This paper examines tropical cyclone (TC) rainfall in the eastern United States from the perspective of documenting accumulated annual water volumes and areas of the precipitation. Volume is a value that merges both rainfall depth and rainfall area into a single metric for each year that can be directly compared between individual years. Area represents the total land area affected by tropical rains. These TC rainfall metrics were then compared to the ENSO and the Atlantic multidecadal oscillation (AMO). Time series of annual TC water volumes show an annual average of 107 km3. The maximum volume was produced in 1985 with 405.8 km3, driven by Hurricanes Bob, Claudette, Danny, Elena, Gloria, Henri, Juan, and Kate as well as by Tropical Storms Henri and Isabel. The lowest TC volume occurred in 1978 with 8.9 km3. ENSO phases did not show any statistical correlation with TC frequency in the eastern United States. However, AMO showed a significant correlation with volume and the number of storms affecting the region. TC rainfall volume and area in the eastern United States showed a strong correlation. However, there are exceptions, whereby 1985 stands out as an exceptional volume year though the area affected is not as impressive. In contrast, 1979 is an example when TCs covered a large area with a corresponding small rainfall volume, in part because of the rapid forward movement of the storms, for example, Hurricanes David and Frederic. Since 1995, TCs have become more numerous, producing larger volumes and affecting larger areas.
Tropical cyclones (TCs) affecting the eastern United States are responsible for causing economic losses and loss of human life (Blake et al. 2007). These events are generally thought of as storms with coastal effects with high winds and surge, but they can also produce heavy rainfall along the coast, as well as much farther inland. As a result, they can be an important contributor to monthly and seasonal rainfall totals during hurricane season. Indeed, this was the case from June to November in the eastern United States as documented by Cry (1967), Knight and Davis (2007), and Nogueira and Keim (2010). However, the total volume of water and total area affected by Atlantic TC rainfall in the eastern United States has never been determined. Implications of TC rainfall input over land include river basin flooding (e.g., from Hurricane Camille in Virginia in 1969), urban flash flooding (e.g., Tropical Storm Allison in Houston, Texas, in 2001), and drought-mitigating rains (e.g., Hurricanes Katrina and Rita across Texas, Louisiana, and Mississippi in 2005). As a result, this paper will address the effects of Atlantic TC rainfall (including storms in the Gulf of Mexico and Mexico) in the eastern United States.
Tropical cyclone activity in the North Atlantic experiences great variability from the intra-annual (e.g., Keim and Robbins 2006), interannual (e.g., Bove et al. 1998), and interdecadal (Gray 2007; Landsea et al. 1999) time scales. Clearly, no two hurricane seasons are identical, but there is some long-term predictability to tropical activity in the North Atlantic basin that stems from atmospheric teleconnections. For example, Landsea et al. (1999) point out that the overall TC activity is associated with North Atlantic sea surface temperatures (SSTs)—termed the Atlantic Multidecadal Oscillation (AMO)—combined with other modulating environmental factors such as El Niño–Southern Oscillation (ENSO). An empirical relationship between Atlantic SSTs and TC activity was identified, whereby warmer Atlantic SSTs enhance the development of TCs (Shapiro and Goldenberg 1998; Molinari and Mestas-Nuñez 2003) and landfalls in the eastern United States (Keim et al. 2007). Furthermore, El Niño events have been shown to reduce TC activity through enhanced upper-tropospheric westerly wind shear over the Caribbean Basin and the equatorial Atlantic, while La Niña events increase activity (Gray 1984; Bove et al. 1998). Hence, the interaction between AMO and ENSO looms particularly large regarding the tropical storm and hurricane effects in the eastern United States. Therefore, the objectives of this study are as follows:
to determine the annual volume of rain produced by tropical storms and hurricanes over the eastern United States;
to determine the total area that received tropical-storm- and hurricane-induced precipitation in the eastern United States each year; and
to determine if teleconnections (AMO and ENSO) are associated with the annual volumes of rain as well as the annual area sizes of the eastern United States that are affected by the tropical cyclones.
