1. Introduction
There is growing evidence of increases in the intensity and frequency of extreme precipitation events in a warming climate (Donat et al. 2016; Guerreiro et al. 2018; IPCC 2021; Kendon et al. 2014; Myhre et al. 2019; Wang et al. 2017; Westra et al. 2014), with potentially serious consequences of more frequent and severe flooding (Sharma et al. 2018; Wasko and Nathan 2019). Precipitation intensities are predicted to increase at a rate of 6%–7% °C−1 based on the increase of atmospheric water vapor as governed by the Clausius–Clapeyron (CC) relation. However, observed and modeled rates of projected increase in extreme precipitation intensity have been found to vary significantly across the globe, with projected rates above 7% (superCC) found in tropical and subtropical regions (Berg et al. 2009; Emori and Brown 2005; Fowler et al. 2021b; Sugiyama et al. 2010; Visser et al. 2021). This has led to increasing interest in the potential intensification of temporal patterns of precipitation (Fowler et al. 2021a; Westra et al. 2014), as after the precipitation depth, the temporal pattern of precipitation typically has the largest influence on flood estimates, with less uniform distributions of extreme precipitation resulting in higher flood peaks (Hettiarachchi et al. 2018; Nathan et al. 2016). However, to date, there is little to no observational evidence that within precipitation events, temporal patterns are changing.
Multiple factors confound the comparison of temporal patterns between individual precipitation events, including the separation of individual storm events (Gaál et al. 2014; Visser et al. 2020), calculating appropriate statistical measures of the precipitation pattern (Gocic and Trajkovic 2013; Martins et al. 2012), and the mixing of dominant precipitation mechanisms (Berg and Haerter 2013; Molnar et al. 2015). The confounding factors also include potential changes in the dominant precipitation mechanism, as convective precipitation exhibits different characteristic spatial and temporal scales compared to stratiform precipitation (Berg et al. 2013). The limited availability of long-term records of observed subdaily precipitation data also limits studies of temporal patterns of precipitation, with studies generally limited to using coarser data resolutions, such as daily. Examining changes in the temporal distribution of daily precipitation, Rajah et al. (2014) found increasing wet-day frequency and wet-day Gini values in Australia, the United States, southern South America, and western Europe, implying an increase in light and extreme events at the expense of dry and moderate events. Examining changes in wet-day frequency between 1950–83 and 1984–2016, Contractor et al. (2021) found that precipitation has intensified across a majority of land areas globally throughout the wet-day distribution, indicating that light, moderate, and heavy wet-day precipitation has become more intense worldwide.
Ranking equal-duration fractions of embedded, subdaily precipitation bursts by intensity, Wasko and Sharma (2015) examined the relationship between temperature and the temporal patterns of varying duration bursts across Australia, finding less uniform temporal patterns of precipitation (i.e., more intense peak precipitation and weaker precipitation during less intense times) at higher temperatures. This “peakier” temporal pattern was more pronounced in the tropics and with increased embedded burst duration (up to 12 h) but was also associated with a decrease in the embedded burst volume, possibly due to storms shorter in duration than the embedded burst being analyzed. Despite altering the natural temporal structure of the precipitation burst, this ranked analytical approach has been applied in the context of examining changes in flooding with higher temperatures, a necessity in designing stormwater infrastructure that performs adequately in a warmer climate.
Using a novel temporal concentration index (TCI) on complete precipitation events from 843 observation stations in a humid region of China, Long et al. (2021) found that precipitation events tend to be more concentrated both temporally and spatially at higher temperatures, with this concentration in both space and time not dependent on precipitation amount or duration. TCI values were found to increase (indicating temporal intensification) within the temperature range of 5°–24°C before plateauing at higher temperatures. However, the TCI reorganizes the temporal center of the precipitation event and, hence, only provides information on a deviation from uniformity. Analyzing complete precipitation events of similar duration, Visser et al. (2021) showed that the average intensity of precipitation events across Australia increases in line with expected CC relation of 6%–7% °C−1, with higher rates for peak 1-h embedded bursts, particularly in the tropics. This result implies that both complete precipitation events and embedded bursts at higher temperatures can be characterized by peakier temporal patterns compared to lower temperature events of similar durations.
While these studies perhaps suggest an intensification of precipitation events, none considers the temporal structure of the event itself. Embedded bursts typically represent only a small portion of the entire precipitation event, while ranked analytical approaches restructure the temporal center of complete precipitation events and remove the natural temporal correlation structure occurring in the precipitation sequence. Figure 1 presents three potential modes of change in the temporal pattern with increasing temperatures. The temporal pattern of a baseline conceptual precipitation event is presented in blue (Fig. 1a); the precipitation event begins at zero, increases to a maximum at the center, and then decreases symmetrically. The thick line presents the cumulative precipitation. All the above reviewed studies suggest a peakier temporal pattern. Such a peakier temporal pattern is presented in red in Fig. 1b, with an increase in the embedded burst of approximately 14% (double the CC rate) and decreased intensity over the remainder of the event. Despite a significant increase in the embedded burst intensity, there is a limited change to the baseline temporal pattern (red line versus blue line).
