Statistical Analysis
Spatial Distribution and Interpretation of Maximum 1-Hour and Daily Precipitation
To evaluate both the short-term intensity and cumulative impact of the July 2024 heavy rainfall event, we adopt two core precipitation metrics: Maximum 1-Hour Precipitation (Max1h) and Maximum Daily Precipitation (MaxDaily).
Max1h is defined as the highest hourly precipitation value recorded at each station during the seven-day event period (16–22 July), based on data from automatic weather stations (AWS) with hourly temporal resolution. This metric is particularly effective in capturing localized convective bursts that are characteristic of short-duration extreme events and serves as a reference for percentile-based and recurrence interval analyses.
In parallel, MaxDaily represents the single largest daily precipitation total observed at each station within the same period. This metric is suitable for quantifying the hydrological consequences of prolonged rainfall episodes and, when analyzed alongside Max1h, enables an integrated understanding of both intensity and accumulation characteristics.
Figure 1
Figure 2
Table 1
Figure 1 illustrates the spatial distribution of Max1h, while Figure 2 presents that of MaxDaily. Both maps show a distinct concentration of high values over central-western Korea, indicating that intense hourly rainfall and significant daily totals co-occurred in this region. Several stations exceeded 60 mm hr⁻¹, with some—such as Paju, Seosan, and Namhae—recording more than 80 mm in a single hour. These peaks occurred mainly on 19 or 20 July as abrupt pulses rather than gradual developments, suggesting the influence of localized convective cells, cell mergers, or mesoscale systems such as squall lines.
MaxDaily distributions further support this interpretation. At multiple sites, including Paju, Seosan, and Namhae, daily totals exceeded 300 mm, far surpassing climatological norms. The concurrence of high Max1h and MaxDaily values implies that short-lived convective extremes were embedded within broader, sustained rainfall systems. This coupling reflects not only isolated convective storms but the organized passage of mesoscale convective systems (MCS), which are capable of producing both intense and prolonged precipitation.
Table 1 summarizes the top five stations that ranked highest in both Max1h and MaxDaily. Stations such as Paju (101.0 mm, 385.7 mm), Namhae (80.5 mm, 207.1 mm), and Seosan (81.1 mm, 153.0 mm) recorded exceptional values in both metrics, indicating the potential for significant hydrological impacts. These findings underscore the importance of a dual-metric framework in extreme rainfall assessment, wherein both sub-hourly intensity and multi-hour accumulation are jointly considered.
Station-Level Metrics and
Box-and-Whisker Analysis
To evaluate the extremity of the 16–22 July 2024 rainfall event in a historical context, three diagnostic precipitation metrics were computed at five representative stations using daily to hourly observational data. For each metric, long-term distributions were derived from the 1991–2020 summer seasons (June–August), and visualized using box-and-whisker plots. The 2024 event values are indicated with red markers to enable direct comparison with historical variability.
Figure 3
Table 2
Table 3
For each year, the highest 1-hour precipitation was identified for every summer day and then averaged to obtain a seasonal mean. This metric captures the typical intensity of convective precipitation on a daily basis, reflecting how strong short-duration rainfall episodes are, on average, within a given summer season. In 2024, all selected stations exhibited values that exceeded the 75th percentile of their historical distributions. Notably, Seoul and Heuksando recorded 20.16 mm and 25.02 mm, respectively, both surpassing their historical maxima of 11.79 mm and 9.34 mm (see Figure 3 and Tables 2–3).
Figure 4
Table 4
Table 5
The maximum number of consecutive hours with non-zero precipitation was calculated for each summer season. This metric provides insight into the temporal persistence of rainfall, as longer durations typically reflect synoptic-scale systems or long-lived convective complexes. During the 2024 event, all four stations showed durations well below their long-term medians, with values clustered around or below the 25th percentile. For instance, Daegu recorded 6.83 hours, Jeonju 6.57 hours, Jeju 5.82 hours, and Heuksando 7.08 hours—compared to historical means exceeding 20 hours at all locations (see Figure 4 and Table 4–5).
Figure 5
Table 6
Table 7
To evaluate the occurrence rate of high-intensity rainfall, the number of hourly instances exceeding 30 mm h⁻¹ was counted for each summer season. The 2024 values were relatively low across all stations, ranging from 1 to 2 exceedance events per station: Seoul recorded two, while Gwangju, Mokpo, Suwon, and Yeosu each recorded one (see Figure 5 and Tables 6–7). These counts are generally lower than the historical medians, which typically ranged from 2 to 3 occurrences per season at these locations.
Percentile-Based Extremity
Methodology
To evaluate how extreme the 2024 maximum 1-hour precipitation values were relative to long-term climatology, we calculated percentile ranks for each station. This was done by comparing each station’s 2024 event value against its distribution of annual maximum 1-hour precipitation values from 1991 to 2020. The percentile rank represents the percentage of historical years in which the observed value was lower than or equal to the 2024 value (see equation below).
The calculation uses the "weak" definition, which includes equality, and was implemented using the percentileofscore function from the SciPy library.
Figure 6
Figure 7
Figure 6 shows the spatial distribution of percentile ranks across all stations for the 16–22 July 2024 event. A significant concentration of high percentile values is observed over the western interior and southern coastal regions. Many stations exceed the 95th percentile, with a large cluster surpassing the 99th percentile. This spatial coherence suggests that the event was not composed of isolated anomalies but was instead shaped by organized mesoscale convective systems that affected broad areas simultaneously.
