Volume 49, Issue 19 e2022GL100277
Research Letter
Free Access

Inconsistent Frequency Trends Between Hourly and Daily Precipitation During Warm Season in Mainland of China

Miao Lei

Miao Lei

Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, College of Atmospheric Sciences, Lanzhou University, Lanzhou, China

Contribution: Methodology, Software, Validation, Formal analysis, ​Investigation, Resources, Data curation, Writing - original draft, Visualization

Search for more papers by this author
Jiming Li

Corresponding Author

Jiming Li

Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, College of Atmospheric Sciences, Lanzhou University, Lanzhou, China

Correspondence to:

J. Li,

[email protected]

Contribution: Conceptualization, Formal analysis, Resources, Writing - review & editing, Supervision, Project administration, Funding acquisition

Search for more papers by this author
Lijie Zhang

Lijie Zhang

Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, College of Atmospheric Sciences, Lanzhou University, Lanzhou, China

Contribution: Software, Data curation

Search for more papers by this author
Cong Deng

Cong Deng

Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, College of Atmospheric Sciences, Lanzhou University, Lanzhou, China

Contribution: ​Investigation, Data curation

Search for more papers by this author
Yarong Li

Yarong Li

Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, College of Atmospheric Sciences, Lanzhou University, Lanzhou, China

Contribution: ​Investigation, Data curation

Search for more papers by this author
Jianjun He

Jianjun He

State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences, Beijing, China

Contribution: Data curation

Search for more papers by this author
First published: 29 September 2022
Citations: 2

Abstract

Precipitation change is determined by changes in frequency and intensity. Based on hourly and daily precipitation data sets in China (1961–2014) and CMIP6 model, this study found that observed inconsistency in the frequency trend at the two time resolutions (hourly and daily) reflected precipitation becoming more concentrated at 13.4% of stations, resulting in the specific precipitation phenomenon of “more hours in one day, but fewer days.” However, models exhibited opposite precipitation phenomenon at 16.5% of stations and cannot reproduce the widespread increase in the proportion of extreme precipitation amounts (PA) in the total PA, especially at an hourly resolution. Although stations with significant trend from CMIP6 approximately twice the observation, both CMIP6 and observations showed that frequency and intensity changes dominated the total PA trend at hourly and daily time resolutions, respectively. These results demonstrated that daily precipitation observations could not fully capture the important features of short-duration rainfall.

Key Points

  • Observed inconsistent frequency trend between daily and hourly precipitation causes “more hours in 1 day, but fewer days” rainfall feature

  • Models show opposite rainfall feature and miss broad increase of extreme precipitation amount (PA) rate in total PA, especially hourly scale

  • Both CMIP6 and observations show frequency and intensity changes dominated the total PA trend at hourly and daily time scales, respectively

Plain Language Summary

Do the hourly precipitation frequency and intensity have trends consistent with those of daily precipitation? Based on hourly and daily rain gauge data sets in China (1961–2014) during May–September (warm season), we found that the daily precipitation frequency widely decreased over past decades, but the hourly precipitation frequency did not decrease significantly in most regions. Such a difference led to the phenomenon of “more hours in 1 day, but fewer days” of warm season precipitation. The contribution calculation further revealed that precipitation observed at a coarse time resolution (e.g., daily time scale) could not sufficiently capture the important features of short-duration rainfall. Compared with the observations, however, the CMIP6 models simulated a significant decreasing trend in hourly precipitation frequency at many sites and thus exhibited the opposite precipitation phenomenon. Moreover, models cannot reproduce the widespread increase in the proportion of extreme precipitation amount (PA) in total PA, especially at an hourly resolution. However, the models exhibited good consistency with the observations in terms of the dominant factors of the PA trend or long-term change, although the percentages of stations were obviously biased by the models, especially for the trend contribution.