Cumulative precipitation (lines) and precipitation density (shaded areas) for conceptual precipitation events of equal duration. (a) Baseline event at a cooler temperature (blue). (b) Intensification of temporal pattern (red), i.e., a more intense embedded burst (approximately 14% increase), and weaker precipitation during less intense times. (c) Lateral shift in the concentration of precipitation, with the bulk of precipitation falling earlier for warmer events (yellow and red). (d) Combined intensification and lateral shift in the temporal pattern of precipitation with increasing temperature.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
However, if we consider a lateral shift in precipitation, with the bulk of the precipitation falling earlier in the event (Fig. 1c), a large shift in the temporal patterns results (red and yellow lines). As the embedded bursts among the three events remain largely comparable in magnitude and duration, the primary mode driving change here is when the embedded burst occurs. Finally, the third mode (Fig. 1d) combines intensification and lateral shift, which results in the increased intensity of shorter-duration embedded bursts occurring earlier in the event.
The event loading, that is, the timing of when the bulk of the precipitation occurs, is a defining characteristic of the temporal pattern of a precipitation event. Modern flood hydrology is increasingly interested in the characteristics of the complete flood hydrograph (e.g., the time of peak; the shape of the hydrograph) as the performance of many natural and constructed systems are dependent on the volume as well as the peak of the inflow flood. To properly simulate flood behavior, it is thus necessary to employ realistic precipitation temporal patterns that represent the complete storm and not just the embedded burst. To date, there is very little evidence available on how the dependence structure of temporal patterns of precipitation events might change with increasing temperature, or how hydrological applications should accommodate such changes. Here, we investigate the relationship between the temperature and temporal pattern of complete precipitation events, with a focus on the event loading. We use a continent-wide dataset for Australia that encompasses a wide variety of climates, to assess historical changes in temporal patterns and quantify their sensitivity to changes in temperature.
2. Data and methods
a. Meteorological data
Quality-controlled, subdaily observed precipitation data were obtained from the Australian Bureau of Meteorology. Precipitation data are available for 1489 stations across Australia and are measured by either tilting syphon pluviographs or tipping-bucket rain gauges (TBRGs) at 6-min intervals. Precipitation data with quality codes 4, 8, and 9 (i.e., interpolated, missing, as recorded but wrong) were dismissed. During 1996, the Australian pluviograph network was switched to TBRGs, increasing the recording threshold from 0.01 mm (pluviograph) to 0.2 mm (one tip for TBRG). Consequently, precipitation records after 1996 contain longer continuous periods of zero rainfall, resulting in larger discrepancies in event-based statistics (e.g., event duration) compared to annual-based statistics (e.g., total annual precipitation) (Wasko et al. 2022). To ensure consistency in precipitation-event statistics across instrument changes, pluviograph records were converted to TBRG records by emulating TBRG operation. Subdaily observed dry-bulb temperature data are available from 1829 synoptic stations across Australia, and subdaily dewpoint temperature are available from 1574 stations across Australia. Recording resolution for both variables varies from two to eight measurements per day. Dewpoint records that were considered low precision (no decimals) were reconstructed to higher precision records for 876 stations, using observations of the dry-bulb and wet-bulb temperatures, surface pressure, and psychrometric data, as described by Visser et al. (2020).
A data year was considered valid if it had less than 20% observations missing, and a station record was considered valid if it had less than 20% missing years over the selected study period. For scaling analyses, qualifying stations required a minimum of 10 valid years of precipitation and temperature data over any period (151 stations). For trend analyses, qualifying stations required at least 80% valid data years for the study period 1960–2016 for precipitation data (55 stations) and additionally for both precipitation and temperature data (28 stations).
b. Selection of precipitation events
To analyze the temporal structure of precipitation events, it is necessary to identify individual precipitation events. Precipitation events were separated by a minimum period of zero rainfall, that is, a minimum interevent time (MIT). Smaller MIT values limit intra-event intermittency, while larger MIT values ensure event independence. A balanced MIT value of 3 h was selected in line with previous studies over Australia (Visser et al. 2021; Wasko et al. 2015). Several attributes for each precipitation event are extracted, such as duration and total precipitation. Precipitation events with duration shorter than 1 h are removed (i.e., fewer than 10 time steps) to limit temporal pattern variability introduced by shorter-duration events.
c. Calculation of D50
To characterize the temporal structure of the precipitation event, the percentage of the event duration at which 50% of the cumulative event precipitation total is reached is denoted as D50 (Fig. 2). Events with a D50 value smaller than 50% are classified as a front-loaded event, while events with a D50 value greater than 50% are classified as a rear-loaded event. By definition, D50 lies between 0% and 100% for each storm of interest. A change in the value of D50 over time or with respect to a covariate (e.g., temperature) can be inferred as an example of nonstationarity in the event temporal-pattern structure. Another advantage of this formulation is the ability to combine events across a range of durations, given the dimensionless measure of location D50 represents.
Calculation of the percentage of event duration at which 50% of the cumulative event precipitation total is reached (D50). Events with a D50 value less than 50% (red line) are classified as front-loaded events, while D50 values greater than 50% (blue line) are classified as rear-loaded events.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
d. Precipitation–temperature pairs
e. Historical changes in D50 and temperature
A study period from 1960 to 2016 was selected to examine potential trends in D50 and temperature variables. Due to the greater availability of subdaily precipitation data compared to temperature data, trend analyses of precipitation-event variables, such as D50, include all stations that meet the sole data requirements for precipitation data (55 stations). Similarly, for variables related to precipitation–temperature pairs, trend analyses were restricted to stations meeting the data requirements for both precipitation and temperature data (28 stations). The Mann–Kendall trend test (Kendall 1948; Mann 1945) was used to test for nonlinear monotonic trends. The Theil–Sen estimator (Sen 1968; Theil 1950) was used to assess the trend magnitude, with trend significance as calculated by the Mann–Kendall trend test presented with α equal to 0.05.
