deviation of 141.9 mm over 192 stations.Annual rainy days range from 60.6 to 158.5 days, with an average of 97.2 days.NEC is located in the northernmost part of China,which is a junction region of the Eastern Monsoon Area and the Northwest Arid Area of China (Wang et al., 2019; Zhao,1983).The impact of the monsoon system on precipitation gradually decays from the southeastern coast to the northwestern inland,resulting in a general southeast-northwest decreasing spatial pattern of precipitation (Fig.2a).Meanwhile, the impact of terrain on the precipitation generates a local variation of precipitation spatial pattern(Fig.2a).Precipitation of NEC is concentrated in the warm season from May to September.The ratio of precipitation occurring in the warm season accounts for 80-90%,and it generally increases from southeast to northwest (Fig.2b).By contrast, precipitation occurring in the cold season accounts for 10-20% of annual precipitation, and decreases from southeast to northwest(Fig.2c).

2.2.Data collection

Two datasets, daily precipitation data during 1961-2020 and hourly precipitation data during 1971-2020, were collected from the National Meteorological Information Center (NMIC) of the China Meteorological Administration(CMA),which had conducted basic quality control (Li et al., 2011).Daily data was observed by simple rain gauges.The record was made twice a day at 8:00 a.m.and 8:00 p.m.Beijing time with the unit of mm,and the sum of two observations was the daily precipitation.Daily data was available all year around and was applied to calculate monthly,seasonal,and annual total precipitation and rainy days, as well as to detect the corresponding long-term trends.Hourly data was available for the warm season from May to September,as it was recorded by siphon or tipping-bucket self-recording rain gauges,which did not operate in the cold season to avoid instrument failures due to the low temperature and snow.As hourly data can better reveal the intrastorm structure than daily data, and determine the kinetic energy and peak intensity more accurately than daily data, it was applied to calculate storm characteristics and rainfall erosivity indices at the event storm scale during the warm season.

Further data quality control was conducted to select stations with satisfied quality and long data series during 1961-2020 for daily data and during 1971-2020 for hourly data.The criteria of quality control for daily data were: (1) If the number of missing days equals or exceeds 7 days in one month, this month was defined as a missing month;(2)If the number of missing months of one year exceeds one month, this year was defined as a missing year;(3)Stations without any missing year during 1961-2020 was used in the analysis.Finally,192 meteorological stations with daily precipitation data were selected(Fig.1).The average percentage of missing days during 60 years over 192 stations was 0.05%.The selection criteria for hourly data (May to September) collected between 1971 and 2020 were as follows(Wang,Yang,et al.,2022):(1)Stations with complete missing data for any given month from May to September were excluded; (2) Stations with more than 25%missing hourly observations during the warm season for any given year were excluded;(3)Stations with more than 5%missing hourly observations for the entire data series during the warm season were excluded.After these criteria were applied,126 stations with hourly data were selected for analysis (Fig.1).The average percentage of missing hours during the warm season of 1971-2020 over 126 stations was 2.1%.Ten stations had several months with missing hours of 20-25%, and for all other stations, missing hours in all months during the warm season were less than 20%.Missing days and missing hours were estimated by linear interpolating precipitation values from preceding and following periods.

2.3.Precipitation indices

For daily data,rainy days with daily precipitation ≥0.1 mm were kept, then analyzed at monthly, seasonal, and annual scales,respectively.The accumulation of precipitation and the total number of rainy days for each month and throughout the year were calculated and the abbreviation was introduced in Table 1.According to Chinese daily precipitation classification, (China Meteorological Administration, 2003), daily precipitation data were classified into three categories by precipitation amount:light precipitation (daily precipitation amount <10 mm), moderate precipitation(10-25 mm),and intense to heavy precipitation(daily precipitation ≥25 mm).