1 Introduction

As a key process of the hydrologic cycle, precipitation has always been considered a critical indicator in climate change studies. Since its long-term trend can greatly impact regional water supply and natural ecosystems, it has received wide attention in recent years (H. Zhang & Zhai, 2011; J. Sun & Ao, 2013). According to the atmospheric energy balance constraints or the Clausius–Clapeyron equation, global extreme precipitation is expected to increase due to the enhanced humidity holding capacity of the atmosphere under climate warming (O’gorman & Schneider, 2009). The changes in precipitation, which are determined by frequency and intensity, exhibit obvious regional characteristics (Lu et al., 20142016) that could cause distinct natural disasters in different climate regions and bring great challenges to agricultural production and socioeconomic systems (Q. Zhang et al., 2012). For example, due to low vegetation coverage over arid and semi-arid regions, heavy precipitation in these regions could easily trigger floods and debris flows (J. Peng et al., 2015). Excessive precipitation in humid regions could result in serious urban waterlogging and threaten human lives and economies (Zhai et al., 2005). To reduce the risks of extreme precipitation induced by climate change, the space-time variations in precipitation amount, frequency and intensity over different climate regions should be further evaluated.

For a station, whether its precipitation change is determined by the change in frequency or intensity is closely linked to the time resolution of precipitation. By using daily precipitation data sets, some studies have indicated that the long-term changes in daily extreme precipitation at most stations in China are mainly determined by frequency (Lu et al., 2014), but the change in total precipitation at most stations is determined by precipitation intensity (Lu et al., 2016). However, given the same daily precipitation amount, precipitation could occur with longer (shorter) hours but weaker (stronger) intensity, which means that the findings based on daily precipitation data could not be directly applied to precipitation at an hourly time resolution (Trenberth et al., 2003; Westra et al., 2014), and some heavy hourly precipitation could be misidentified as normal precipitation at a daily time resolution. Some studies have focused on the relationship between hourly extreme precipitation intensity and temperature (Guo et al., 2020; Xiao et al., 2016) or hourly precipitation trends in the warm season (H. Zhang & Zhai, 2011; Qin et al., 2021); nonetheless, it remains unclear whether the changes in the total (or extreme) precipitation amount at hourly and daily time resolutions are determined by the same dominant factor (e.g., frequency or intensity change), especially over different climate regions of China.

To address the abovementioned knowledge gap, this study aims to answer the following question: are the contributions of frequency and intensity changes to the observed precipitation trend (and long-term change) consistent at different time resolutions and over different climate regions? In addition, as the most important tool for projecting future climate change, some studies have shown that general circulation models always underestimate heavy precipitation intensity and overestimate the weak precipitation frequency at daily or lower time resolutions (Chen & Dai, 2019; Y. Sun et al., 2006), but the evaluation of simulated precipitation at higher time resolutions over China has received less attention. Therefore, it is also necessary to evaluate the performance of CMIP6 models in reproducing hourly precipitation characteristics and to further determine whether the precipitation intensity or frequency bias in CMIP6 models will result in inconsistent contributions of frequency and intensity changes with that of observations, particularly for extreme precipitation.

2 Data and Methodology

2.1 Precipitation Data Set and Division of Climate Regions

The hourly precipitation data of 2,455 meteorological observation stations in China (1961–2014) were obtained from the China Meteorological Administration. The hourly precipitation data mainly came from the digital product of the self-recording precipitation paper before automatic meteorological observations were established. Since precipitation mainly occurs in the warm season, we mainly selected precipitation events from May to September (warm season) for further analysis. The stations with shorter precipitation records (e.g., <11 years) were excluded. The precipitation data with an annual missing rate <10% were retained for the analysis. Finally, 2,087 stations passed the quality control. Note that due to the limited sensitivity of surface observation instruments, the minimum observation recording value of surface precipitation is 0.1 mm/hr (or day) (e.g., Guo et al., 2019; Li et al., 2016; Qin et al., 2021; Vergara-Temprado et al., 2021). This observational limitation possibly will cause a little uncertainty in our results but will not change the main conclusions of this study. In addition, because precipitation data came from cumulative values during the related time period, drizzle (precipitation amount <0.1 mm/hr) may be ignored by the hourly precipitation observations but can be observed in the daily precipitation, so the daily precipitation data were derived from the surface daily precipitation data set (V3.0) in the same period.

