A 131-year evidence of more extreme and higher total amount of hourly precipitation in Hong Kong

Based on the observations of hourly precipitation for 131 years from Hong Kong Observatory Headquarters, this study examined the long-term changes in the characteristics of hourly precipitation extremes in terms of intensity, total precipitation amount, duration, and frequency. Results show that the hourly precipitation extremes have significantly intensified by 29%–38% from 1885 to 2022. The 131-year observations evidence that the more extreme the hourly precipitation is (i.e. higher percentiles), the faster the increasing rate it has. Specifically, the magnitudes of hourly precipitation with the 95th, 97.5th, 99th, and 99.9th percentiles increased by rates of 0.03 mm, 0.05 mm, 0.07 mm, and 0.12 mm per year, respectively. Through the secular trend analysis, we found that only the maximum intensity of extreme precipitation events (i.e. events with maximum intensity exceeding the 95th percentiles) shows a significant increasing trend during 1885–2022, while the trends in the total precipitation amount, duration, and mean intensity are not significant. However, by comparing the percentile bin values between three sub-periods of the 131-year record, we found a significant rise over time in the total precipitation amount, mean intensity, and maximum intensity of extreme precipitation events with different intensities (i.e. 95th, 96th, 97th, 98th, and 99th percentiles), while the change in the duration is not significant. The analysis of the frequency of precipitation events shows significant increases in the proportion of extreme precipitation events during 1885–2022. The observations of 53 stations across Hong Kong from 1986 to 2022 show significant intensification and increasing frequency in the hourly precipitation extremes in most areas of Hong Kong. Meanwhile, the precipitation duration shows a decreasing tendency, which may explain the insignificant changes in the total precipitation amount. These findings provide important insights into the longer-term variations in the characteristics of hourly precipitation extremes.


Introduction
As one of the world's most common hazards, extreme precipitation triggers flooding and has caused enormous damage to human society [1][2][3][4][5][6].In recent years, sub-daily precipitation extremes have received more and more attention, because urban flooding has become more frequent and extreme due to the intensification of short-term rainstorms [7].It is well known that the intensity of daily extreme precipitation is expected to increase by an approximate Clausius-Clapeyron rate (i.e.7% per degree as temperature rises, CC rate) as the climate warms [8][9][10].However, due to the regional dynamical responses, the dependence of sub-daily precipitation extremes on temperatures shows large regional deviation, with scaling rates exceeding the CC rate being detected in areas such as the US [11], Europe [12], and Australia [13].Urban flooding (also called pluvial flooding or waterlogging) associated with short-term precipitation has become a crucial threat to the sustainability of major cities such as Beijing [14], Hong Kong [15,16], Kuala Lumpur [17,18], and New York City [19].These studies show that the changes in the sub-daily precipitation extremes under climate warming may be more complex than the CC relation, highlighting the necessity of more research efforts on investigating the changes in precipitation extremes in different regions and at finer time scales (i.e.sub-daily and sub-hourly).Previous studies have reported increases in sub-daily extreme precipitation in regions such as the US [10], Australia [13], Japan [20,21], Peninsular Malaysia [22], and Indian [23].Some regions such as China and Italy exhibit considerable regional differences in the changes of sub-daily extremes, and no obvious trends at the whole scale were detected [24][25][26].These studies provided important results in revealing the spatial and temporal characteristics of hourly precipitation extremes in these areas, but were mostly limited to relatively short length of records (e.g.mostly based on observations of the past 30-50 years).
Studies based on a relatively short period may be challenging to identify whether the trends are a part of interdecadal variations owing to natural variability or long-term changes caused by global warming.For instance, an analysis of the hourly precipitation in Tokyo from 1890 to 1999 found two peaks of intensity and frequency of hourly precipitation extremes in the 1940s and the 1990s, and one valley in the 1970s [21].
In that case, one may conclude that the precipitation extremes in the 1990s are unprecedentedly strong if he or she conducts the trend analysis using data after the 1970s, even though the hourly precipitation extreme around the 1940s was even stronger and more frequent than that in the 1990s.With data records of ∼100 years, it is found that the temporal changes in the hourly precipitation extremes were affected by both long-term increasing trends and inter-decadal variations with a timescale of ∼30 years and longer [27].These results indicate the necessity of investigating the effects of climate changes and internal variability on the hourly precipitation extremes using data with longer records (e.g.>100 years).
Other than extreme precipitation at a specific time scale (e.g.maximum precipitation intensity at an hourly or daily scale), precipitation events, defined as rainy spells with continuous precipitation records (e.g.⩾1 h), are another important perspective to characterize precipitation behaviors.Different features of precipitation events, such as maximum intensity, mean intensity, duration, total amount, and frequency, provide very detailed knowledge of the changes in hydrometeorological behaviors [28][29][30].Total precipitation amount and duration are closely related to hydrological processes, and have significant impacts on overland flow and groundwater [31].Moreover, the duration is an important feature that reflects the mechanisms of precipitation events.Short-duration and intense precipitation is usually triggered by strong mesoscale convective activity, while long-duration precipitation is closely related to large-scale circulation such as monsoon [32].However, only a few studies have investigated the changes in the duration-related characteristics of hourly precipitation events.
In this study, hourly precipitation observations of 131 years in Hong Kong were used to examine the changes in the behaviors of extreme hourly precipitation and the characteristics (i.e. total precipitation amount, duration, mean intensity, maximum intensity, and frequency) of precipitation events at the hourly scale.Specifically, we first evaluated the changes in extreme hourly precipitation at the 95th-99.9thpercentiles over the last 131 years.Second, the changes in the characteristics of extreme precipitation events at the hourly scale were analyzed.Lastly, we further extended our analysis to 53 stations across Hong Kong in the more recent period of 1986-2022.Our study, based on the century-long hourly precipitation records, can provide important insight into the historical changes of precipitation extremes (in aspects of both hourly precipitation and precipitation events) affected by both decadal-scale climate variability and global warming.