2. Data and methods
Separating rainfall into tropical and nontropical cyclone components is a challenging task, as TCs can produce rainfall for hundreds of kilometers from their centers (Larson et al. 2005). However, previous research provides guidance regarding the size of the rainfall swath produced by TCs. For example, Cry (1967) considered rainfall to be tropical within the limits of the TC circulation ranging from less than 100 to more than 800 km, depending on each storm’s rainfall characteristics. Rao and Macarthur (1994) gridded each storm’s rainfall swath and determined the rainfall within each grid cell. Gleason (2006) used a simple partition method, classifying any rainfall ≤600 km from the center of the storm as “tropical.” Englehart and Douglas (2001) found that in 90% of cases, TC rainfall occurs within 600 km from the center. In the end, they used a 550-km radius from the center of each storm to assign surface weather stations as receiving TC-derived rainfall data. Knight and Davis (2007) included all rainfall data associated with the tropical storm, even after becoming extratropical or associating with a frontal system. This approach yields a relatively high contribution of TC rainfall in monthly totals.
In this study, a conservative approach was used in considering tropical rainfall related to the distance of the center of TC. Using guidance from these previous efforts, a 500-km (∼310 miles) radius centered on each storm was used to delineate the area affected by tropical precipitation. The TC rain shield was assumed symmetric around the storm center, and sensitivity analysis performed by Larson et al. (2005) found this to be reasonable. Rainfall totals, however, tend to be larger on the right side of the track because of the onshore flow of moisture (Jones 1987; Powell 1987) and smaller on the left side because of the entrainment of drier air from the continent. However, when TCs begin interacting with extratropical weather systems, rainfall patterns get increasingly asymmetric (Larson et al. 2005). The 500-km criterion is an operational definition to reduce the influences of these systems.
TC-related precipitation was considered as any precipitation produced by landfalling tropical storms and hurricanes as well as for those storms that tracked within an offshore distance of 500 km (determined by the ArcMap buffer), whereas land lies within the tropical cyclone rain swath. The 6-hourly data available through the National Hurricane Center (NHC) in the Atlantic basin Hurricane Database (HURDAT; Jarvinen et al. 1984) was used to determine the timing and location of the storm to properly appropriate the rainfall. If a storm was denoted in this database as either subtropical, extratropical, or as a tropical depression, then the associated rainfall with those systems was no longer included in this analysis. Remnant lows can in some instances produce very high rainfall totals (i.e., remnants of Tropical Storm Amelia in 1978 or Tropical Storm Erin in 2007); however, these systems often interact with extratropical weather systems as well and thus were not included in this analysis. Compared to other studies that include remnants of tropical storms (i.e., Knight and Davis 2007), these methods would underestimate rainfall totals. We also note that prior to 1968, there is no designation of subtropical storms in HURDAT, and a small percentage of events that would now be subtropical are designated as tropical (Landsea et al. 2008). As performed by Cry (1967), rainfall data were divided into two subsets: one called tropical rainfall (TR), which includes the accumulated TC daily rainfall for each site for each month, and the other called nontropical rainfall (NTR), which is derived as the difference between TR and U.S. Historical Climatology Network (USHCN) monthly precipitation totals at each site.
Rainfall data from 1960 to 2007 were extracted from monthly rainfall observations from the USHCN monthly precipitation and temperature data (Williams et al. 2007). This dataset contains 1221 high-quality stations from the U.S. Cooperative Observer Network within the 48 contiguous United States, and it has undergone extensive quality assurance checks and includes only the most reliable and unbiased long-term records. Of the 1221 stations, monthly precipitation data were obtained from a subset of these data totaling 717 stations in the eastern United States (Fig. 1). TC rainfall contributions are less than 1% west of Minnesota and Iowa, and north of Kansas (Cry 1967; Knight and Davis 2007); hence, these areas were not considered in this study. Daily precipitation data were obtained for the same 717 stations through the Southern Regional Climate Center (SRCC) Applied Climate Information System (ACIS; see Hubbard et al. 2004; ACIS 2009) to update the USHCN dataset to 2007 for the delineated region. The source for both datasets (UNHCN and ACIS) is National Oceanic and Atmospheric Administration (NOAA)’s National Climatic Data Center. A sample of stations was cross-checked from both networks, and very few differences between the datasets were found. When differences arose in our perusal, it was because there were missing data in the ACIS dataset. In such cases, USHCN made efforts to fill in short periods of missing data using nearby stations.