Until now, only five CMIP6 models with historical or AMIP simulations may provide hourly and daily precipitation outputs from 1961 to 2014, the details of these CMIP6 models were shown in Table S1 in Supporting Information S1. To facilitate comparison between observation and model, CMIP6 models outputted gridded precipitation data sets with different horizontal resolutions but the same period observations were interpolated to the station through the inverse distance weighting method. In addition, precipitation frequency and intensity were sensitive to thresholds that identify wet hours or wet days (Trenberth & Zhang, 2018). For consistency, 0.1 mm/hr was used in both observations and CMIP6 as the threshold to define wet day/hour, and the same threshold used in the model and observations has been proven to have a weak effect on the results (Chen et al., 2021). In the following analysis, only the precipitation characteristics from the multimode ensemble mean (MEM) were provided.

To classify the climate region of the stations, we used the aridity index (AI) with a spatial resolution of 0.5° × 0.5° provide by Feng and Fu (2013), which is the ratio of annual precipitation to annual potential evapotranspiration. By bilinear interpolating the gridded AI index to the site, all stations were further divided into arid (0.05 < AI < 0.2), semiarid (0.2 ≤ AI < 0.5), semi-humid (0.5 ≤ AI < 0.65) and humid (AI > 0.65) regions (Feng & Fu, 2013). Due to the very few stations in arid regions, arid and semiarid regions were combined as arid regions for further analysis.

2.2 Threshold and Trend Test

Determination of the threshold for extreme precipitation is one of the key steps in the investigation because varying thresholds can lead to different extreme precipitation trends (Schär et al., 2016) and affect the response of extreme precipitation to global warming (Pendergrass, 2018). Due to the complex precipitation characteristics in China, the threshold of extreme precipitation should consider the local climate conditions. In this paper, the cutoff scale was selected as the threshold of extreme precipitation. The probability density distribution of daily precipitation Pp is usually fitted using a gamma distribution (Cho et al., 2004) through the following formula:
urn:x-wiley:00948276:media:grl64905:grl64905-math-0001(1)
where the scale parameter urn:x-wiley:00948276:media:grl64905:grl64905-math-0002 in the gamma distribution can be used as the cutoff scale, and urn:x-wiley:00948276:media:grl64905:grl64905-math-0003 controls the rate of decay of the distribution in the power-law range, and P represent temporally averaged precipitation (P ≥ 0.1 mm). The probability density decreases sharply with increasing precipitation intensity and drops after the cutoff scale. This finding means that the cutoff scale controls the tail characteristics of probability density. Recent studies have shown that sub-daily precipitation can also be described using the gamma distribution (Kaptué et al., 2015; Martinez-Villalobos & Neelin, 2019). The multivariate linear regression of the binned probabilities (used to estimate urn:x-wiley:00948276:media:grl64905:grl64905-math-0004 urn:x-wiley:00948276:media:grl64905:grl64905-math-0005) has a small dependence on the binning scheme (Martinez-Villalobos & Neelin, 2019), and urn:x-wiley:00948276:media:grl64905:grl64905-math-0006 is proportional to the moment ratio urn:x-wiley:00948276:media:grl64905:grl64905-math-0007 (urn:x-wiley:00948276:media:grl64905:grl64905-math-0008 urn:x-wiley:00948276:media:grl64905:grl64905-math-0009 is the sequence of precipitation intensity, i = 1,…,N). Following recent studies (Chang et al., 2020; Martinez-Villalobos & Neelin, 2019), we used urn:x-wiley:00948276:media:grl64905:grl64905-math-0010 to replace the cutoff scale urn:x-wiley:00948276:media:grl64905:grl64905-math-0011 as the threshold for extreme precipitation due to its simplicity.