Data
A high-quality precipitation dataset with observational hourly records at the Hong Kong Observatory (HKO) Headquarters from 1885-2022 was collected from HKO to examine the temporal evolution of hourly precipitation and the characteristics (i.e.intensity, duration, and the total precipitation amount) of precipitation events at the hourly scale in Hong Kong [15,33].The precipitation records during 1940-1946 are not available due to WWII.Therefore, the overall record length of this dataset is 131 years.The instrument used in HKO Headquarters is an ordinary 203 mm raingauge, which is checked with the records of a Casella 100573E tipping-bucket raingauge nearby.The instruments in the HKO Headquarters were slightly relocated (<20 m) in 1933.The homogeneity of precipitation data in the HKO Headquarters was examined using a standard normal homogeneity test [34], which suggested that the impact of changes in the instruments (e.g.relocation and updates of instruments) on the data consistency is minor and negligible.More detailed information on the changes of instruments used in the HKO Headquarters since 1884 can be found in [33,35].Besides, the hourly precipitation data during 1986-2022 of 53 stations in Hong Kong were collected to evaluate the spatial and temporal changes in hourly precipitation extremes during the more recent decades.The locations of the precipitation stations and a brief introduction to the climate of Hong Kong are provided in appendix A, Supplementary data.

Secular trend analysis of hourly precipitation
We used a 15 year sliding window analysis method to analyze the trends of extreme hourly precipitation at high percentiles [27].Taking the hourly precipitation from 1885 to 2022 as an example, the generalized Pareto distribution (GPD) was fitted using the upper 10% data of wet hours (hours with precipitation amount ⩾0.1 mm) for each 15 year sliding window (i.e.1885-1899, 1886-1900, 1887-1901, …, 2008-2022) [12].Afterward, the 95th, 97.5th, 99th, and 99.9th percentile values of each 15 year window were computed from the GPDs.The 95% confidence interval was calculated based on the GPDs to show the error bands.For clarity, the last year of each time window was used to represent that time window.For example, the percentile values calculated in the time window 1885-1899 were taken as results of the year 1899.The changes in the 95th, 97.5th, 99th, and 99.9th percentile values were estimated to examine the long-term trends of the extreme hourly precipitation.The Modified Mann-Kendall test was used to examine the significance level of the monotonic increasing or decreasing trends [36].The change rate was estimated using Sen's slope [37].The 15year sliding window was selected with considerations of (1) obtaining enough samples for determination of precipitation extremes, and (2) capturing interdecadal variations [27].The sliding windows of 10and 20 year were also used to test the sensitivity of results to different lengths of sliding windows.The same method was applied to the data during 1986-2022 of 53 stations but using a 5-year sliding window due to the relatively short recording length.