One incongruity we addressed is that the HURDAT dataset provides data in 6-hourly intervals, yet the rainfall data are provided daily. Most National Weather Service cooperative stations take observations in the morning, near 1200 UTC, thereby representing an observational day of 1200 UTC from the previous day to 1200 UTC on the day of the observation. When a storm transitions to extratropical, or is downgraded to a depression within this 24-h “observational day” window of time, the entire daily rainfall logged at the end of the observational day was included in this analysis. This practice was consistent for all stations regardless of the time of observation, since the vast majority of stations have morning observation times. Basically, the choice was either to be inclusive or exclusive of the tropical rainfall in this circumstance, and we chose to be inclusive. Our decision was based on the findings that the transition declaration by the National Hurricane Center is a subjective decision and that there is considerable uncertainty in the accuracy of the specific point in time when this transition occurs (Hart and Evans 2001), hence including the entire observational day of rainfall when this occurred seemed reasonable.
Environmental Systems Research Institute, Inc. (ESRI)’s ArcMap 9.2 was used to plot and display all weather stations by year, subset those stations by each storm’s buffer region, and perform spatial analyses (Johnston et al. 2001). This was accomplished by importing hurricane track shapefiles from NOAA’s Coastal Services Center. These data are generated from the NHC’s HURDAT dataset. For each storm track, a 500-km buffer was produced. A simple kriging quartile tool was then used to create an interpolated TC rainfall surface (Chapman and Thornes 2003). Kriging is a stochastic technique—similar to inverse distance weighting (IDW)—that uses linear combinations of weight at known points to estimate values to unknown points. Kriging has been used effectively to interpolate rainfall data (Earls and Dixon 2007; Mirás-Avalos et al. 2007).
The Lambert equal-area conic projection was used to minimize errors in calculating the rainfall areas and volumes of TCs. The first step in this endeavor selected all USHCN stations in the study area that fell within 500 km of the center of each storm’s position and then computed the accumulated rainfall total related to TCs at each station. The second step interpolated a rainfall pattern using simple kriging, which is then converted to a raster format (cell size = 2 km × 2 km). Lastly, the raster pattern over the buffer area is used to calculate volume by implementing the following formula:
annual rainfall volume (km3) = annual pixel average TC rainfall (km) × annual number of pixels × total buffer area by year (km2).
After interpolating the rainfall pattern over land in the eastern United States, the depth and area of rainfall for each hurricane season is converted into the total TC-induced volume of water. Volume, therefore, is a value that merges both rainfall depth and rainfall area into a single metric for each year that can be directly compared between individual years. It is presented in units of cubic kilometers.
TC rainfall is characterized by interannual and interdecadal variability. Those variations may be related to climate teleconnections such as ENSO and AMO. In this paper, effects of ENSO and AMO in the TC rainfall over the eastern United States are examined. Listed next are seven variables related to TC rainfalls chosen to test for correlations with ENSO and AMO: Month_tot represents all rainfall during hurricane season extending from June–November rainfall, including both tropical and nontropical rainfall; TC_tot is the total amount of rain by month and season produced by TCs only; Non_tc tot is the total rainfall produced by nontropical systems and is derived by subtracting TC_tot from Month_tot; Percentage_tc is the percentage of total rainfall produced by TCs; Number of Storms is the number of storms by year affecting the eastern United States; VOLUME represents the total rainfall volume produced by TCs; AREA is the area in square kilometers affected by the TC rain shield (based on the 500-km buffer).
Henderson-Sellers et al. (1998) pointed out that El Niño events are related to the seasonal frequency and interannual variations of tropical cyclone activity. Furthermore, several studies relate ENSO with tropical cyclone activity (Landsea et al. 1996; Bove et al. 1998; Landsea et al. 1999; Pielke and Landsea 1999; Tonkin et al. 1997; Enfield et al. 2001; Gray 2007). Pielke and Landsea (1999) noted that during the cold phase ENSO (La Niña) events, the United States experiences a larger number of TCs and more damage compared to the warm ENSO (El Niño) phases.