Figure S1 in Supporting Information S1 shows the distribution of the hourly (daily) threshold value and the corresponding percentile of wet hours (days) for the observation and MEM. For the observation (Figure S1a in Supporting Information S1), the pattern is similar to the 99.9th percentile of wet hours in 1981–2013 (Luo et al., 2016). Overall, the threshold gradually trended upward from northwest to southeast, and the two extreme value centers were located on the North China Plain and Hainan Island. In particular, we also tested the effectiveness of selection method of extreme precipitation threshold (see the text and Figure S2 in Supporting Information S1). The threshold in CMIP6 is far below that in the observations (see Figures S1d and S1h in Supporting Information S1), which is caused by the “drizzling” bias (high precipitation frequency but low intensity) (Dai, 2006).

The Mann-Kendall nonparametric trend test (Mann, 1945) was applied to identify statistically significant trends in the time series of all and extreme precipitation events. Sen's slope (Sen, 1968) was used to quantify the trend. The trend is statistically significant if it passes the 95% significance test; otherwise, it is nonsignificant. Because the scattered distribution of station trends and the time duration might be inconsistent, we evaluated the significance of trends using a field significance resampling-based procedure (Q. Sun et al., 2021; Westra et al., 2013). The time series of precipitation events at the stations were disrupted and resampled 1000 times. We generated the percentage distribution of sites with a significant trend to determine the 2.5%–97.5% confidence levels of the null hypothesis without a trend and examined whether the proportion of stations with a significant trend from the actual observations was statistically different from the null hypothesis without a trend.

2.3 Contribution Rate

To study the contributions of frequency (PF) and intensity (PI) to the precipitation amount (PA) trend within a given region, multiple linear regression between the PA anomaly and the PF and PI anomalies needs to be established. Anomalies are defined as follows:
urn:x-wiley:00948276:media:grl64905:grl64905-math-0012(2)
where k is the year, urn:x-wiley:00948276:media:grl64905:grl64905-math-0013 represents the P value of the kth year, urn:x-wiley:00948276:media:grl64905:grl64905-math-0014 represents the mean P value of all years, and urn:x-wiley:00948276:media:grl64905:grl64905-math-0015 represents the anomaly. The regression model can be expressed as follows (Jian et al., 2018; Lu et al., 2016):
urn:x-wiley:00948276:media:grl64905:grl64905-math-0016(3)
where all variables are subjected to normalized procedures (mean = 0, standard deviation = 1), ΔP, ΔF and ΔI represent the anomalies of amount, frequency and intensity, respectively, and C is a constant term. Based on the regression model, we calculated the relative contribution of variables (e.g., Ci) to the significant trend of PA using the following formula:
urn:x-wiley:00948276:media:grl64905:grl64905-math-0017(4)
where urn:x-wiley:00948276:media:grl64905:grl64905-math-0018 is the regression coefficient and urn:x-wiley:00948276:media:grl64905:grl64905-math-0019 is the difference in average PA (PF or PI) between the first and last 10 years.
The contribution rate of long-term change is calculated by the following formula (Huang & Yi, 1991):
urn:x-wiley:00948276:media:grl64905:grl64905-math-0020(5)
where k is the length of the data time series, n is the number of variables in the regression model, urn:x-wiley:00948276:media:grl64905:grl64905-math-0021, urn:x-wiley:00948276:media:grl64905:grl64905-math-0022 represents the regression coefficient from (Equation 3) and urn:x-wiley:00948276:media:grl64905:grl64905-math-0023 is the variable in the regression model.

