Sharp rises in large-scale, long-duration precipitation extremes with higher temperatures over Japan

The Clausius–Clapeyron (C–C) relation suggests atmospheric moisture capacity and, by extension, extreme precipitation should increase by about 7% per °C absent major circulation changes. Observations, however, show short-duration extremes often increase at super C–C rates, especially at higher temperatures, with implications for flash floods. Prior work indicates super C–C scaling is largely linked to convective, short-lived events, while synoptic-scale systems (e.g., atmospheric rivers, ARs) also produce extreme, persistent precipitation and widespread flooding in mid-latitudes such as East Asia. Daily (fixed-interval) analyses can mask event characteristics, particularly for long-lasting storms. This study asks how synoptic patterns and event duration modulate precipitation–temperature scaling in Japan’s warm season. Using event-based data, the authors classify storms by duration to separate local convective from synoptic-scale, AR-like events, and assess how peak intensity and accumulated precipitation extremes scale with temperature, elucidating risks for floods and landslides.

Previous studies documented super C–C scaling of sub-daily precipitation extremes and linked it predominantly to convective events. Event typologies have been inferred from cloud type, lightning, circulation patterns, and duration. While convective precipitation often drives super C–C behavior, synoptic-scale storms and atmospheric rivers also cause extremes with long durations and large accumulations in mid-latitudes, including East Asia, Western Europe, and western North America. In East Asia’s warm season, interactions between monsoon flow and synoptic systems favor persistent precipitation with embedded convective bursts. Prior scaling analyses in this region even showed longer-duration (daily) extremes can scale more strongly than shorter durations, but the role of synoptic context for persistent events remained unclear. Event-based approaches are recommended to capture storm lifecycle and background circulation and to address biases of daily statistics.

Data and study period: 10-minute rain gauge observations from 646 AMeDAS stations across Japan (mostly below 500 m elevation), covering 1994–2018 warm seasons (May–September). Daily mean surface air temperature from the same network was paired with events. Synoptic conditions were taken from JRA-55 reanalysis (6-hourly, 1.25° grid). Event detection: Individual precipitation events were defined using a minimum inter-event dry spell of at least 3 hours. Sensitivity tests with 1 and 6 hours showed negligible impact on scaling. Events with total amount >1.0 mm were retained. Days influenced by tropical cyclones were excluded using JMA best-track data, defined as days when a TC center was within 1000 km of a station. Event properties: For each detected event, three properties were computed: duration; peak hourly intensity (maximum 60-minute rolling total computed every 10 minutes within the event); and total-event precipitation accumulation. In total, 790,554 events were detected across 25 warm seasons. Event classification: Short-duration events were defined as events lasting <5 h whose peak intensity occurred in the afternoon/evening (12:00–00:00 local time), reflecting local thunderstorms. Long-duration events were those lasting >10 h. Mid-duration events (5–10 h) and short-duration events with morning peaks were excluded from the main short/long comparison to minimize mixing of storm types. Diurnal patterns confirmed distinct mechanisms: short-duration extremes peak in late afternoon; long-duration extremes dominate midnight–morning. Synoptic composites: For days when any station’s event peak intensity exceeded the 99th percentile within each temperature bin, JRA-55 fields were composited to characterize large-scale moisture flux, convergence, and 500 hPa temperature anomalies. This confirmed long-duration events align with AR-like synoptic patterns. Scaling analysis: Event precipitation metrics (99th percentiles of peak hourly intensity or total-event accumulation) were related to daily mean surface temperature via binning: 2 °C-wide, overlapping bins stepped by 1 °C, using temperatures 11–27 °C to ensure >80% station coverage per bin. If events spanned multiple days, the average of those days’ temperatures was used. Uncertainties (95% confidence intervals) were estimated via bootstrap with 1000 resamples using half-sample size per bin, assuming normality of the bootstrap percentile distribution. Exponential regression was applied to log-transformed extremes versus median bin temperature: log(P)=β0+β1 T, with scaling rate ΔP(% per °C)=100×(e^{β1}−1).