Definition of characteristics of precipitation events
The precipitation events were identified as continuous precipitation hours with intensity ⩾0.1 mm hr −1 that were separated by at least 2 dry hours (the hours with precipitation <0.1 mm) [29,38].On this basis, extreme precipitation events were precipitation events with at least one extreme precipitation record exceeding the 95th percentile.The total precipitation amount is the accumulated precipitation during a precipitation event.The duration of a precipitation event is the number of hours from the beginning to the end of the precipitation event (including the dry hours within a precipitation event; also see figure S2).The mean intensity was calculated as the total precipitation amount divided by the duration.The maximum intensity is the highest precipitation intensity of a precipitation event.Moreover, the changes in frequency, i.e. the number of extreme precipitation events, were also analyzed.More details about the definition of (extreme) precipitation events can be seen in appendix B.

Comparison analysis between sub-periods
Based on the very long time series at the HKO Headquarters, we further examined the changes in hourly precipitation and the characteristics of precipitation events by comparing the mean values in various percentile bins for the earliest, the middle, and the latest 30 year periods, i.e. 1885-1914, 1947-1976, and 1993-2022, respectively.Take the hourly precipitation as an example, for each sub-period, the hourly precipitation was binned by every 2 percentiles (i.e.[90th, 92nd), [92nd, 94th), [94th, 96th), [96th, 98th), and ⩾98th).The percentile values were calculated based on the fitted GPDs.The mean for each bin and each sub-period was calculated and compared to examine the temporal changes of precipitation extremes of various percentiles.A Mann-Whitney-Wilcoxon test at the 5% significance level was used to assess differences in the hourly precipitation among different sub-periods [11,13,39].To examine the changes in the characteristics of extreme precipitation events with different intensities, the extreme precipitation events were binned with their maximum intensities by every 1 percentile interval (i.e.(95th, 96th], (96th, 97th], (97th, 98th], (98th, 99th], and >99th) for each sub-period.Then we calculated and compared the mean values of total precipitation amount, duration, mean intensity, and maximum intensity of extreme precipitation events in different percentile bins and sub-periods.

Long-term trends in the magnitude of extreme hourly precipitation
We first assessed the long-term trends of high percentiles of hourly precipitation during 1885-2022 using a 15 year sliding window analysis.The precipitation extremes at the 95th, 97.5th, 99th, and 99.9th percentiles show significant increasing trends with obvious inter-decadal oscillations during 1885-2022 (figure 1).The rates of increase of the four percentiles are 0.03, 0.05, 0.07, and 0.12 mm per year, respectively.Here, we found that the higher the percentiles of hourly precipitation, the faster the magnitude of hourly precipitation increases (figure 1(e)).The percentage increases (calculated as the difference between the end year and the start year divided by the value at the start year) of hourly precipitation extremes at the 95th, 97.5th, 99th, and 99.9th percentiles are 37%, 38%, 37%, and 29%, respectively (tables S2).The upward trends are still significant and even more pronounced when considering the time series after WWII, i.e. 1947-2022 (table S2).When using a different length of sliding window such as 10 years or 20 years, the trends in the extreme hourly precipitation do not change considerably (figures S3 and S4).This result is in agreement with that in [27], which also showed increasing trends in extreme hourly precipitation in Hong Kong.However, the aim of [27] was to examine the scaling relationship between extreme hourly precipitation and dew point temperatures, while our study focuses on the temporal evolution of hourly precipitation, especially characteristics of precipitation events.
Most recently on 7-8 September 2023, the Guangdong-Hong Kong-Macao Greater Bay Area including cities Hong Kong, Shenzhen, and Guangzhou experienced an unprecedented extreme precipitation event with hourly precipitation reaching 158 mm in Hong Kong, which is the highest since records began in 1884.We further collected the latest hourly precipitation data at HKO Headquarters to September 2023, and analyzed the changes in hourly precipitation during the wet season of Hong Kong, i.e.May-September from 1884 to 2023 to examine the trends in extreme hourly precipitation when taking this record-breaking extreme event.Results show that the increasing trends in the extreme hourly precipitation are still significant (figure S6), and this individual extreme event does not influence the longterm trends substantially.