3. Results
a. Historical changes in D50 and temperature
Figure 3 presents station-based historical trends for the median annual D50 (Figs. 3a,b) and median annual temperature (Figs. 3c,d) from 1960 to 2016. Figure 3a indicates that nearly all stations have experienced decreasing trends (green triangles) in the median annual D50 value, with numerous sites having statistically significant trends (black outlines). This result indicates that the bulk of the precipitation is falling earlier in the event. Figures in the appendix show the percentage of front-loaded events per year has also increased over time (Fig. A1a); the percentage varies from 50% in the south of the country to 68% in the tropical north (Fig. A2). A histogram of station location trends (Fig. 3b) indicates a clear tendency toward lower D50 values for almost all stations. Trend results for the median annual temperature (Fig. 3c) indicate increasing temperatures before storm events across most of the country, with larger and more significant trends in the south.
Map of trends in the median annual (a) D50 and (c) representative storm temperature across Australia over the period 1960–2016 using station-based observations. Triangles and inverse triangles represent increasing and decreasing trends at station locations, respectively, with black-bordered triangles indicating stations with significant trends at p < 0.05. (b),(d) Stacked histograms of observed station location trends in the median annual (a) D50 and (c) representative storm temperature. Hobart Airport (ID 094008), Sydney Observatory Hill (ID 066062), Rockhampton Airport (ID 039083), Broome Airport (ID 003003), and Darwin Airport (ID 014015) stations are indicated as Hobart, Sydney, Rockhampton, Broome, and Darwin, respectively.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
Trend results for the median annual dewpoint (Fig. A1c) indicate overall smaller trends and fewer stations with significant trends across the country compared to temperature. Slightly decreasing trends in the median dewpoint are found in southern regions, while slightly increasing trends are found along the northern coastline and interior, comparable to mean annual dewpoint trends produced by Denson et al. (2021). Due to the complex interplay between numerous climatic variables for different storm mechanisms, pinpointing a predominant causative factor behind the increase in front-loaded precipitation events is complicated. However, the greater uniformity and significance in temperature trends points to a predominant influence of increasing temperature over increasing dewpoint. Hence, we proceed with using temperature as a scaling variable.
b. Temporal patterns versus temperature
To ensure D50 is relevant to characterizing potential changes in the temporal pattern structure of precipitation, Fig. 4 presents the median temporal pattern of precipitation events based on individual temperature subsets (bins) for two stations, Sydney (Fig. 4a) and Darwin (Fig. 4b), as indicated in Fig. 3. The two stations are selected based on differences in trend results for both D50 and temperature. For the temperate climate station Sydney (Fig. 4a), median temporal patterns reveal a more uniform distribution of precipitation over a wider temperature range. With increasing temperature, a small but noteworthy shift to lower D50 values is evident. The limited number of events in the extreme temperature bins increases the variability of these median lines.
Median temporal pattern of precipitation events per representative storm temperature bin (lines) for (a) Sydney and (b) Darwin. Minimum requirement of 20 events per temperature bin.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
For the tropical station Darwin (Fig. 4b), median temporal patterns reveal a clearer systematic decrease in D50 values over a much narrower temperature range. Lower D50 values (<50%) for all temperature bins are representative of the tropical north’s high percentage of front-loaded events per year (Fig. A2). Differences in the median temporal patterns for the two locations can be partially explained by differences in the dominant storm mechanisms, whereby Darwin’s warmer climate experiences predominantly shorter-duration convective events compared to longer-duration frontal systems for Sydney. The rate of decrease in D50 values with temperature, therefore, appears dependent on the dominant storm duration or storm mechanism of the selected location and temperature range.
c. D50 versus temperature
To illustrate this rate of decrease of D50 with temperature, Fig. 5 presents the relationship between the median D50 and temperature for five stations across Australia, including Sydney and Darwin. For each station, the median D50 is calculated for precipitation events group per 1°C temperature bin. Figure 5a presents the relationship between D50 and temperature for all precipitation events, and Fig. 5b presents the same relationship for precipitation events with durations longer than 9 h.
Relationship between median D50 and the representative storm temperature for (a) all events and (b) events with duration longer than 9 h for five stations: Hobart (ID 094008), Sydney (ID 066062), Rockhampton (ID 039083), Broome (ID 003003), and Darwin (ID 014015). Mean D50 values (points) are calculated for binned precipitation events per 1°C temperature bins, with a minimum requirement of 20 events per bin. Station latitudes are indicated next to station names, and station locations are shown in Fig. 3.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
Considering all event durations (Fig. 5a), D50 values are found to decrease with increasing temperature for all five stations analyzed. The rate of D50 decrease with increasing temperature is lower for temperatures below 25°C (roughly −0.6% °C−1), while steeper rates of decrease are experienced above 25°C (roughly −4% °C−1). Results from Fig. 5a indicate that increases in representative storm temperatures are associated with increased front loading of precipitation events. Even small increases in temperatures in tropical locations, such as Darwin and Broome, correspond to a significant decrease in median D50 values. These tropical stations have a greater dominance of shorter-duration, intense, convective precipitation events associated with elevated temperatures (Visser et al. 2021).