Initially,the twelve months were divided into two study periods based on the seasonal hydrothermal combination:the cold season(October to April) and the warm season (May to September).Further, the warm season was divided into two sub-periods, the early warm season(May to June)and the late warm season(July to September), to examine differences in vegetation coverage and precipitation patterns among months.The seasonal precipitation amount and the number of rainy days were analyzed separately for the cold season, early warm season, and late warm season.

Fig.2.Spatial distributions of annual precipitation(a),ratio of warm season(May to Sep.)precipitation(b)and cold season(Oct.to April)precipitation(c)for Northeast China based on 192 stations with daily precipitation data.

Table 1 Definition of precipitation and erosivity indices at annual,seasonal, and storm scales.

Hourly precipitation data were used to analyze storm-scale characteristics and trends for the warm season due to the data availability.Continuous observed data series were divided into individual storms firstly by a minimum inter-event dry time of 6 h(Huff, 1967; Wischmeier, 1959).Then storm precipitation (mm),storm duration(h),storm intensity(mm h-1),and several erosionrelated indices, including storm energy, the maximum 30-min intensity, storm rainfall erosivity, and total rainfall erosivity were calculated,respectively, which were introduced in Table 1.

2.4.Rainfall erosivity indices

The storm rainfall erosivity index, EI30, is the product of storm rainfall kinetic energy (E)and the maximum 30-min intensity of a storm (I30) (Foster, 2004; Renard et al., 1997; Wischmeier, 1959).Theoretically, E, I30, and EI30require the pluviograph breakpoint rainfall data to be calculated.Considering the limited availability of a long series of breakpoint data and the relatively easier availability of hourly data, E, I30,and EI30were approximately estimated by hourly data first and then adjusted to equivalent values by breakpoint data (Yin et al., 2007).The unit rainfall energy per hour per unit area per unit rainfall depth for hourly data was calculated with the equation recommended in the RUSLE2 documentation (Foster,2004) as follows:

where emwas the kinetic energy for the mth hour, in which the rainfall intensity was considered uniform (MJ ha-1mm-1), and imwas the rainfall intensity for the mthhour (mm h-1).The total energy for an event storm (MJ ha-1) obtained with the hourly data was as follows:

in which,Pmwas the rainfall depth for the mthhour(mm)and n was the number of precipitation hours.Then E,I30,and EI30of all storms(including erosive and non-erosive storms defined by RUSLE2)during the warm season were approximately estimated from Eh,H60,and EH60by multiplying adjustment factors of 1.105,1.668,1.73 which were developed by Yin et al.(2007).The conversion factors were derived from 456 storms at five stations in eastern China,and their fitted accuracy was quite satisfactory, with R2values of 0.99,0.90, and 0.93 for E, I30, and EI30, respectively (Yin et al., 2007).In addition, storm rainfall erosivity for all years in the early (late)warm season was ranked from small to large for each station, and the 90th thresholds of storm rainfall erosivity were determined.Storms with rainfall erosivity greater than the 90th threshold were denoted as extreme storms.The frequency and average of storm rainfall erosivity for extreme storms were calculated (Table 1).

Hourly precipitation is only available during the warm season during the period 1971-2020, which is shorter than that of daily precipitation.To obtain the characteristics of rainfall erosivity during the cold season and the whole year, rainfall erosivity was estimated with the following equation based on daily precipitation as well(Xie et al., 2016):

where rdrepresents rainfall erosivity for a rainy day, j represents the month from 1 (January) to 12 (December), and pdrepresents daily rainfall (mm).By summing daily erosivity during the cold season, the early warm season, the late warm season, and the whole year, seasonal rainfall and annual erosivity can be obtained year by year (Table 1).