3 Results and Discussion

Figure 1a shows the spatial distribution of the climatological mean of warm season precipitation. The annual warm season PA increases from less than 200 mm/year in the northwest inland area to approximately 1,500 mm/year on the southern coast of China. This result is similar to the annual average precipitation distribution in China (Li et al., 2016; Ma et al., 2015). The hourly precipitation from MEM basically reproduced the spatial distribution of observed precipitation but obviously overestimated the hourly precipitation amount (HPA) over most areas, especially in Southwest China and South China (Figure 1c). In Figure 1b, HPA exhibited significant increasing trends (>60 mm/decade) at some of the observation sites over the lower reaches of the Yangtze River, Northeast China and the Shanxi Province and decreasing trends over some observation sites of the Yunnan-Guizhou Plateau. Some sites in East China exhibited increasing PA, which might be related to the weakening of the East Asian summer monsoon (Wu et al., 2016). Different from the observations, however, MEM simulated the widespread decreasing trend of HPA over the Shanxi, Gansu and Sichuan Provinces (Figure 1d). The daily precipitation amount (DPA) showed a spatial distribution and trend similar to those of the HPA observations (Figures S3a and S3b in Supporting Information S1), but the models exhibited an increasing trend of DPA at more sites (proportion of stations is approximately 15%) over arid regions than those of the simulated HPA (approximately 7%), while they showed a decreasing trend of DPA at fewer sites over semi-humid/humid regions (Figure 1e and Figure S3e in Supporting Information S1). Overall, observations and models both indicated that a slightly wider wetting trend occurred in the arid/semi-arid regions, consistent with previous studies based on daily precipitation and different precipitation indices (Deng et al., 2014). D. Peng and Zhou (2017) suggested the close relationship of the wetting trend over arid/semi-arid regions to the convergence of water vapor flux and global warming-enhanced evaporation.

Details are in the caption following the image

The spatial distribution of hourly precipitation amount, hourly precipitation frequency, and hourly precipitation intensity during warm season from observations (a, f, k) and MME (c, h, m), and corresponding trend distribution from observations (b, g, l) and MME (d, i, n), (e, j, o) are the proportions of stations with significant increases (red) or significant decreases (blue), the dot represents the observation and the star represents MME.

For the hourly precipitation frequency (HPF), its distribution is similar to the annual average precipitation days (i.e., DPF) (Ma et al., 2015), whether for observation or MEM (also see Figures 1f and 1h, Figures S3f and S3h in Supporting Information S1). Although the MEM may reasonably describe the spatial distribution of HPF and DPF, due to the “drizzling” bias, the model always overestimated the precipitation frequency, and the HPF and DPF in Southwest China even exceeded 2,000 hr/year and 150 days/year, respectively. Interestingly, we found that the observed HPF only experienced an obvious decreasing trend at some sites on the Yunnan-Guizhou Plateau and an increasing trend in the Shanxi and Shandong Provinces (Figures 1g and 1j), but a widespread decreasing trend in the observed DPF can be found over Southwest and Northeast China (Figures S3g and S3j in Supporting Information S1). The inconsistent trends between the observed HPF and DPF over the past decades might indicate that precipitation has become more concentrated, resulting in an obvious decreasing trend in the number of precipitation days. Figures S4b and S4c in Supporting Information S1 clearly demonstrate that the trend of precipitation hours with one wet day over all climate regions experienced a significant and wider increase (reaching 0.3 hr/d/decade and percentages of stations with increasing trends >10% over three climate regions) in the study period. Thus, it formed the specific precipitation phenomenon of “more hours in 1 day, but fewer days” over 13.4% of stations. In contrast, however, MEM cannot reproduce this phenomenon and even simulated a broad and significant decreasing trend of precipitation hours with 1 day at 16.5% of stations which mainly over the Shanxi, Gansu and Sichuan Provinces and the southeast coast of China (Figures S4e and S4f in Supporting Information S1). This is principally because simulated HPF has a wide downward trend over these regions (percentages of stations with decreasing trends >12% over three climate regions), while the trend of the simulated DPF is not obvious over most areas (Figures 1i and 1j, Figures S3i and S3j in Supporting Information S1). In addition, the models simulated more precipitation hours with 1 day and failed to reproduce the observed maximum value centers of precipitation hours with 1 day (see Figures S4a and S4d in Supporting Information S1).