The ENSO monthly SST anomalies used in this study were obtained from the Niño-3.4 region (5°N–5°S, 170°–120°W) from NOAA’s Climate Prediction Center (CPC), based on a threshold of 0.3°C, as suggested by Trenberth (1997). The ENSO index is identified using 6-month (June–November) averages of SST anomalies. El Niño was defined as when the SST anomaly average was greater than 0.3°C and a La Niña as when the SST anomaly average was at least 0.3°C below average. Figure 2 shows the ENSO-3.4 SST anomaly time series and the ENSO phases (cold, warm, and neutral).
AMO is defined by the SST between warm and cold phases within a 65–80-yr cycle (Kerr 2000). AMO showed warm phases from periods 1860–80 and 1940–60 and cold phases from periods 1905–25 and 1970–94. A new AMO warm phase started circa 1995 (Enfield et al. 2001) and continued through the end of the study period in 2007. The relationship between TC frequencies and decadal-scale SSTs has been well documented (Knight et al. 2005; Kerr 2000; Landsea et al. 1999; Shapiro and Goldenberg 1998; Rao and Macarthur 1994).
AMO was named by Kerr (2000), and it is an index of North Atlantic sea surface temperatures between 0° and 60°N latitude and 7.5° and 75°W longitude. The AMO phase affects weather patterns, such as rainfall and river flow, over the continental United States (Enfield et al. 2001) and is linked to Atlantic hurricane activity (Landsea et al. 1999; Goldenberg et al. 2001; Gray 2007). The AMO dataset was obtained from the NOAA Earth System Research Laboratory’s Physical Sciences Division laboratory. Note that AMO data were averaged by the 12-month (January–December) anomaly values. These values are plotted as a time series in Fig. 3.
3. Results and discussion
The study period showed an average of 313 (of a total of 717) stations were affected by TC rainfall per year (Fig. 4a). Note that a station can be counted more than once in the same year, if more than one storm affects the site. Also, the number of stations is constant over time over the study area at 717. The year 1990 had the lowest number of stations affected at 52 and 1985 had the highest at 936, demonstrating the double—or even triple—counting (or more) of some stations in a single year. The number of affected stations showed a slight positive trend, but it was not statistically significant based on the Kendall tau-b correlation test. Instead, it shows a multidecadal oscillation, with affected stations above average from the 1960s to the mid-1970s and after the mid-1990s, and affected stations below average from the late 1970s to the early 1990s. The number of affected stations showed a high correlation with the AMO phase, significant at the 99% confidence level using Student’s t test. The 5-yr moving average shows two periods with a lower number of stations affected by TC rainfall: 1971–82 and 1987–94, and a positive trend after 1995. The number of stations affected by TC rainfall per year is directly related to the number of storms. However, averaging the number of stations affected by the number of TC per year shows a different result (Fig. 4b). This time series serves as an index of the annual average spatial extent of the storms that occurred each year. The greater the value, one can assume that the storms of that year affected larger areas than if this value was smaller. However, this is not an index for the number of storms or the total area affected each year, but rather the average area affected per storm event. The average number of affected stations per storm shows less interannual variability when compared with the total number of affected stations per year.
Total rainfall (TC and non-TC rainfall) accumulated for all 717 stations from June to November in the study period is shown in Fig. 5a. Figure 5b represents the anomalies from average, in units of standard deviations (std devs) from the mean. Overall, total rainfall exhibits great interannual variation (2.5 × standard deviations). The time series presents a drier period in the first decade (1960–70) that is likely related to the positive phase of AMO (see Fig. 3), as described by Enfield et al. (2001). Rainfall after 1970 was characterized by high interannual variability, and from 1992–96, there was a sequence of 5 yr with values higher than average followed by 5 yr (1997–2001) with values lower than average. Total rainfall has a slight overall positive trend, and the Kendall tau-b test indicates this trend is statistically significant (α ≤ 0.05) and similar to Karl and Knight (1998) and as described in the Intergovernmental Panel on Climate Change (IPCC) special report on climate change (Houghton et al. 1996, chapter 8).