2.5.Data analysis

The rank-based Mann-Kendall method was applied to estimate the monotonic trend of the time series and corresponding trend significance for indices lists in Table 1 (Kendall,1949;Mann,1945;Wang et al.,2021;Xu et al.,2003).Since the auto-correlation in the time series can influence the MK test results (Yue & Wang, 2002),the Trend-Free Pre-Whitening (TFPW) method was adopted to remove the serial correlations before applying the MK test (Chu et al., 2010).The Pre-whitening time series were applied MK test following Kendall(1949)and Mann(1945)and the Z statistics were calculated.The null hypothesis is trend absence.If |Z|>1.645, the null hypothesis will be rejected at a 90% significant level; if |Z|>1.96,the null hypothesis will be rejected at a 95%significant level;and if |Z|>2.576, the null hypothesis will be rejected at a 99%significant level.It will be considered as an increasing trend if Z>0,a decreasing trend if Z < 0, and no trend while Z = 0.

Then the non-parametric Theil and Sen's approach (TSA) were applied to calculate the steepness estimator β to illustrate the tendency and its magnitude (Sen,1968; Theil,1992, pp.345-381;Xu et al., 2003),

in which,1

In addition, the relative tendency was also determined.By comparing T with the corresponding average value during the study period 1961-2020 for indices based on daily data, or 1971-2020 for indices based on hourly data,the relative tendency was defined as,

Following the steps above, the trend and its magnitude for storm characteristics of each station can be detected.

For a specific station,the indices in Table 1 were calculated on an annual basis,and the equations outlined in section 2.5 were utilized to evaluate the long-term trends.The trends for the entire study area of NEC were obtained as follows: (1) The annual value from each station was interpolated to cover the entire NEC region,using an ordinary kriging method with a resolution of 10 km,resulting in a corresponding value of each pixel;(2)The mean value of all pixels within the study area was calculated;(3)The temporal series of the specific index was obtained and analyzed for long-term trends using the MK test and Theil-Sen estimator.

Besides, the field significance test was conducted to assess the actual statistical significance of trends for the entire study region(Livezey & Chen,1983).There are a series of approaches that can detect the field significance test.Renard et al.(2008) proposed resampling-based bootstrap and False Discovery Rate(FDR),which are adequate and robust in detecting field significance.In this study,the resampling-based bootstrap procedure was applied to determine the field significance of the MK test for all indices (Douglas et al., 2000; Renard et al., 2008; Wang, Yang, et al., 2022).Taking one index calculated with daily rainfall from 192 stations as an example, the original data set, referred to as S0, is comprised of records from all 192 stations over 60 years.The following steps were taken: (1) 60 years are randomly selected with replacement from years in the original data set S0, and records from the corresponding years for each station are used to form a new data set,S1;(2)A statistical test is performed on each station for data set S1,and the number of stations with significant results is recorded as N(1);(3) Steps (1) and (2) are repeated 60 times to form additional bootstrapped data sets, S2, S3, …, Sm, resulting in N(1), N(2), …N(m); (4) The field significance is determined based on the distribution of N(i)(i=1,…,m)at a specified level of significance,α',by calculating the critical value corresponding to the 1-α' quantile of the N(i) values.The filed significance can be determined by comparing the ratio of stations that show significant trends with the critical value (1-α').This approach provides a measure of the overall significance of the trends in the data,considering the spatial distribution of the trends across the field.

3.Results

3.1.Change in precipitation and rainfall erosivity revealed by daily data during 1961-2020

3.1.1.Annual precipitation

Trends of Pannualfor stations located in the southwestern and northeastern parts of NEC are opposite(Fig.3a).Stations located in the northeastern part show increasing trends consistently,whereas those in the southwestern part show decreasing trends.However,trends of Pannualfor most stations of NEC are insignificant, which results in an insignificant increasing trend for the entire region(Table 2).In contrast,RDannualshows a significant decreasing trend over the study area with a magnitude of-1.6 day(10a)-1(Fig.3e),and the ratio of stations with significant trends reaches up to 41.7%of all stations (Fig.4e).