The distributions of mean hourly precipitation intensity (HPI) are displayed in Figure 1k. Overall, the HPI gradually decreased from southeast to northwest China, and the maximum and minimum values were approximately 4 mm/hr and 0.5 mm/hr over Hainan Island and Xinjiang, respectively (Figure 1k). The daily precipitation intensity (DPI) had a similar distribution but also showed a stronger intensity over central China in addition to South China (Figure S3k in Supporting Information S1). For the trend of precipitation intensity, we found that the observed HPI and DPI both experienced wider enhancement over humid regions, where the percentages of stations with increasing trends were approximately 10% (Figure 1l and 1o, Figures S3l and S3o in Supporting Information S1). This result may be due to the increase in anthropogenic aerosols in the humid area of East China inhibiting mild precipitation (Wang et al., 2016), which significantly enhanced precipitation intensity. In addition, the DPI exhibited wide and significant enhancement over North China, whereas the HPI had a decreasing trend over some sites of the Beijing and Tianjin regions. The hourly light precipitation frequency increased (Figure S7g in Supporting Information S1), and high pollution may restrain the HPI in this area (Jiang et al., 2021). For the models, however, the results indicated that MEM underestimated precipitation intensity and even could not reproduce the spatial distributions of mean values and trends for the observed HPI and DPI (Figure 1m and 1n, Figures S3m and S3n in Supporting Information S1). For example, strong hourly precipitation simulated by the MEM tended to occur in the vast eastern part of China, whereas the observed maximum values of the HPI were mainly located in South China. In contrast, MEM also missed the observed maximum values of the DPI over central China. In particular, we also found that the MEM obviously underestimated the extent of the increasing trend for the HPI and DPI over the humid region; in contrast, it showed higher percentages of stations with increasing trends of HPI and DPI over the arid region than over the humid region (Figure 1o and Figure S3o in Supporting Information S1).

The increase in HPI may have added the risk of short-term extreme precipitation, especially over southeastern and southern China. Here, we further examined the spatial distributions of the means and trends of warm season hourly extreme precipitation amounts (HEPA, Figures 2a–2d). Compared with the observations, the simulated HEPA showed a significant overestimation over all of China (Figure 2c), especially over the southern part of China. This finding is also reflected in previous studies on daily extreme precipitation (Xu et al., 2022) (see Figures S5a and S5c in Supporting Information S1). The observed HEPA exhibited a wide and significant increasing trend over Southeast China (Figure 2b). This result is related to the heating of the TP, which causes the South Asia high to strengthen and extend eastward, the West Pacific subtropical high to strengthen westward, and the southwest water transport in the middle and lower reaches of the Yangtze River to increase, thus increasing the extreme precipitation in this region (Ge et al., 2019; Ning et al., 2017). However, the MEM cannot reproduce the wide increasing trend of HEPA, especially over the humid region, and it even simulates a decreasing trend around the Qinling region (Figure 2d). Figures S5b and S5d in Supporting Information S1 indicate that the daily extreme precipitation amount (DEPA) has a similar trend distribution as HEPA both for the observations and models, but the proportion of stations with a significant increasing trend is relatively lower than that of HEPA over the humid region (Figure 2e and Figure S5e in Supporting Information S1). The observed extreme precipitation amount (EPA) exhibited a wider increasing trend than the total PA, indicating that the proportion of EPA in the total PA experienced a widespread increase over past decades, especially at the hourly observation resolution (see Figure S6c in Supporting Information S1). However, this feature was not captured by the MEM; thus, the model would underestimate the probability and risk of short-term extreme precipitation in the climate prediction (see Figures S6b and S6d in Supporting Information S1).