For shorter-duration events, each precipitation interval (6 min) will constitute a larger proportion of the total duration and likely precipitation total, compared to longer-duration events. Therefore, shorter duration events are likely to have greater variability in their temporal patterns and D50 compared to longer-duration events. Only considering longer-duration events (>9 h), Fig. 5b reveals greater consistency in median D50 values (near 50%) compared to all events across the same temperature range. Precipitation events with durations longer than 9 h are limited to a cooler temperature range for all stations, as longer duration events typically require a continuous inflow of moisture, which is more readily sustained by a lower saturation vapor pressure (Visser et al. 2021). Clear differences in D50–temperature sensitivities between Figs. 5a and 5b indicate a dependence on event duration, which requires further examination.
d. D50 versus event duration
Figure 6 presents the relationship between D50 and event duration for Sydney (Fig. 6a) and Darwin (Fig. 6b). Comparison of the two panels in Fig. 6 reveals unique distributions in precipitation events (points), event density (purple to yellow shading), and the median D50 value (black line) for each station per selected duration bin. For both stations, the range of D50 values narrows with increasing duration, with D50 values trending toward 50% for longer-duration events. This suggests increased temporal uniformity in precipitation intensity for longer-duration events. However, with increasing duration, a single 6-min time step represents a smaller proportion of the total event and the probability that 50% of the event precipitation total will fall within the first or last few time steps decreases.
Relationship between D50 and event duration for (a) Sydney and (b) Darwin. Color shading indicates the density of precipitation events (points) as a percentage of the total events that fall within a neighborhood area. Higher density is indicated by yellow shading; lower density is indicated by blue-purple shading. The median D50 value (black line) is calculated for unequal-duration bins and drawn using locally estimated scatterplot smoothing (LOESS) with a default span (α) of 0.75.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
Based on the median D50 value per duration bin (black line), Sydney precipitation events are roughly divided in equal numbers between front- and rear-loaded events (Fig. 6a). However, an increased concentration of front-loaded events is evident at shorter durations (2–4 h). For Darwin (Fig. 6b), a much higher number of front-loaded events is present at shorter durations (yellow shading). This is likely due to the region’s summer-dominant rainfall regime, characterized by predominantly short-duration convective events. Figure 6 demonstrates that D50 is dependent on event duration, particularly for tropical locations. Therefore, accurately assessing the changes in D50 with increasing temperature requires comparison of precipitation events with similar durations (and, hence, likely similar storm mechanisms).
e. D50 versus precipitation intensity
Precipitation events typically have different intensity characteristics based on event duration. Shorter-duration (convective) events tend to produce higher peak intensities, while longer-duration (stratiform) precipitation events tend to have lower average intensities (Visser et al. 2021). Periods of high-intensity precipitation during events, therefore, will exhibit steeper temporal pattern slopes, potentially inducing differences in D50 values compared to lower-intensity events (refer to Fig. 1). Figure 7 presents the relationship between D50 and average precipitation intensity for Sydney (Fig. 7a) and Darwin (Fig. 7b) for different event-duration categories.
Relationship between D50 and average precipitation intensity for (a) Sydney and (b) Darwin. Median D50 values (lines) are calculated for unequal event duration bins of 1–2 h (red), 2–6 h (green), longer than 6 h (blue), and all events (black). Gray points represent individual precipitation events. Median lines are drawn using LOESS with a default span (α) of 0.75.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
Figure 7 reveals that for both stations, higher-average-intensity precipitation events tend to be more front loaded (decreasing D50 lines) compared to lower-intensity events. This is particularly evident for shorter-duration events (<6 h), in which median D50 values (lines) decrease with increasing average precipitation intensity up to a certain extent (≈8 mm h−1) before increasing. Longer-duration events (blue lines) undergo limited change in D50 values with increasing average precipitation intensity, with slight decreases observed for Darwin (Fig. 7b).
Like event duration, a trending of the D50 range toward 50% is noted with increasing average precipitation intensity for both stations (points). Precipitation events with higher average precipitation intensity (rarer events) will have less intermittency (periods of zero rainfall) compared to less intense (more frequent) events of similar duration. With increasing rarity for a given duration, precipitation events have higher precipitation totals than more frequent events, which are typically distributed over more wet time steps, reducing the impact of a single time step. With increasing average precipitation intensity, the probability that 50% of the event precipitation total will fall within the first or last few time steps decreases, resulting in the range of D50 values trending toward 50%.
f. Scaling of D50 with temperature
Results from the previous sections reveal the D50–temperature relationship is dependent on both event duration and precipitation intensity. Scaling analysis of the D50–temperature sensitivity is thus conducted based on event-duration and precipitation-intensity subsets. For each station, precipitation events are placed in subsets based on unequal-duration bins (see Fig. 8), followed by an additional subset based on the 75th-percentile average precipitation intensity. These two subsets result in two event samples per duration bin, one for low-average-intensity precipitation events (<75th percentile) and one for high-average-intensity precipitation events (≥75th percentile). To increase the number of available events per analysis, standardized temperature–precipitation pairs are pooled based on Köppen–Geiger climate zones (see section 2d) before the scaling (α) of D50 with temperature is calculated.