Daily precipitation of NEC was further divided into three categories to distinguish the contribution of different rainy days to the reduction of RDannual:(1)rainy days with light rainfall(daily rainfall<10 mm), (2) rainy days with moderate rainfall (daily rainfall between 10 and 25 mm),and rainy days with intense to heavy rainfall(daily rainfall ≥25 mm).Stations showing significant decreasing trends occupy nearly half the proportion of total stations for rainy days with light rainfall, whereas stations with significant decreasing trends take up 2.6% and 1.0% for rainy days with moderate and intense to heavy rainfall(Fig.4e).It can be concluded that the decreasing trend of annual rainy days mainly resulted from the reduction of rainy days with light rainfall.

3.1.2.Seasonal precipitation

Precipitation for cold season, PCS, shows increasing trends for most stations in NEC (Fig.3b).Stations with significant increasing trends are mainly located in the northeastern NEC with high latitude (Fig.3b).At the regional scale, PCSincreased with a relative rate of 4.4% (10a)-1compared with the average during the study period.Moreover, both the light and moderate precipitation increased significantly (Fig.4b, Table 2).For RDCS, stations with negative and positive trends are distributed scattered across the study area and most stations are with insignificant trends (Fig.3f;Fig.4f).However, rainy days with moderate precipitation(10-25 mm) increased significantly(Fig.4f).

Precipitation occurring in the early warm season(PMJ)accounts for 24.3%of annual precipitation and 29.0%of the total precipitation of the warm season.Stations across the study area show consistent increasing trends for PMJ(Fig.3c).Ratio of stations with significant increasing trends mostly spreads over central and northern NEC.Stations located in northern NEC changed much obviously and with greater magnitude than southern stations.At the regional scale,PMJincreased significantly at a rate of 4.0 mm (10a)-1.The significant increasing trend detected for PMJis mainly due to the increase in moderate and intense to heavy precipitation(daily precipitation ≥10 mm)(Table 2;Fig.4c).For RDMJ,no significant change has been observed except for several stations located in the central NEC(Fig.3g).However,rainy days with moderate and intense to heavy precipitation in the early warm season increased significantly at relative rates of 2.9% and 8.4% (10a)-1, respectively(Fig.4g).

Contrary to the significant increasing trend found for PMJand insignificant increasing trend for RDMJin the early warm season,an insignificant decreasing trend and significant decreasing trend have been detected for the precipitation amount, PJAS, and rainy days, RDJAS, in the late warm season, respectively at rates of-1.7 mm(10a)-1and-0.6 day(10a)-1(Fig.3d&h,Table 2).For RDJAS, 55.7% of stations are showing significant downward trends,which distribute across the whole study area(Figs.3h and 4h).The downward of PJASand RDJASmainly results from the decrease of light precipitation (Fig.4d & h).The ratio of decreasing stations accounts for 31.3% and 62.5%, respectively for light precipitation amounts and corresponding rainy days.

Fig.3.The 10-year trends of annual and seasonal precipitation (a)-(d), and annual and seasonal rainy days (e)-(h).Circles indicate insignificant trends and triangle (inversetriangle) indicates significant increasing (decreasing) trends at the p = 0.05 significance level.

Table 2 The 10-year trends (Trend) and relative trends (RT, %) for six precipitation indexes of the Northeast region during 1961-2020, and the ratio of stations show significant increasing and decreasing trends at the p= 0.5 significant.

Fig.4.The histogram represents the ratio of stations showing significant increasing and decreasing trends to all stations at the p = 0.05 significant level for six seasonal indices.

3.1.3.Monthly precipitation

The temporal distribution of precipitation around the year is influenced by the monsoon climate.Monthly precipitation and rainy days of NEC show obvious annual cycles.Precipitation and rainy days mainly concentrate in the warm season through May to September, while the dry season last from October to April coinciding with the cold season.More than 80%of precipitation and 60%of rainy days occur from May to September,and the rest occur from October to April.