Changes in characteristics of precipitation events
The total amount, duration, and intensity of precipitation events are key features reflecting precipitation behaviors.We examined the long-term trends in the annual mean total precipitation amount, duration, mean intensity, maximum intensity, and the frequency of all and extreme precipitation events in the HKO Headquarters, respectively.Due to limited space, we mainly focused on the extreme precipitation events in this section, while the results of all precipitation events are provided in appendix D.
For extreme precipitation events, the total precipitation amount, mean intensity, and maximum intensity show increasing trends during 1885-2022, but only the increase (0.015 mm hr −1 per year) in maximum intensity is significant (figure 2).The mean duration of extreme precipitation events shows an insignificant decreasing trend (figure 2(b)).The time series of the annual maxima of these characteristics show consistent trends as using the annual mean values (figure S10).Our analysis of the frequency found that even though the total number of all precipitation events decreased significantly, the number of extreme precipitation events increased significantly by 0.042 events per year during 1885-2022 (figure 2(e)).In that case, the proportion of extreme precipitation events (which is calculated by dividing the number of extreme precipitation events using the number of all precipitation events) also shows a significant increasing trend of 0.035% per year (figure 2(f)).
To further examine the changes in characteristics of extreme precipitation with different maximum intensities, the mean values of total precipitation amount, duration, mean intensity, and maximum intensity of extreme precipitation events with different maximum intensities (i.e.exceeding the 95th, 96th, 97th, 98th, and 99th percentiles, respectively) at three sub-periods were calculated and compared.We found that the mean values of total precipitation amount, mean intensity, and maximum intensity are higher for more extreme precipitation events, and increased over time (i.e. from P1 to P2, and from P2 to P3), while the differences in the mean duration between precipitation events with different maximum intensities and different sub-periods are not obvious (figure 3).
To test the sensitivity of the results to different selections of years in each sub-period, we repeated this comparison analysis 100 times by randomly sampling the 30 year data over each of the three periods 1885-1939 (P1), 1947-1984 (P2), and 1985-2022 (P3).Results show that significant increases in the total precipitation amount, mean intensity, and maximum intensity of extreme precipitation events with different maximum intensities were found in most cases, while exceptions such as insignificant increases and even decreases also exist (figure S11).As for the duration, the changes are not significant in most cases, and are very sensitive to the selection of sub-periods.In general, the duration of extreme precipitation events in percentile bins <99th tended to increase from P1 to P2, and then decrease from P2 to P3 (figure S11b).This change pattern is consistent with that of all precipitation events (figures S8(b) and S9(b)).

Changes in the hourly precipitation extremes in extensive Hong Kong
We further analyzed the temporal evolution of hourly precipitation intensity, the total precipitation amount, duration, mean intensity, and maximum intensity of precipitation events during 1986-2022 in 53 stations in Hong Kong.Figure 4 shows the change rates (i.e.slope) of the 95th, 97.5th, 99th, and 99.9th percentile values of hourly precipitation.In most locations, the extreme hourly precipitation in different percentiles shows increasing trends during 1986-2022, especially in northern New Territories, and Hong Kong Island.
For extreme precipitation events, the total precipitation amount shows insignificant decreasing trends in most stations (figure 5(a)).The duration shows significant decreasing trends in northern New Territories, Kowloon, and Hong Kong Island (figure 5(b)).The mean intensity of extreme precipitation events shows significant increasing trends in most stations (figure 5(c)).Whereas, over this shorter period (i.e.1986-2022), the increasing trends in the maximum intensity are not significant in most stations (figure 5(d)), even though the trend in the maximum intensity over the 131 years is significant.The less obvious increase in intensity could not offset the effect of shortened duration, which explains the decreasing trends or insignificant increasing trends in the total precipitation amount.For all precipitation events, significant increases in the total precipitation, mean intensity, and maximum intensity were detected in most areas in Hong Kong, while the duration shows decreasing trends in most stations (figure S12).The decreasing trends in the duration of all and extreme precipitation events in the recent few decades further validate the changing pattern of increasing first and then decreasing in duration as mentioned in section 3.2.
The analysis of the changes in the frequency of extreme precipitation events finds significant increasing trends in the number of extreme precipitation events and its proportion to the total number of precipitation events in most areas of Hong Kong, indicating more and more frequent extreme precipitation events (figure S13).