Figure 7 compares historical and event-period percentile ranks at selected high-impact stations. All five stations shown—Boseong-gun, Heuksando, Paju, Jindo-gun, and Seosan—recorded 2024 values that fall at or extremely close to the 100th percentile. In contrast, their historical median percentiles range from 47 to 73, indicating that the 2024 event represented an unprecedented deviation from the local norm.
Taken together, these results confirm that the July 2024 event produced historically extreme hourly rainfall across a wide portion of the Korean Peninsula. The percentiles provide robust statistical evidence that the observed values were the highest in at least three decades at many stations, not as isolated outliers, but as part of a spatially coherent and climatologically exceptional episode.
Return Period via GEV Fitting
Methodology
To estimate the statistical rarity of the 2024 maximum 1-hour rainfall values, we apply Generalized Extreme Value (GEV) distribution fitting to each station’s annual Max1h data over the 1991–2020 base period. The GEV distribution is characterized by three parameters—shape, location, and scale—which are estimated using the method of maximum likelihood.
Once the GEV parameters are fitted, the return period for the 2024 event is computed as the inverse of the survival function evaluated at the event’s Max1h value at each station.
This approach quantifies the exceedance probability of an observed value in terms of how often such an event would be expected under historical conditions.
Figure 8
Figure 9
Figure 8 shows the spatial distribution of return periods derived from the GEV-fitted models for the 16–22 July 2024 event. A distinct cluster of stations across the southwestern and central-western regions exhibits return periods exceeding 50 years, with several surpassing the 100-year threshold. These high return period values suggest that many locations experienced rainfall intensities far beyond those expected within the historical climate regime.
Figure 9 compares historical mean return periods (based on annual maxima from 1991–2020) with the event-specific values for 2024 at five representative stations. Gunsan, Baengnyeongdo, and Bukchangwon all show dramatic increases—from historical means of around 2 years to nearly or above 275 years—demonstrating the extremity of the event. Even at sites with historically longer return periods, such as Uiryeong-gun and Cheongju, 2024 values still show substantial upward deviations.
Taken together, these findings highlight the statistical singularity of the July 2024 rainfall event. The widespread occurrence of return periods exceeding 75, 100, or even 250 years across multiple regions strongly indicates that this event lies far outside the expected range of natural variability. Its spatial coherence, heavy-tailed distribution of station-level return periods, and regional clustering of extremes suggest a compound risk scenario driven by persistent mesoscale forcing.
Regional-Scale Concentration (CI10%)
Methodology
To evaluate the regional-scale spatial asymmetry of the 2024 rainfall event, we employ the Concentration Index (CI10%), which measures the proportion of total precipitation contributed by the top 10% of stations ranked by event-scale maximum 1-hour precipitation. This metric highlights whether a small number of sites disproportionately accounted for the overall hydrologic impact.
A high CI10% indicates pronounced spatial clustering, where a few stations received the majority of rainfall. Such concentration can exacerbate local flood hazards, particularly in urbanized catchments or terrain-constrained basins.
Figure 10
Figure 11
Figure 10 displays the spatial distribution of CI10% across Korea during the 16–22 July 2024 event. Notable peaks appear in Hongseong and central-eastern regions, with some values exceeding 60%. These areas acted as focal points of convective convergence, accumulating extreme rainfall while surrounding areas received significantly less.
Figure 12 compares CI10% values from 2024 against historical distributions at four representative stations. At all locations, the 2024 values exceed the 75th percentile, and in several cases—such as Hongseong and Samcheok—they approach or reach the historical maximum. This strongly suggests that the spatial concentration observed in 2024 was not a product of typical interannual variability but reflects an outlier event in terms of mesoscale organization.
Taken together, the elevated CI10% values and their deviation from historical norms confirm that the 2024 event exhibited exceptional spatial asymmetry at the regional scale. The clustering of high-intensity rainfall in a few hotspots underscores the need for subnational hazard assessments that account for spatially uneven precipitation impacts.
Local-Scale Extremes (LIR)
While CI10% captures broader regional concentration, localized anomalies can be missed without high-resolution comparisons. To address this, we use the Local Intensity Ratio (LIR), which evaluates the deviation of a station’s precipitation relative to its immediate neighborhood within a 30 km radius.
LIR values above 1.0 suggest a station received more rainfall than its neighbors, and values exceeding 1.5 are indicative of strong local hotspots. This metric is particularly useful for identifying localized convective enhancements that may not register at broader scales.
Figure 12
Figure 13
Figure 12 maps LIR values for the 2024 event, revealing distinct hotspots across the central-west, southern coast, and isolated interior zones. Several stations exhibit values above 2.0, with the highest exceeding 4.0, implying localized bursts of rainfall likely driven by terrain effects, land–atmosphere interactions, or storm cell mergers.
Figure 13 presents historical distributions of LIR at key stations, with 2024 values superimposed. Many stations—such as Seosan, Miryang, and Sancheong—record 2024 values well above historical medians. This suggests that local-scale intensification during this event was not only more severe but also more geographically widespread than in typical years.
These findings indicate that the 2024 rainfall event featured an unusually high density of localized extremes, in addition to its broader spatial asymmetry. Such local hotspots represent heightened flash flood risk zones and pose significant challenges for nowcasting, which often struggles to resolve sub-grid convective structures in real time.