Previous studies have indicated that extreme precipitation intensity is increasing over time, and has been attributed to anthropogenic warming. Generally these studies have limited analyses to data from daily rainfall totals. We extend those studies by examining characteristics associated with storms of varying duration. We find that significant differences exist in the character of long-duration storms (those of twenty consecutive hours or more) from 1948 to 2004. Specifically we find that, although long-duration storms are becoming wetter, (a) they are occurring less frequently and, consequently, comprising a progressively smaller proportion of the total storm number, and (b) they are contributing a smaller proportion of the total rainfall. Geographically, these storms are more likely to influence the Gulf States (particularly in autumn) and the central west coastal area of northern California. Fundamentally, this study suggests that evaluating precipitation over daily time frames may not capture the full complexities in extreme rainfall events.
 Such studies do not easily capture the complex character of the storms given that, in general, they are based on daily precipitation totals. For example, even though two storms may produce equivalent amounts of rainfall, the resulting hydrologics of an intense storm lasting a few hours are different from that of a storm lasting many more hours but is less intense. Conversely, if one examines storms for a certain specified time interval (such as one day), the internal distribution of the precipitation within that time interval is not known.
 For this study, we have examined extreme rainfall as precipitation fundamentally expressed as a function of the duration. For example, a storm producing hourly measurable precipitation (≥0.254 mm per hour) for three consecutive hours was classified as a three-hour storm while a storm producing hourly measurable precipitation for thirty consecutive hours was classified as thirty-hour storm. For a given analysis (e.g., trend analysis or spatial mapping), we computed mean precipitation for the averaging period (e.g., 1 hour rains, 2 hour rains, etc.) for each station or across the network. These averaging periods permit study of the spatial and temporal variations in storm character of varying duration thereby revealing an additional dimension of precipitation changes over time.
2. Hourly Precipitation Data
 In order to conduct these analyses, we first collected hourly rainfall from 89 first-order National Weather Service stations in continuous operation for the period 1948 to 2004. We categorized event data by two indicators: (a) the overall continuous duration of the storm and (b) the total amount of precipitation recorded for that duration of rainfall. The 89-station network produced 624,607 separate precipitation events ranging from 1 hour to 44 hours. The number of storms of greater than 44-hour duration over the network and length of record were extremely limited and therefore were removed from consideration in this study. For certain statistical tests, we then segregated these events by date of occurrence into an early data period (1948–1975) and a late data period (1976–2004); the early period represents a time of hemispheric cooling (or no change in temperature) while the later period is marked by hemispheric warming. A similar number of events occurred in each of the two time periods.
3. Spatial and Temporal Analyses
 For the network of 89 first-order stations, a distinct and significant difference is evident between the early and later periods of record for precipitation totals that are of storm duration 20 hours or more (Figure 1). Storms of duration less than 20 hours have roughly the same amount of precipitation in the earlier and later portions of the records. For storms of 20 hours or greater, the early period had a mean of 51.46 mm of precipitation while long-duration storms of the later period demonstrated a mean of 55.75 mm. This difference creates a t-value of 2.19 (ρ = 0.03). This finding provides a new verification of Karl and Knight's  and many others' conclusions indicating an increase in extreme rainfall events. However, in contrast to these studies, we demonstrate that this tendency for a greater extreme rainfall over time is also a function of the duration of the rainfall. The average rainfall of short-duration storms (less than 20 hours) did not vary significantly between the early and later halves of the last fifty years. It is only in the long-duration storms that a marked increase in rainfall production is evident. Consequently, for this study, we define ‘extreme rainfall events’ with specific reference to duration as those storms of 20 hours or greater in duration.
 Because several of the statistical techniques used in our study assumed that the data are normally distributed (a Gaussian distribution), we tested the precipitation associated with long-duration (20 hours or more) storms and for storms of all durations for this property using the standardized coefficients of skewness, z1, and kurtosis, z2, calculated as:
where the resulting z values are compared against a t-value deemed appropriate for a selected level of confidence. If the absolute value of z1 or z2 exceeds the selected value of t, a significant deviation from the normal distribution is confirmed. Otherwise, no statistically significant deviation from a normal distribution is determined (the null hypothesis that the samples came from a normal distribution cannot be rejected). We also used the Kolmogorov-Smirnov one-sample test to further evaluate the normality of each variable. These tests revealed no significant deviations from normality at the ρ = 0.01 or ρ = 0.05 levels of confidence.
 Consequently, we defined “long duration storms” as those of 20 hours or more, and produced a map showing the average amount of precipitation for each long duration storm (Figure 2). The results reveal a marked gradient near the Gulf Coast and Florida with high precipitation amounts rapidly decreasing inland from the coast. The seasonal distribution of precipitation of long-duration storms is seen in Figure 3.
 The strong likelihood is that many of these heavy long-duration precipitation events are associated with tropical cyclone activity in which the cyclones spatially dissipate rapidly as they move inland (as seen by the stronger autumn gradients in precipitation in Figure 3c). A second area of relatively heavy, long-duration precipitation is found along the central western coast of the United States. This region is frequently struck by late winter and spring decaying frontal storms particularly during El Niño episodes (as seen by the stronger winter and spring gradients in precipitations in Figures 3a and 3b). These frontal storms tend to stagnate over the coastal mountains and, consequently, produce heavy and long-lasting precipitation. In contrast, marked minima in long-duration precipitation events occur across the entire northern states with particular emphasis on the northern Rocky Mountains states. Synoptic systems crossing through this region tend to be faster-moving and therefore less likely to produce long-duration precipitation.