Discussion
In this study, we analyzed the long-term trends and temporal changes in the hourly precipitation extremes by employing very long-term observational data from 1885 to 2022.Our results show that the magnitudes of higher percentiles increased faster, which might be because extreme precipitation at higher percentiles responds more pronounced to the increasing temperature [12].The upward trends in extreme hourly precipitation are not monotonic, but show inter-decadal variations.For example, in all percentiles, there is a valley period during the 1950s with less strong precipitation followed by a peak during the 1970s (figure 1).This inter-decadal scale variation is not apparent in previous studies using shorter recording data [16,25].The analyses of the annual maximum of short-duration (5 min-3 h) precipitation in Hong Kong during 1984-2010 only found increasing trends with an abrupt change in the early 1990s [16].As for the causes of the temporal changes in hourly precipitation, the increasing trend in the extreme hourly precipitation was primarily attributed to the warmer climate and urbanization according to previous studies [27,40,41], while the ENSO and winter monsoon can also influence the inter-annual variations of precipitation in Hong Kong [42].The long-term trends in the duration of extreme precipitation events are not statistically significant (figure 2(b)).Nevertheless, significant differences in the duration of extreme precipitation events between different sub-periods were detected in some cases (figure S11).Although the differences are not significant in most cases and are very sensitive to the selection of sub-periods, the duration shows a pattern of increasing from P1 to P2 and decreasing from P2 to P3 for percentile bins <99th.This pattern is clearer for duration of higher percentiles (e.g.>96th) in all precipitation events (figures S8(b) and S9(b)).The decrease in duration in recent decades might be related to the increase in short-duration precipitation events [28,43].For example, the enhancement of hourly precipitation extremes in urban areas of the Pearl River Delta was mainly attributed to the significant increase of precipitation events with durations <6 h [43].However, further research efforts are needed to explore the causes of the changes in duration on a longer-term scale.
Our results also suggest significant increases in the intensity, and total precipitation amount of all and extreme precipitation events in recent decades compared to a century ago.The frequency of extreme precipitation events increased significantly even though the total number of precipitation events decreased.These results imply greater flood risks in Hong Kong.Since the statistics of hourly precipitation are critical parameters for the design of flood protection infrastructures such as the drainage system, identifying the changes in extreme precipitation and understanding the drivers are necessary for better design and planning [44].Except for extreme precipitation, Hong Kong is also facing challenges from more intense storms, sea-level rise, and higher exposure due to socio-economic development [45,46].Therefore, apart from traditional engineering measures, sufficient strategic planning and more flexible approaches are needed to enhance the resilience of the city [47].For example, non-traditional approaches using big data and social media to improve the preparation and response of communities to floods have emerged recently [48,49].These results are not only applicable in Hong Kong, but also of practical significance to other cities to enhance climate readiness, especially flood prevention and management.

Conclusions
In this study, we investigated the long-term changes in hourly precipitation extremes in aspects of hourly precipitation and precipitation events based on observations of 131 years at HKO Headquarters.By analyzing the higher percentiles (i.e.95th, 97.5th, 99th, and 99.9th) of hourly precipitation, we found that the extreme hourly precipitation has intensified significantly by 29%-38% during 1885-2022.The more extreme hourly precipitation intensified faster, i.e., the magnitude of the 95th, 97.5th, 99th, and 99.9th percentiles increased at the rates of 0.03, 0.05, 0.07, and 0.12 mm per year, respectively.
We further analyzed the characteristics including the total precipitation amount, duration, mean intensity, maximum intensity, and frequency of extreme precipitation events to better understand the evolution of precipitation behavior during the past century.The secular trend analysis suggested a significant increase in the maximum intensity of extreme precipitation events during 1885-2022, while the trends in the total precipitation, duration, and mean intensity are not significant.However, when comparing these indices between sub-periods, significant changes were detected.The mean values of the total precipitation amount, mean intensity, and maximum intensity of extreme precipitation events with different intensities show significant increases over time.As for duration, the changes are very sensitive to the selection of sub-periods, but there is a changing pattern of increasing first and then decreasing over the 131-year period in general.We also found that significant increases in the number of extreme precipitation events and its proportion to the total number of all precipitation events during the past century.
The intensification of hourly precipitation extremes was widely detected in most areas of Hong Kong during 1986-2022.However, due to the shortening of precipitation duration, the changes in the total precipitation amount of extreme precipitation events are not significant in this period.In the shorter period, the frequency of extreme precipitation events, and its proportion still show significant increasing trends in most areas of Hong Kong.
As summarized above, hourly precipitation extreme events tended to be more intense, higher total precipitation amount, and more frequent from 1885 to 2022.These changes imply greater pluvial and flash flood risks from the more intense and frequent hourly precipitation extremes.