By separating the study period into two climate states,1961-1990 and 1991-2020, obvious changes can be seen (Fig.5).The main difference exists in May and July as the average monthly precipitation of May over all stations increase from 44.4 mm to 53.2 mm, and that of July decreased from 154.5 mm to 141.9 mm.Compared with 1961-1990, rainy days for most months in 1991-2020 show consistent downward except for May and December.Average annual rainy days during 1991-2020 reduce by 7.56 days compared with that during 1961-1990.Particularly, the average rainy days in July, August, and September in 1991-2020 were reduced by 2.1,1.6, and 1.9 days, respectively.

Table 3 summarizes long-term trends of monthly precipitation and rainy days revealed by the MK test and field significance test.Monthly precipitation from October to June show increasing trends.Among these months, precipitation in March, May, June,November, and December increased significantly.Meanwhile,precipitation occurring from July to September show decreasing trends,and the trend in July is significant.Rainy days in July,August,and September decreased significantly, while the rest months changed insignificantly.

3.1.4.Annual and seasonal rainfall erosivity

Trends found in annual and seasonal rainfall erosivity are consistent with trends found in precipitation (Table 2, Table 4).Pannualand PJASfor most stations changed insignificantly,and trends in Rannualand RJAS-dfor most stations are not significant as well.For RCS-dand RMJ-d, which seasons are characterized by significant upward trends in seasonal precipitation, the seasonal erosivity increased significantly as a consequence.This is mainly because daily precipitation is the only variable involved in the daily erosivity estimation model (Eq.(3)).Therefore, the significant change in precipitation amount will lead to an increase in rainfall erosivity directly.

3.2.Change in storm characteristics and rainfall erosivity revealed by hourly data during 1971-2020

3.2.1.Storm precipitation characteristics

Spatial-temporal variation of precipitation within the study period was further analyzed in terms of storm characteristics and potential rainfall erosivity based on 126 stations with hourly precipitation.Basic information on storms occurring in two stages of the warm season and magnitudes of trends are shown in Table 5.In general, storms occurred in the late warm season have larger average storm characteristics and rainfall erosivity compared with those in the early warm season.

Table 5 and Fig.6 indicate all storm-related indices adopted in this study show consistent upward trends for both stages of the warm season except the number of storms for the late warm season.The average storm precipitation(P),duration(D),intensity(I), maximum 30-min intensity (I30), and rainfall energy (E) of storms occurring in early and late warm seasons increased significantly during the past five decades (Table 5).Compared with the average during 1971-2020, storm characteristics changed with relative rates of 1.3%-3.3%(10a)-1.What should be noticed is that,compared with the mean intensity(I)during storms,the maximum 30-min intensity (I30) shows a higher increasing rate, with more stations showing significant increasing trends.The ratio of stations showing a significant increase for the late warm season is more than that of the early warm season.Indices derived from hourly data during two stages of the warm season show a similar spatial pattern in trends, which for the late warm season are displayed in Fig.7, and that of the early warm is not shown in the article.

The exceedance probability density of storm precipitation and storm EI30for the warm season was depicted separately for two periods,1971-1995 and 1996-2020.The comparison reveals that the probability density of storms with heavier precipitation in the second period (1996-2020) is higher than that in the first period(1971-1995), and the difference between the two periods gets larger as storm precipitation increases (Fig.8).It can be inferred that the heavy to extreme storms during the warm season occurs more frequently.

3.2.2.Rainfall erosivity

Based on hourly data, rainfall erosivity at the event storm scale can be estimated (Table 5, Fig.6).For both stages in the warm season, significant upward trends found for I30and E result in significant increasing changes in rainfall erosivity at the storm scale(EI30).Moreover, two extreme erosivity indices, (EI30)90thfreand(EI30)90thave, which presents the frequency and the average erosivity of storms with extreme erosivity that exceeds the 90th threshold,increased significantly for both two stages as well, indicating that storms with extreme rainfall erosivity tend to occur more frequently and with heavier rainfall erosivity than before.