- Duration-based classification separated storm types: short-duration extremes peaked in late afternoon, while long-duration extremes peaked midnight–early morning, consistent with local convective versus synoptic-scale (AR-like) systems. - Synoptic patterns during extreme days for long-duration events showed strong southwesterly, vertically integrated moisture flux and convergence typical of atmospheric rivers during Baiu/Meiyu season. - Extreme hourly peak intensity versus temperature: 99th-percentile scaling rates were 11.1% °C⁻¹ for long-duration events, 9.8% °C⁻¹ for short-duration events, and 9.7% °C⁻¹ for all events. Upper-temperature bins exhibited peak-like behavior (declines at highest T) likely due to reduced relative humidity at higher temperatures. - Total-event precipitation for long-duration events showed super C–C scaling between 18–24 °C. Median scaling rates were approximately 11.8% °C⁻¹ (events selected by extreme peak intensity) and 11.4% °C⁻¹ (events selected by extreme total accumulation). Event durations did not vary systematically with temperature in this range, suggesting the scaling arises from intensity and storm structural changes (e.g., intensified convective cells, increased convective area, possibly stronger stratiform components). - In the cold regime (<18 °C), extremes of accumulated precipitation and event durations were largely insensitive to temperature. In the warm regime (>24 °C), extremes tended to decrease with temperature, consistent with moisture supply limitations and shorter event durations, and subject to larger sampling uncertainty due to fewer events. - Long-duration extreme events predominantly occurred in May–July (Baiu/Meiyu season), when sustained monsoonal moisture maintains near-saturation conditions for local temperatures.

Findings show that long-duration, synoptic-scale precipitation events in Japan are highly sensitive to local surface temperatures, exhibiting super C–C scaling in both peak intensity and total-event accumulation within an 18–24 °C range. The AR-like moisture transport during Baiu/Meiyu likely maintains near-saturated conditions, enhancing convective activity embedded within larger stratiform systems. Peak-like declines at the highest temperatures align with reduced relative humidity from land–ocean temperature contrasts. The results imply that as temperatures rise, persistent, large-scale storms may intensify disproportionately, elevating risks of widespread flooding and landslides. With climate models projecting increased AR frequency in the Northern mid-latitudes, these scaling behaviors suggest notable increases in hydrological hazards in mid-latitude coastal regions. While peak-shaped scaling raises questions about future limits, model projections indicate both the peak intensity and the temperature at which the peak occurs may increase in warmer climates, implying current observed peaks may not cap future extremes. Continuous monitoring of precipitation and atmospheric moisture, and refined modeling, are essential for accurate future risk assessment.

By applying an event-based framework and duration-based classification, the study demonstrates that long-duration, AR-like precipitation events exhibit super C–C scaling in both hourly peak intensity and total-event accumulation at higher temperatures, exceeding thermodynamic expectations. This indicates that long-lasting storms respond more strongly to warming than short-lived convective storms, heightening flood and landslide risk in mid-latitude coastal regions. Future work should examine how storm spatial structures change with temperature and leverage convection-permitting climate models to improve projections of large-scale, long-duration precipitation extremes, alongside sustained observations to track potential shifts in scaling behavior.

- Sampling uncertainty at the warmest temperatures due to fewer events leads to wide confidence intervals and peak-like behavior that may be sensitive to data scarcity. - Event classification excluded mid-duration and morning-peak short-duration events from main analyses to avoid mixing storm types; while supplementary analyses exist, residual classification uncertainty remains. - Use of daily mean surface temperature (averaged across days for multi-day events) may not perfectly represent the temperature governing convective intensity or moisture source regions. - Exclusion of tropical cyclone days avoids dynamical influences but limits generalizability to all extreme precipitation contexts in Japan. - Station network predominantly at elevations below 500 m and pooling across stations may mask elevation or local effects; synoptic reanalysis resolution (1.25°) may blur mesoscale features. - Mechanistic attribution of super C–C scaling (e.g., detailed convective dynamics) is beyond the scope of this study.