(a) Scaling of 50th-percentile D50 (%) with temperature across event durations for events with average precipitation intensities below the 75th percentile, pooled per Köppen–Geiger climate zone for Australia. Results are presented per main climate group for (b) equatorial, (c) arid, and (d) temperate zones. Quantile regression is used to calculate scaling rates (y axis) based on subsets of precipitation events for each event-duration bin (x axis), e.g., 1–2 h. Lines are drawn to connect calculated scaling rates (solid lines) and 95th-percentile rank–based confidence intervals (dashed lines) (Koenker 1994) across duration bins using LOESS with a default span (α) of 0.75. The Köppen–Geiger classification coverage was obtained from Beck et al. (2018) at 0.083° resolution. Minimum requirement: 100 events per climate zone per duration bin.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
Figures 8 and 9 present the scaling results for the 50th-percentile D50 value to temperature per climate zone and duration bin for lower-average-intensity events (<75th percentile) (Fig. 8) and higher-average-intensity events (≥75th percentile) (Fig. 9). It is necessary to increase the width of the duration bins to ensure sufficient events per analysis, with a minimum requirement of 100 events per duration bin. Three main climate types are considered, and these are represented with an initial capital letter: A (equatorial), B (arid), and C (warm temperate) groups. Due to a limited number of stations in certain climate zones, selected climate zones have been combined, including Af, Am, and Cwa along the northeast coastline, and BSk and BWk along the southern inland region.
As in Fig. 8, but for events with average precipitation intensities equal or above the 75th percentile.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
For lower-average-intensity events (Fig. 8), scaling results for equatorial groups (Fig. 8b) indicate negative scaling rates of D50 (increased front loading) for almost all durations, with the greatest shift to front loading between 1- and 4-h durations and for equatorial regions possibly due to the greater prevalence of convective events. With increasing duration, scaling rates increase toward zero for longer durations (>12 h). Moving south, arid groups (Fig. 8c) BSh and BWh have similar negative patterns to equatorial groups, but scaling rates are closer to zero. Southern arid groups BSk and BWk, along with warm-temperature groups (Fig. 8d), have the most uniform distribution of scaling rates across durations, with slight negative rates for most durations. For most event durations and regions, an increase in temperature is associated with an increased front loading of lower-intensity precipitation events.
The 95% confidence interval for each climate zone (dashed lines in Fig. 8) indicates the negative scaling rates for shorter durations (<9 h) are statistically significant in the equatorial group, zones BWh and BSh in the Arid group, and zones Cfa and Cfb in the warm-temperature group. The uncertainty increases for longer durations in certain climate zones due to smaller event sample sizes, with the 95% confidence interval widths generally varying from less than 1% for shorter durations, up to 2.5% for longer durations.
Scaling results for higher-average-intensity events (Fig. 9) are comparable to lower-intensity results (Fig. 8), with increased variability due to smaller sample sizes. A limited number of climate zones (BSh, BWh, Cfa) exhibit greater negative scaling rates over a wider range of event durations compared to lower-intensity events. For both lower- and higher-intensity categories, the greatest (negative) D50 scaling rates are observed in the equatorial zone between events durations of 2 and 6 h. To examine the difference in temporal patterns between precipitation-intensity categories, Fig. 10 presents a comparison of temporal patterns of 2–6-h-duration events for Darwin between events of lower-average intensity (Fig. 10a) and higher-average intensity (Fig. 10b).
Median temporal pattern of 2–6-h-duration precipitation events per representative storm temperature bin (lines) for Darwin. Precipitation events with average intensities (a) smaller than the 75th percentile and (b) larger than the 75th percentile. Minimum requirement: 20 events per temperature bin.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
For lower-intensity events (Fig. 10a), temporal patterns are comparable to Fig. 3b, which includes all precipitation events for Darwin. According to subsets based on higher-average-intensity precipitation events (Fig. 10b), precipitation temporal patterns show increased front loading (lower D50 values) but decreased separation in D50 values between temperature bins (lines closer together). For the highest temperature bin of 31°C (yellow line), D50 values are comparable between the two intensity categories (roughly 18% of the event duration). However, the higher-intensity temporal pattern (Fig. 10b) exhibits a high rate of precipitation (steep slope) for a greater proportion of the event duration, reaching 75% of the precipitation total much quicker than its lower-intensity counterpart. Both Figs. 10a and 10b indicate a combined intensification and lateral shift of temporal patterns with temperature (conceptualized in Fig. 1d). This combined change is associated with increased intensity of shorter-duration embedded bursts occurring earlier in the event, demonstrated by increasing temporal-pattern slopes with increasing temperature.
4. Discussion
a. Physical links to changing temporal patterns of precipitation
Precipitation intensities are predicted to increase with temperature rise due to an increase of atmospheric water vapor. The results suggest that this increased water vapor is preferentially precipitating at the start of the precipitation event, leading to a shift in the bulk of the precipitation. Importantly, the magnitude of temporal pattern shift depends on region, precipitation-event duration, and precipitation intensity. Shorter-duration storms tend to be more front loaded compared to longer-duration events (Fig. 6), particularly in the tropics, while higher-intensity events are also more likely to be front loaded (Fig. 7). It appears that increases in the percentage of front-loaded precipitation events (as shown in Fig. A1) may be linked to the greater occurrence of high-intensity, shorter-duration events at higher temperatures. This increased prevalence of short-duration events at higher temperatures is supported by the increased precipitation–temperature scaling rates found in the tropics (Visser et al. 2021), with stronger increases in precipitation intensity with temperatures characteristic of shorter-duration convective precipitation compared to longer-duration stratiform precipitation (Berg et al. 2013).