Seasonal rainfall erosivity in the early and late warm season(REI30) were also estimated by accumulating EI30during the early and late warm seasons from 1971 to 2020,respectively.REI30values during both stages in the warm season increase significantly, with relative rates of 6.1% (10a)-1and 5.5% (10a)-1, respectively(Table 5).Stations with significant increasing trends mainly distribute in the central and eastern NEC (Fig.7i).

3.3.Comparison of trends in rainfall erosivity by hourly data with daily data

Fig.5.Comparison of the intra-annual variation of monthly precipitation (a) and rainy days (b) during 1961-1990 and 1991-2020.

Table 3 The 10-year trends (Trend)and relative trends (RT, %) for monthly rainfall amount (rainy days) of Northeastern China during 1961-2020, and the ratio of stations showing significant increasing and decreasing trends at the p= 0.05 significant level.

Table 4 Trends in rainfall erosivity for the whole year (Rannual)and three seasons, the cold season (RCS), the early warm season (RMJ), and the late warm season (RJAS).

Table 5 Basic statistics for storm-based rainfall and erosivity indices, and its 10-year trends (Trend) and relative trends (RT) for the NEC at the p = 0.05 significant level during 1971-2020.

Rainfall erosivity is defined as the product of kinetic energy and maximum 30-min intensity during a storm, using breakpoint precipitation data or fine interval data less than 5 min to calculate.Coarser temporal resolution data,such as daily data,can be used to estimate rainfall erosivity by empirical models and determine its temporal trend in the absence of precipitation data with fine resolution.When using the daily erosive model(such as Eq.(3)in this study), the estimated rainfall erosivity is mainly influenced by the accumulation of daily precipitation amount.However, changes in the instant rainfall intensity, rainfall energy, and peak rainfall intensity,which cannot be captured by daily precipitation data,may play predominant roles in the change of rainfall erosivity.

To explore the influence of data resolution on the trend of rainfall erosivity, two methods were used to estimate seasonal rainfall erosivity from 1971 to 2020: (1)using hourly precipitation and Eqs.(1),(2)and(2)using daily precipitation and Eq.(3).Results show that rainfall erosivities by hourly data for both stages of warm seasons increase significantly while those by daily data increase insignificant (Tables 4 and 5).The ratio of stations showing significant trends by hourly data was higher than those by daily data.To provide a further comparison,Fig.9 depicts inter-annual variations and trends of seasonal rainfall erosivity over NEC using two methods.The fitted trend line of REI30for the early warm season shows a slightly steeper slope than that of RMJ-d,while that for the late warm season shows a much steeper trend than that of RJAS-d (Fig.9).

Fig.6.The histogram represents the ratio of stations showing significant increasing trends for the early(late)warm season at the p =0.05 significant level for nine storm characteristics and rainfall erosivity relating indices during 1971-2020.The significant decreasing ratio was omitted and not plotted in the figure because the percentage for indices are zero or nearly zero.

4.Discussion

4.1.Possible reasons for the change in seasonal precipitation

Precipitation for the cold season (Pcs) increases significantly,especially for the precipitation in December and rainy days with moderate precipitation (10-25 mm, Fig.4 and Table 2).Precipitation for most stations within the study area showed consistently increasing trends, and the more northerly sites are located, the higher trends are.Results achieved here broadly confirm the conclusion reported by Wang and He (2012) that winter precipitation (from December to February) from 1986 to 2010 is 20%higher than that of 1951-1985.As pointed out by Wang and He(2012), the weakening of the East Asian Winter Monsoon(EAWM) since the 1980s may be the direct driver that led to the increase of precipitation in the cold season of NEC.As the EAWM was weakened, the water vapor transports from the southern coastal marine area of Northeast Asia,and the west side of NEC was strengthened in the cold season which provides an increasing water vapor condition for precipitation.The convergence in the lower atmosphere from the west, south, and east sides of the NEC was intensified, and due to the weakening of the EAWM, the divergence in the upper layers was intensified as well.Meanwhile,the rise of surface sea temperature (SST) along the coast of Northeast Asia may lead to more water vapor content in the air over NEC (Wang & He, 2012).