Details are in the caption following the image

Same as in Figure 1 but for hourly extreme precipitation.

The significant overestimation of HEPA and DEPA from the MEM was almost caused by the overestimation of hourly and daily extreme precipitation frequency (EPF) in the models, which is particularly obvious for the Yunnan-Guizhou Plateau (Figure 2h and Figure S5h in Supporting Information S1). However, the trend of EPF has a similar distribution to that of EPA at two time scales for both the observations and models (Figures 2f–2j and Figures S5f–S5j in Supporting Information S1). In addition, the decreasing trend of HPF was mainly concentrated in the Yunnan-Guizhou Plateau, but the trend of HEPF over this region was significant only at several stations. This finding means that hourly light or moderate precipitation over the Yunnan-Guizhou Plateau may have turned infrequent in the past decades (Figures S7g–S7i in Supporting Information S1). These statistical results are consistent with conclusions based on hourly precipitation in Li et al. (2016) (relative threshold) and Miao et al. (2016) (fixed threshold). Moreover, we also found that the observed significant and wide decreasing trend of DPF (Figure S3g in Supporting Information S1) was related to the significant decrease in light and moderate precipitation rather than daily extreme precipitation (Figures S7a–S7c and Figure S5g in Supporting Information S1). As a result, it would be helpful to investigate the cause of the observed reduction in hourly and daily light or moderate precipitation in these regions (Miao et al., 2016), which could adversely affect water and soil conservation (Piao et al., 2009) and crop growth. However, for MEM, Figures S7j–S7l in Supporting Information S1 and Figure 2i indicate that the decreasing trend of the simulated HPF over the Shanxi, Gansu and Sichuan Provinces is related to precipitation of all intensities, but the decreasing trend of the simulated HPF over the southeastern coast of China is totally unrelated to the HEPF in the models. For the hourly extreme precipitation intensity (HEPI) (Figure 2k) from the observations, the distribution is similar to the mean intensity during the hours exceeding the 99.9th percentile (Luo et al., 2016), indicating the applicability of the threshold method selected in this study. The low extreme precipitation threshold in the model resulted in the low HEPI in the model, and its spatial distribution (Figure 2m) was similar to that of the HPI. Compared with the HPI and DPI, the observed HEPI and DEPI were significantly enhanced over relatively few stations in the humid region (Figure 2o). For the models, the MEM obviously overestimated the percentage of stations with a significant increasing trend of extreme precipitation intensity over the arid and arid/semi-humid regions for hourly and daily precipitation, respectively (Figure 2o and Figure S5o in Supporting Information S1).

Using Equations 3 and 4, we further quantified the contributions of precipitation frequency and intensity to the total and extreme precipitation amount trends at hourly and daily time resolutions during warm season. When the contribution of frequency (or intensity) exceeds 50% at a given station, the precipitation trend of this station is dominated by frequency (or intensity). Finally, we calculated the percentage of stations by frequency-dominated and intensity-dominated trends over the three climate regions (see Figure 3). At the daily resolution, precipitation intensity change is always the dominant factor of the long-term trend for both the observed and simulated total PA over the three climatic regions, although the percentages of stations with intensity-dominated trends are higher for the models (see Figure 3a). In contrast, at the hourly resolution, precipitation frequency became the dominant factor affecting the observed PA trend except in the humid region, where the percentages of stations with frequency- and intensity-dominated trends were comparable with the observations (see Figure 3c). For the model, however, we found that precipitation frequency dominated the simulated PA trend mainly over the semi-humid and humid regions. For extreme precipitation, the observations and models both indicated that the change in frequency is always the dominant factor of the EPA trend regardless of hourly or daily precipitation (Figures 3b and 3d). The intensity of extreme precipitation is adequately strong and does not fluctuate easily (Lu et al., 2014), while the EPF is low, and small changes will have a great impact on EPA. In addition, frequency-dominated contributions can be detected by hourly extreme precipitation observations at more stations in humid regions and should receive more attention. For the long-term change in PA, Figure S8 in Supporting Information S1 also showed that the daily precipitation change is totally dominated by intensity change, whereas hourly precipitation change is mainly dominated by frequency change. The observed results are partly consistent with Lu et al. (2016), who used the daily precipitation data set to confirm that intensity is the dominant factor of the PA trend in China. Similarly, for extreme precipitation, frequency change totally dominated the long-term change in EPA regardless of whether the resolution was hourly or daily. In summary, the models were consistent with observation in the aspect of dominated factors of the PA trend or long-term change, although stations with significant trend from CMIP6 approximately twice the observation.