Figure 1 .
Figure 1.Temporal changes of the (a) 95th, (b) 97.5th, (c) 99th, and (d) 99.9th percentiles of hourly precipitation at HKO Headquarters.The percentile values were estimated based on the generalized Pareto distribution (GPD) fitted using the upper 10% data of each 15 year sliding window.The last year of each time window was used to represent that time window.The grey shadings indicate the 95% confidence interval calculated based on the GPD.The colored solid (dashed) straight line indicates the significant trend (insignificant) at the 5% level based on the Modified Mann-Kendall test.The black straight lines in (a)-(d) indicate the trends during two sub-periods (i.e.1885-1939 and 1947-2022).Sen's slopes during 1885-2022 are shown.(e) shows the four percentiles in the same panel.

Figure 2 .
Figure 2. Temporal changes in the characteristics of extreme precipitation events (i.e. at least one extreme precipitation record with precipitation exceeding the 95th percentile) at HKO Headquarters.Panels (a)-(d) are the scatter plots of the total precipitation amount (mm), duration (hr), mean intensity (mm hr −1 ), and maximum intensity (mm hr −1 ) of extreme precipitation events, respectively.Panel (e) shows the annual number of extreme and all precipitation events.Panel (f) shows the proportion (%) of the number of extreme precipitation events to that of all precipitation events.In panels (a)-(d) and (f), the black solid lines indicate the time evolution of the annual mean values, while the solid (dashed) straight lines indicate the significant (insignificant) trends at the 5% level based on the Modified Mann-Kendall test.In panel (e), the red (blue) lines indicate the time evolution of the annual number of extreme (all) precipitation events, while the solid straight lines indicate the significant (insignificant) trends at the 5% level based on the Modified Mann-Kendall test.Sen's slopes during 1885-2022 are shown.In panels (a)-(d), the natural logarithmic y-axis is used.

Figure 3 .
Figure 3.Comparison of (a) total precipitation amount (mm), (b) duration (hr), (c) mean intensity (mm hr −1 ), and (d) maximum intensity (mm hr −1 ) of extreme precipitation events (i.e. at least one extreme precipitation records with precipitation exceeding the 95th percentile) with different maximum intensity among three sub-periods.The percentile bins were binned based on the maximum intensity of precipitation events.The colors indicate three different sub-periods: 1885-1914 (red boxes), 1947-1976 (green boxes), and 1993-2022 (blue boxes).The whisker represents the range between the 5th and 95th percentiles of the data sample.The box represents the interquartile range of the data sample.The horizontal black line inside each box indicates the median value.In (d), the natural logarithmic y-axis is used.

Y
Lai et al

Figure 4 .
Figure 4. Changes in extreme hourly precipitation of the (a) 95th, (b) 97.5th, (c) 99th, and (d) 99.9th percentiles during 1986-2022 in Hong Kong.The colors indicate the change rates estimated using Sen's slope.The triangles (inverted triangles) indicate that the increasing (decreasing) trends are significant at the 5% level based on the Modified Mann-Kendall test, while the circles indicate insignificant trends.

Figure 5 .
Figure 5. Changes (i.e.Sen's slope) in (a) precipitation amount, (b) duration, (c) mean intensity, and (d) maximum intensity of extreme precipitation events (i.e. at least one extreme precipitation records with precipitation exceeding the 95th percentile) during 1986-2022 in Hong Kong.Red (blue) circles indicate increasing (decreasing) trends.The colors indicate the change rates estimated using Sen's slope.The triangles (inverted triangles) indicate that the increasing (decreasing) trends are significant at the 5% level based on the Modified Mann-Kendall test, while the circles indicate insignificant trends.