 As seen in Figure 4 for the entire time series, the percentage of total precipitation from long duration storms compared to precipitation from all storms is relatively high in the central west coast area and the Northeast; both areas are frequented by decaying synoptic-scale events. The high overall annual precipitation totals in Florida and the Gulf Coast result in a low relative contribution from long duration storms to the overall precipitation total.
 Beyond the basic climatological patterns associated with the long duration storms, Figure 1 reveals a shift to a higher average precipitation for long-duration storms from the early to late halves of the time period. This shift can be resolved through analyses of (a) the changes in the relative number of long-duration storms in comparison to storms of all durations and (b) the ratio of the amount of rainfall of these long-durations storms in comparison to the rainfall of storms of all durations. For the complete time period (1948–2004) over all stations, storms of 20 hours or more of duration make up 0.53% ± 0.40% of the total number of storms but their rainfall totals comprise 5.24% ± 4.44% of the average precipitation from all storms.
 The strength of this temporal variability can initially be evaluated through the use of linear regression with the year as the independent variable and the average amount of precipitation for storms of 20 hours duration or more as the dependent variable. Linear upward trend from 1948 to 2004 explains 22.4% of the total variance (ρ < 0.01) for the long-duration storms. Repeating the analysis for storms of all durations produced a linear upward trend explaining only 1.1% of the total variance (ρ = 0.46). Additionally, linear regression analysis revealed a downward trend in the number of long-duration storms over the time of record. The trend downward in the number of long-duration storms explains 8.7% of the variance over the 1948–2004 time period (ρ = 0.03), suggesting long-duration storms have become less frequent. In addition to simple regression analysis, we tested for trend using the non-parametric Mann-Kendall trend test that reconfirmed the existence of trend in these data.
 As a second distinct measure of these trends, we statistically compared the means in long duration storm proportions between early (1948–1975) and late (1976–2004) time periods. Student's t-test analyses indicated there are statistically significant differences in long duration precipitation proportions between the two periods. For the early time period, long duration storms comprise 0.62% ± 0.44% of the total number of all duration storms but, conversely, for the later time period, long duration storms comprise 0.39% ± 0.25% of the total number of storms. In short, this test confirms the earlier regression analysis that the relative number of long duration storms is decreasing. Additionally, for the early time period, long duration storms contribute 5.91% ± 4.85% of the precipitation and, in the later time period, contribute 3.90% ± 2.75% of the total precipitation of the storms. Thus, while the t-test indicates that the relative contribution of long duration storms has decreased, it has not decreased to the extent of the reduction in relative number. So while long-duration storms have decreased in number, the remaining smaller number of such storms has increased their relative contribution to the overall precipitation; recall that the regression analyses indicated that long duration storms have increased their precipitation total per storm over the length of record. These differences in the relative number and relative rainfall contribution of long-duration storms in comparison to storms of all durations are statistically significant (for percent frequency, t = −4.39, p < 0.01; for percent total, t = −3.47, p < 0.01).
 The largest differences between the number of long-term precipitation events in the early (1949–1975) and late (1976–2004) time periods are apparent in the central west coast area and the Northeast; both areas are frequented by decaying synoptic-scale events (Figure 5). The recent time period displays few long-duration events in these areas. The Gulf coastal region, in terms of long-duration precipitation events, has remained relatively constant between the early and late time periods.
4. Summary and Conclusions
 Previous studies have indicated that extreme rainfall intensity is increasing over time, and has been attributed to anthropogenic warming. A potential limitation of many of those these studies, however, is the use of data computed using a fixed duration for the storm (normally averaged over a 24-hour daily period). We have created a different classification approach than is commonly used in which storms are segregated by the relative duration of the storm. This allows us to identify new and important characteristics of precipitation climatology.
 When one examines storm amounts from hourly precipitation data, distinct differences are apparent in amount of rain associated with storms of varying duration. In particular, although long-duration storms (20 hours or more in duration) are becoming wetter over the last half century, (a) they comprise a progressively smaller proportion of the total number of storms and (b) they contribute a smaller proportion of the total rainfall. Geographically, these long-duration storms are more likely to influence the Gulf States and the central West Coastal area of northern California.
 By classifying storms by their duration rather than by their precipitation amounts, we are able to examine a new aspect of storm climatology. Long-duration storms can have a profound impact on population centers through a greater potential for flooding because of saturated soil conditions (e.g., Hurricane Mitch's massive flooding of Nicaragua). Identification and understanding of such storms' characteristics becomes more important as our findings point to an increasingly wetter nature of these storms. The IPCC's [2007, p. 12] assertion that “based on a range of models, it is likely that future tropical cyclones will [produce] more heavy precipitation…” might link to our findings that long duration (20 hours or more of duration) storms are particularly concentrated along the Gulf Coast (and therefore likely of tropical origin), prevalent more in the summer, and that these storms have become wetter in recent decades.