Significant increasing trends in temperature, over other climatic variables such as dewpoint, aligns with findings of decreasing mean relative humidity over Australia (Denson et al. 2021). Decreasing relative humidity can contribute to a reduction in the frequency of longer-duration precipitation events, as a complex interplay of atmospheric process is needed to sustain moisture inflow over longer durations. The increased uniformity in the temporal patterns of longer-duration events (minimal D50 scaling) aligns with lower precipitation–temperature scaling rates for longer-duration events (Visser et al. 2021).
b. Implications for flooding and flood design
The temporal and spatial patterns of precipitation affect catchment response, impacting streamflow and sediment transport volumes and peaks (Peleg et al. 2020). Indeed, after the precipitation depth, the temporal pattern of precipitation typically has the biggest influence on flood estimates, with less uniform distributions of extreme precipitation resulting in higher flood peaks (Hettiarachchi et al. 2018; Nathan et al. 2016). Although flood response is catchment specific and depends on the relative size of the storm to the catchment, the results here suggest that flood peaks could increase due to an intensification of the temporal pattern associated with a shift to greater front loading. However, changes in other catchment characteristics, such as drier initial conditions, could offset this change (Ho et al. 2022).
Combined with precipitation volume, precipitation temporal patterns are used in a variety of rainfall-based flood-estimation methods for the design of engineering infrastructure. Design flood models that rely on representing the temporal patterns of precipitation may need to consider a more front-loaded temporal pattern of increased peak intensity, while applications that use Monte Carlo sampling methods of multiple temporal patterns may require adjustments to sample a greater proportion of front-loaded temporal patterns to be representative of higher temperatures. Continuous simulation methods which use historical rainfall sequences may not be representative of future conditions, with stochastic-generation approaches requiring nonstationarity in the parameters to represent changes in the temporal pattern (Wasko and Sharma 2017). Further investigation will be required to conclude if the shift to more front-loaded temporal patterns is evident in precipitation events of extreme rarity, such as those used in probable maximum precipitation estimates.
5. Conclusions
To investigate the relationship between temperature and the temporal pattern of precipitation, we introduced an event-loading variable termed D50, which identifies the percentage of the event duration at which 50% of the total event precipitation is reached. Historical trend analyses based on station observations reveal D50 values have decreased across Australia over the past 60 years, resulting in an increased percentage of front-loaded (D50 < 50%) precipitation events per year. This increased front loading of precipitation events coincides with increasing trends in representative storm temperatures, with higher temperatures associated with a greater proportion of short-duration convective events. D50 values are shown to decrease with increasing temperature, with a greater rate of decrease above 25°C. However, due to the mixing of storm mechanisms across temperature ranges, quantifying changes in D50 to temperature requires careful consideration of event duration and precipitation intensity.
Shorter-duration precipitation events are associated with lower D50 values (more front loading) compared to longer-duration events. With increasing event duration, the variability in D50 decreases. Greater uniformity in the temporal patterns of longer-duration events is partly due to the reduced impact of individual temporal pattern intervals over longer-duration events of greater precipitation totals. The majority of high-average-intensity precipitation events are associated with front-loaded storms. D50 values are found to decrease with increasing average precipitation intensity for shorter-duration events (<6 h), but only to a certain precipitation intensity before increasing. D50 values for longer-duration events (>6 h) exhibit limited change with increasing average precipitation intensity. Similar to event duration, the range of D50 values trends toward 50% with increasing average precipitation intensity. This is likely due to a combined effect of decreasing event intermittency with increasing average intensity, and a possible lower limit of D50 values, which will be limited by the rate of precipitation.
For lower-average-intensity precipitation events (<75th percentile), D50 values decrease with increasing temperature across nearly all event durations, with greater decreasing rates for shorter-duration events (<6 h). The greatest decreasing rates (−2%°C−1) are found for shorter-duration events in equatorial climate zones. Scaling results for higher-average-intensity precipitation events (≥75th percentile) are found to be roughly comparable to lower-intensity events, with increased variability in results due to smaller sample sizes.
With higher representative storm temperatures, we find a clear systematic shift to increased, front-loaded temporal patterns of precipitation, particularly in the tropics, which contributes to the intensification of embedded precipitation bursts. Hydrological applications reliant on temporal pattern inputs, such as design flood estimation, may need to consider if temporal characteristics obtained from historical data are representative of storms at higher temperatures. The increased front loading of high-intensity precipitation events will likely result in higher flood peaks, particularly for shorter-duration events, while a reduction of rear-loaded, longer-duration precipitation events can potentially impact long-term runoff statistics of volume-sensitive systems.
Acknowledgments.
Conrad Wasko receives funding from the Australian Research Council (ARC) project DE210100479. This research was supported by the ARC Discovery project DP200101326 and by industry support from Hydro Tasmania, Melbourne Water, Murray-Darling Basin Authority, Queensland Department of Natural Resources Mines and Energy, Seqwater, Snowy Hydro, Sunwater, West Australian Water Corporation, and WaterNSW.
Data availability statement.
Observed data were obtained from the Australian Bureau of Meteorology and can be found online (http://www.bom.gov.au/climate/data/stations/).
APPENDIX
Percentage Front-Loaded Events per Year
Figure A1 is a reproduction of Fig. 3 using the mean percentage of front-loaded events per year and the median representative storm dewpoint. Figure A2 shows the mean percentage front-loaded precipitation events per year across Australia based on station observations.