During the past six decades, both annual rainy days (RDannual)and rainy days in the late warm season (RDJAS) have decreased significantly, and it was found that decreasing trends of RDannualand RDJASwere mainly attributed to the decrease of rainy days with light rainfall (Fig.4, Table 2).This is consistent with the result achieved in the previous study(Qian et al.,2007).Qian et al.(2007)reported an overall decreasing trend in the frequency of light rainfall in China, which was accompanied by a regional warming process.The increasing nighttime temperature and daily mean temperature led to the warmer air being harder to reach dew-point temperature than normal, which further resulted in a decrease in light rainfall (Qian et al., 2007).

Fig.7.The 10-year trend of storm characteristics and rainfall erosivity for storms occurring in the late warm season (July to September) during 1971-2020.Circles indicate insignificant trends and triangle (inverse-triangle) indicates significant increasing (decreasing) trends at the p = 0.05 significance level.

Fig.8.The probability density function of exceedance for (a)storm precipitation and (b) storm rainfall erosivity at station #54076 in Shulan during two time periods:1971-1995 and 1996-2020.

Fig.9.The inter-annual variation and trend of seasonal rainfall erosivity during the early(a)and late(b)warm season derived from daily precipitation and daily erosive model(Rd)and hourly precipitation and EI30 (REI30) during 1971-2020.

4.2.Influence of data resolution on the trend of rainfall erosivity

The significance and rates of trends for seasonal rainfall erosivity by hourly data and daily data during 1971-2020 are different.The difference may be from two aspects: (1) Daily precipitation data cannot describe the instant rainfall intensity,rainfall energy, and peak rainfall intensity as well as hourly data, which decreases the accuracy in the estimation of rainfall erosivity.(2)Responses of daily and hourly precipitation to the warming may differ.While the Clausius-Clapeyron equation predicts an increase in atmospheric water content and total precipitation of 7% per degree Celsius with warming (Wentz et al., 2007), Lenderink and Van Meijgaard (2008) have shown that the response of extreme hourly precipitation to temperature change for temperatures above 10°C exceeds this rate, indicating a faster increase than that predicted by the Clausius-Clapeyron equation.Storms occurring through the late warm season have undergone significant changes,which shift towards less number of total events, larger storm precipitation, longer storm duration, and heavier storm intensity(Table 5).Moreover, the frequency of extreme storms increases(Fig.8), and storms with extreme rainfall erosivity occur more frequently and with heavier rainfall erosivity than before, which results in a significant increase in rainfall erosivity estimated from hourly data.These findings suggest that rainfall erosivity estimated from daily precipitation and the daily erosive model may not be able to capture the trend fully under the warming background.Therefore, precipitation data at higher resolutions than daily scale are necessary to detect trends of rainfall erosivity more accurately.

4.3.Implication of changing rainfall erosivity on regional soil erosion

The identified changes in RCS-dsuggest that NEC is experiencing an increase in regional soil erosion risk during the cold season.Soil erosion induced by snow-melt runoff occupies a considerable fraction of total annual soil erosion in mid-high latitude regions(Ollesch et al., 2005; Wu et al., 2018).NEC is the second largest stable snow cover area in China due to its high-latitude geographical location.Also, snowfall is the main precipitation form of NEC during the cold season.For the northernmost part of NEC, the snowing period is as long as 190 days (Pan, 2022).After snowfall reaches the ground surface, it is redistributed by strong wind and accumulates in gullies, furrows of farmland, and forests,which will turn into the main source of snow-melt runoff in the following Spring (Tang et al., 2021).Snow melts rapidly as the temperature increases in spring,and the snowmelt runoff will lead to water-induced soil erosion.At the same time,the surface cover is relatively low,especially for cropland.The increasing precipitation during the cold season will increase the snowmelt runoff in the following spring, which may create a greater risk of regional soil erosion for bare cropland as a result.