Figure 5c is the TC-accumulated total rainfall by year, summed for all USHCN stations in the eastern United States. There is some bias introduced in this figure because of geographically varying station densities. The TC rainfall component accumulated each year is influenced by the number of storms (based on the 500-km threshold), the duration of each storm, the station density most affected, and most likely the storm’s forward velocity. Results are nevertheless intriguing. The time series shows large interannual variability of TC rainfall over land in the eastern United States (Figs. 5c,d), perhaps with some relationship to known patterns of AMO. During the AMO negative phase, from 1970 to the mid-1990s, annual TC rainfall sums were predominantly below average and then mostly above average after 1995 (see Fig. 3). The year 1985 had the largest total (3.6 standard deviations from the mean) and 1978 had the smallest. The moving average showed a positive trend after the mid-1990s, and the overall time series shows a slight positive trend; however, it was not statistically significant. Subsequent analysis using kriging largely eliminates the varying station density problem.
The TC rainfall contribution to total rainfall (total found in Fig. 5a) presents a high yearly variation, from 2.5% in 1978 to 16% in 1985 (Fig. 5e). The time series shows a negative trend during the first half of the study period (1960–84) and a positive trend during the second half (1985–2007). Overall, the TC rainfall contribution has a slight positive trend; however, it is not statistically significant.
Figure 6 displays a time series of the annual TC rainfall volume for the eastern United States. These annual values are derived by interpolating the spatial patterns of annual TC rainfall through kriging. The time series shows an annual average of 107 km3 of water volume. Although the data show an increasing TC rainfall volume rate of 1.5 km3 yr−1, especially evident after 1995, the Kendall test for trend indicates that it is not statistically significant. The TC rainfall consists of interdecadal and interannual variations, also found by Ren et al. (2007). The volume distribution shows 16 yr with values above average and 32 yr with values below average, indicating a right skew to the distribution (Fig. 6b). However, 65% of those years with positive values occurred after 1984, indicating that TCs are producing more volume of rainfall in the latter portion of the time series. The maximum of 405.8 km3 (3.6 standard deviations in Fig. 6b) occurred in 1985, driven by Hurricanes Bob, Claudette, Danny, Elena, Gloria, Henri, Juan, and Kate, and Tropical Storms Henri and Isabel. The second highest value occurred in 2004 with 313.7 km3 (2.6 × standard deviations on Fig. 6b), driven by Hurricanes Cindy, Dennis, Emily, Katrina, Ophelia, Rita, and Wilma, and Tropical Storms Arlene and Tammy. The lowest TC volume occurred in 1978 with 8.9 km3 with only Tropical Storms Amelia, Debra, and Hope making any contributions. The block of years from 1973 to 1984, with the exception of 1979, was a protracted period with anomalously low input of TC rainfall in the eastern United States.
Comparing TC area (Figs. 6c,d) with TC rainfall volume (Figs. 6a,b) shows a strong correlation (Kendall tau-b correlation = 0.674, p < 0.001). However, there are some interesting exceptions in the data, whereby 1985 stands out as an exceptional volume year though the area affected is not as impressive. Storms that year tracked close to the coast (e.g., Hurricane Juan that persisted along Louisiana’s coast for four days) and those that penetrated inland were quickly downgraded to tropical depression status (Hurricanes Danny and Elena). In these latter cases, the heavy coastal rains are tropical, but rains farther inland induced by storm remnants were not included as tropical. In some years TCs traveled greater distances, thereby covering larger areas, increasing the potential for rainfall. In other years even with many storms covering a large area, the total volume of rainfall was relatively low. The year 1979 is an example when TCs covered a large area with a corresponding small rainfall volume, in part because of the rapid forward movement of the storms over the eastern United States, for example, Hurricanes David and Frederic. After 1995, TCs have become more numerous, producing larger volumes and affecting larger areas.