Map of trends in (a) the mean annual percentage of front-loaded events per year and (c) the median representative storm dewpoint across Australia over the period 1960–2016 using station-based observations. Triangles and inverse triangles represent increasing and decreasing trends at station locations, respectively, with black-bordered triangles indicating stations with significant trends at p < 0.05. (b),(d) Stacked histogram of observed station location trends in (b) the mean annual percentage of front-loaded events per year and (c) the median representative storm dewpoint.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
Mean percentage front-loaded precipitation events per year across Australia for the period 1960–2016 using station-based observations.
Citation: Journal of Climate 36, 18; 10.1175/JCLI-D-22-0694.1
REFERENCES
Beck, H. E., N. E. Zimmermann, T. R. McVicar, N. Vergopolan, A. Berg, and E. F. Wood, 2018: Present and future Köppen-Geiger climate classification maps at 1-km resolution. Sci. Data, 5, 180214, https://doi.org/10.1038/sdata.2018.214.
Berg, P., and J. O. Haerter, 2013: Unexpected increase in precipitation intensity with temperature—A result of mixing of precipitation types? Atmos. Res., 119, 56–61, https://doi.org/10.1016/j.atmosres.2011.05.012.
Berg, P., J. O. Haerter, P. Thejll, C. Piani, S. Hagemann, and J. H. Christensen, 2009: Seasonal characteristics of the relationship between daily precipitation intensity and surface temperature. J. Geophys. Res., 114, D18102, https://doi.org/10.1029/2009JD012008.
Berg, P., C. Moseley, and J. O. Haerter, 2013: Strong increase in convective precipitation in response to higher temperatures. Nat. Geosci., 6, 181–185, https://doi.org/10.1038/ngeo1731.
Contractor, S., M. G. Donat, and L. V. Alexander, 2021: Changes in observed daily precipitation over global land areas since 1950. J. Climate, 34, 3–19, https://doi.org/10.1175/JCLI-D-19-0965.1.
Denson, E., C. Wasko, and M. C. Peel, 2021: Decreases in relative humidity across Australia. Environ. Res. Lett., 16, 074023, https://doi.org/10.1088/1748-9326/ac0aca.
Donat, M. G., A. L. Lowry, L. V. Alexander, P. A. O’Gorman, and N. Maher, 2016: More extreme precipitation in the world’s dry and wet regions. Nat. Climate Change, 6, 508–513, https://doi.org/10.1038/nclimate2941.
Emori, S., and S. J. Brown, 2005: Dynamic and thermodynamic changes in mean and extreme precipitation under changed climate. Geophys. Res. Lett., 32, L17706, https://doi.org/10.1029/2005GL023272.
Fowler, H. J., C. Wasko, and A. F. Prein, 2021a: Intensification of short-duration rainfall extremes and implications for flood risk: Current state of the art and future directions. Philos. Trans. Roy. Soc., 379, 20190541, https://doi.org/10.1098/rsta.2019.0541.
Fowler, H. J., and Coauthors, 2021b: Anthropogenic intensification of short-duration rainfall extremes. Nat. Rev. Earth Environ., 2, 107–122, https://doi.org/10.1038/s43017-020-00128-6.
Gaál, L., P. Molnar, and J. Szolgay, 2014: Selection of intense rainfall events based on intensity thresholds and lightning data in Switzerland. Hydrol. Earth Syst. Sci., 18, 1561–1573, https://doi.org/10.5194/hess-18-1561-2014.
Gocic, M., and S. Trajkovic, 2013: Spatio-temporal patterns of precipitation in Serbia. Theor. Appl. Climatol., 117, 419–431, https://doi.org/10.1007/s00704-013-1017-7.
Guerreiro, S. B., H. J. Fowler, R. Barbero, S. Westra, G. Lenderink, S. Blenkinsop, E. Lewis, and X.-F. Li, 2018: Detection of continental-scale intensification of hourly rainfall extremes. Nat. Climate Change, 8, 803–807, https://doi.org/10.1038/s41558-018-0245-3.
Hettiarachchi, S., C. Wasko, and A. Sharma, 2018: Increase in flood risk resulting from climate change in a developed urban watershed—The role of storm temporal patterns. Hydrol. Earth Syst. Sci., 22, 2041–2056, https://doi.org/10.5194/hess-22-2041-2018.
Ho, M., R. Nathan, C. Wasko, E. Vogel, and A. Sharma, 2022: Projecting changes in flood event runoff coefficients under climate change. J. Hydrol., 615, 128689, https://doi.org/10.1016/j.jhydrol.2022.128689.
IPCC, 2021: Climate Change 2021: The Physical Science Basis. V. Masson-Delmotte et al., Eds., Cambridge University Press, 2391 pp., https://www.ipcc.ch/report/ar6/wg1/.
Kendall, M. G., 1948: Rank Correlation Methods. Griffin, 160 pp.
Kendon, E. J., N. M. Roberts, H. J. Fowler, M. J. Roberts, S. C. Chan, and C. A. Senior, 2014: Heavier summer downpours with climate change revealed by weather forecast resolution model. Nat. Climate Change, 4, 570–576, https://doi.org/10.1038/nclimate2258.
Koenker, R., 1994: Confidence intervals for regression quantiles. Asymptotic Statistics, P. Mandl and M. Hušková, Eds., Springer-Verlag, 349–359.