Among factors that have impacts on the regional soil erosion process,rainfall erosivity,and surface coverage are the main factors that vary with the season during the year in NEC.Rainfall erosivity describes the potential capacity of rainfall and runoff that can cause soil loss,which is decided by storm characteristics.Precipitation in NEC begins to increase from May and rainfall erosivity increase at the same time.Vegetation cover can protect surface soil from the detachment of raindrops,and greatly reduce the occurrence of soil erosion.The growing season of the crop in NEC starts after 140 days of the year(mid-May)(Tang et al.,2015;Zhang et al.,2017).Hence,the coverage of cropland in the early warm season is relatively low,whereas rainfall erosivity during this period was getting higher,which may result in high erosion risk on the surface soil in the early warm season, especially from several extreme storm events.Results achieved in this study confirm that total rainfall erosivity and erosivity from extreme events during the early warm season had increased significantly during the past five decades, which may further increase the regional soil erosion risk in this period.

During the late warm season,although the coverage of cropland is the highest during the whole year,the precipitation amount and the rainfall erosivity are the highest as well.Considering that rainfall erosivity from extreme storms during the late warm season is getting higher, the corresponding soil and water conservation measurements should be further enhanced for preventing soil erosion from these extreme storms correspondingly.

5.Conclusions

Northeast China(NEC)is one of the most important commercial grain bases in China, which plays an important strategic role in ensuring China's food security.Long-term daily precipitation data from 192 stations during 1961-2020 and hourly precipitation data from 126 stations during 1971-2020 distributed across NEC were analyzed to determine long-term trends of precipitation and rainfall erosivity over NEC at different temporal scales.Three seasons,including the cold season(October to April),the early warm season(May to June), and the late warm season (July to September) were divided according to the combination of the precipitation and vegetation coverage among months.The following conclusions can be drawn:

(1) Daily precipitation data during 1961-2020 reveals that no significant change has been detected for annual precipitation and rainfall erosivity in NEC, but the intra-annual distribution has been altered.Significant increasing trends have been found for seasonal precipitation and rainfall erosivity in the cold season and the early warm season at relative rates within the range of 3.1-6.1% (10a)-1, respectively, which mainly resulted from the increase of rainy days with moderate to intense precipitation.Total precipitation and rainfall erosivity in the late warm season decrease insignificantly,with seasonal rainy days showing a significant decreasing trend of -1.5% (10a)-1, which is mainly related to the decrease of rainy days with light precipitation.

(2) Hourly precipitation data during 1971-2020 reveals storms occurring in the early and late warm seasons have undergone significant increasing changes, which shift towards longer event duration,larger event amount,average intensity,peak intensity,kinetic energy,and rainfall erosivity.Moreover,the frequency of extreme storms increases.

(3) Trends of seasonal rainfall erosivity estimated by daily precipitation data and hourly precipitation data for 1971-2020 are not identical.Rainfall erosivities estimated from daily precipitation during 1971-2020 increased insignificantly at rates of 4.2% and 0.9% for the early and late warm season,respectively, whereas those estimated from hourly precipitation during 1971-2020 increase significantly at rates of 6.1% and 5.5%, respectively.These findings suggest that rainfall erosivity estimated from daily precipitation may not be able to capture the trend fully under the warming background,and precipitation data at higher resolutions than the daily scale are necessary to detect trends of rainfall erosivity more accurately.

Data availability

Gauge daily and hourly precipitation data were provided by the National Meteorological Information Center(CMA).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgment

This work was supported by the National Key Research and Development Program of China (Grant No.2021YFE0113800), the National Key Research and Development Program of China (Grant No.2021YFD1500705),and the Project for Recruited Talents to Start Up Their Work and Research in Beijing Normal University at Zhuhai(310432116).