Details are in the caption following the image

(a, b, c and d) represent the percentages of stations dominated by frequency (blue) or intensity (gray) of daily precipitation amount, daily extreme precipitation amount, hourly precipitation amount, hourly extreme precipitation amounts, respectively (the trend of all stations passed the significance test).

4 Conclusions

This study utilized hourly and daily precipitation data sets during warm season to determine the factor that dominated the trends of observed and simulated total (and extreme) precipitation at sub-daily and daily time resolutions in mainland China. Observational results indicated that the DPF has widely decreased over the past decades, but the HPF has not decreased significantly in most regions. The inconsistent frequency trend at the two time resolutions led the precipitation to become more concentrated, thus forming a specific precipitation phenomenon of “more hours in 1 day, but fewer days” over most of the sites. However, due to CMIP6 models simulating an unrealistically wide decreasing trend of HPF over the Shanxi, Gansu, and Sichuan Provinces and southeast coast of China, the models exhibited an opposite precipitation phenomenon. Moreover, the models simulated more precipitation hours within 1 day and failed to reproduce the observed maximum value centers of precipitation hours within 1 day. In addition, the model also failed to reproduce the widespread increase in the proportion of EPA in the total PA, especially at an hourly resolution; therefore, the model would underestimate the probability and risk of short-term extreme precipitation in climate predictions. It is worth noting that above conclusions were almost derived from the stations with longer record time, and the selection of data length thus will not affect the results.

The observed increase in the proportion of EPA in the total PA may be related to the impacts of local water vapor, temperature, convective available potential energy or aerosols on the formation and intensity or frequency of hourly precipitation (e.g., Guo et al., 2019; Ng et al., 2022). Based on the contribution calculation, the CMIP6 model and observations both revealed that frequency and intensity changes dominated the total PA trend at hourly and daily time scales, respectively. However, changes in frequency always dominated the extreme PA trend. Thus, these results demonstrate two viewpoints: (a) precipitation with coarse temporal resolution (e.g., daily) could not fully capture or resolve some features of short-term rainfall frequency and intensity; and (b) models still exhibited good consistency with observations in the aspect of dominant factors of PA trend or long-term change even if they cannot reproduce the trend of total (or extreme) precipitation.

Acknowledgments

This work was jointly supported by the NSFC Major Project (41991231), National Science Fund for Excellent Young Scholars (42022037) and National Natural Science Foundation of China (41805028 and 41905003).

    Conflict of Interest

    The authors declare no conflicts of interest relevant to this study.

    Data Availability Statement

    The recent hourly rainfall data set can be downloaded from (http://data.cma.cn/en/?r=data/detail&dataCode=A.0012.0001). The daily precipitation data set derived from national operational rain gauges are not publicly available for legal/ethical reasons but are available to researchers subject to a non-disclosure agreement based on reasonable request. Data can be requested at http://data.cma.cn/. CMIP6 output is available from the Earth System Grid Federation (https://esgf-node.llnl.gov/search/cmip6/). CMIP6 models used can be found in Table S1 in Supporting Information S1.