Koenker, R., and G. Bassett, 1978: Regression quantiles. Econometrica, 46, 33–50, https://doi.org/10.2307/1913643.
Long, K., D. Wang, G. Wang, J. Zhu, S. Wang, and S. Xie, 2021: Higher temperature enhances spatiotemporal concentration of rainfall. J. Hydrometeor., 22, 3159–3169, https://doi.org/10.1175/JHM-D-21-0034.1.
Mann, H. B., 1945: Nonparametric tests against trend. Econometrica, 13, 245–259, https://doi.org/10.2307/1907187.
Martins, D. S., T. Raziei, A. A. Paulo, and L. S. Pereira, 2012: Spatial and temporal variability of precipitation and drought in Portugal. Nat. Hazards Earth Syst. Sci., 12, 1493–1501, https://doi.org/10.5194/nhess-12-1493-2012.
Molnar, P., S. Fatichi, L. Gaál, J. Szolgay, and P. Burlando, 2015: Storm type effects on super Clausius–Clapeyron scaling of intense rainstorm properties with air temperature. Hydrol. Earth Syst. Sci., 19, 1753–1766, https://doi.org/10.5194/hess-19-1753-2015.
Myhre, G., and Coauthors, 2019: Frequency of extreme precipitation increases extensively with event rareness under global warming. Sci. Rep., 9, 16 063, https://doi.org/10.1038/s41598-019-52277-4.
Nathan, R., D. Stephens, M. Smith, P. Jordan, M. Scorah, D. Shepherd, P. Hill, and B. Syme, 2016: Impact of natural variability on design flood flows and levels. 37th Hydrology & Water Resources Symposium 2016,, Queenstown, New Zealand, Engineers Australia, 335–345, https://search.informit.org/doi/epdf/10.3316/informit.688989822440058.
Peleg, N., C. Skinner, S. Fatichi, and P. Molnar, 2020: Temperature effects on the spatial structure of heavy rainfall modify catchment hydro-morphological response. Earth Surf. Dyn., 8, 17–36, https://doi.org/10.5194/esurf-8-17-2020.
Rajah, K., T. O’Leary, A. Turner, G. Petrakis, M. Leonard, and S. Westra, 2014: Changes to the temporal distribution of daily precipitation. Geophys. Res. Lett., 41, 8887–8894, https://doi.org/10.1002/2014GL062156.
Sen, P. K., 1968: Estimates of the regression coefficient based on Kendall’s tau. J. Amer. Stat. Assoc., 63, 1379–1389, https://doi.org/10.1080/01621459.1968.10480934.
Sharma, A., C. Wasko, and D. P. Lettenmaier, 2018: If precipitation extremes are increasing, why aren’t floods? Water Resour. Res., 54, 8545–8551, https://doi.org/10.1029/2018WR023749.
Sugiyama, M., H. Shiogama, and S. Emori, 2010: Precipitation extreme changes exceeding moisture content increases in MIROC and IPCC climate models. Proc. Natl. Acad. Sci. USA, 107, 571–575, https://doi.org/10.1073/pnas.0903186107.
Theil, H., 1950: A rank-invariant method of linear and polynomial regression analysis. Indag. Math., 12, 85–91.
Visser, J. B., C. Wasko, A. Sharma, and R. Nathan, 2020: Resolving inconsistencies in extreme precipitation-temperature sensitivities. Geophys. Res. Lett., 47, e2020GL089723, https://doi.org/10.1029/2020GL089723.
Visser, J. B., C. Wasko, A. Sharma, and R. Nathan, 2021: Eliminating the “hook” in precipitation–temperature scaling. J. Climate, 34, 9535–9549, https://doi.org/10.1175/JCLI-D-21-0292.1.
Wang, G., D. Wang, K. E. Trenberth, A. Erfanian, M. Yu, and G. Michael, 2017: The peak structure and future changes of the relationships between extreme precipitation and temperature. Nat. Climate Change, 7, 268–274, https://doi.org/10.1038/nclimate3239.
Wasko, C., and A. Sharma, 2015: Steeper temporal distribution of rain intensity at higher temperatures within Australian storms. Nat. Geosci., 8, 527–529, https://doi.org/10.1038/ngeo2456.
Wasko, C., and A. Sharma, 2017: Continuous rainfall generation for a warmer climate using observed temperature sensitivities. J. Hydrol., 544, 575–590, https://doi.org/10.1016/j.jhydrol.2016.12.002.
Wasko, C., and R. Nathan, 2019: Influence of changes in rainfall and soil moisture on trends in flooding. J. Hydrol., 575, 432–441, https://doi.org/10.1016/j.jhydrol.2019.05.054.
Wasko, C., A. Sharma, and F. Johnson, 2015: Does storm duration modulate the extreme precipitation-temperature scaling relationship? Geophys. Res. Lett., 42, 8783–8790, https://doi.org/10.1002/2015GL066274.
Wasko, C., J. B. Visser, R. Nathan, M. Ho, and A. Sharma, 2022: Automating rainfall recording: Ensuring homogeneity when instruments change. J. Hydrol., 609, 127758, https://doi.org/10.1016/j.jhydrol.2022.127758.
Westra, S., and Coauthors, 2014: Future changes to the intensity and frequency of short-duration extreme rainfall. Rev. Geophys., 52, 522–555, https://doi.org/10.1002/2014RG000464.