The Levene (1960) test for equality of variances and the t test were used to compare ENSO warm and cold phases with TC variables (Table 1). Levene’s test is used to test if k samples have equal variances, with no requirement of a normal distribution. Equal variances across samples are called homogeneity of variance, and Levene’s test can be used to verify that assumption. The test for equality of variances show that only the percentage of TC rainfall to total rainfall (Percentage_tc) was statistically significant (at the 95% confidence level); hence, the null hypothesis of equal variance between La Niña and El Niño is rejected. However, the t- test for Percentage_tc under either positive or negative ENSO conditions is insignificant. The variable Number of Storms shows a value significant at α = 0.095, suggesting somewhat different storm frequencies during La Niña events when compared to El Niño. Gray (1984) pointed out that differences in TC frequency between ENSO phases is related to the storm track, whereby during non–El Niño years, TCs cross the Caribbean more frequently. However, despite this result, there is no ENSO effect with the amount of TC rainfall over the eastern United States.
Wendland (1977) noted that the frequency and intensity of TCs are associated with the magnitude and distribution of SSTs. Furthermore, Nyberg et al. (2007) found that the increase in hurricane activity since 1995 could be considered a return to normal TC activity when compared with other periods.
The AMO phases were tested against the suite of variables. Levene’s test for equality of variances shows that all variables are insignificant; hence, the variances in TC rainfall do not differ significantly between warm and cold SST phases. Using a t test, Month_tot, TC_tot, VOLUME, and Percentage_tc are not significant at α ≤ 0.05. However, VOLUME was significant at α = 0.054. Nontropical, AREA, and Number of Storms are found to have significantly different mean values between AMO phases at the 95% confident interval. There are significant differences in the values related to the AMO phase that cannot be attributed to variability alone. The North Atlantic SST works as fuel to power TC by providing moist enthalpy and instability (Elsner et al. 2008). This suggests that the AMO positive phase could increase the seasonal number of storms and the large area covered by those storms, and affect nontropical rainfall. The Kendall tau-b test was used to determine the possible correlations (Table 3). AMO is shown to have a high correlation with almost all variables. Month_tot showed a negative correlation with AMO; however, it was not statistically significant. TC_tot and Percentage_tc showed a positive correlation with AMO, significant at α levels of 0.062 and 0.067, respectively. The variables related to TC rainfall: VOLUME, Number of Storms, and AREA have positive correlations with AMO at α = <.01.
In conclusion, ENSO and AMO phases play different roles in relation to their influence on TC rainfall in the United States. ENSO has a strong signal in relation to the number of storms; however, there is an insignificant relationship to the amount of TC rainfall, explained by El Niño increasing the upper-atmosphere wind shear over the Caribbean Sea and tropical Atlantic (Gray 1984). On other hand, AMO has a statistically significant correlation with all variables related to TC rainfall in the eastern United States. Warmer SST clearly leads to increases in TC frequency, resulting in increased inland TC rainfall.
This study represents a first effort to examine TC rainfall in the eastern United States from the perspective of documenting accumulated annual water volumes and areas of the precipitation. These TC rainfall metrics were then compared to ENSO and AMO. Time series of annual TC water volumes show an annual average of 107 km3. The single year with the maximum volume of TC-induced precipitation was in 1985 with 405.8 km3 driven by Hurricanes Bob, Claudette, Danny, Elena, Gloria, Henri, Juan, and Kate, and Tropical Storms Henri and Isabel. The second highest value occurred in 2004 with 313.7 km3. The lowest TC volume occurred in 1978 with 8.9 km3 with only Tropical Storms Amelia, Debra, and Hope making any contributions. The years 1973–84, with the exception of 1979, was a protracted period with anomalously low input of TC rainfall in the eastern United States.
In the Atlantic basin in total, TC frequency is related to ENSO phases, where the warm phase (El Niño) has fewer storms and the cold phase (La Niña) has more storms. However, ENSO phases did not show any statistical correlation with TC frequency in the eastern United States. AMO showed a significant correlation with volume produced by TC rainfall and the number of storms. The AMO phases have a negative correlation with nontropical rainfall; however, monthly rainfall is not statistically significant.
When comparing TC area in the eastern United States with TC rainfall volume, a strong correlation was found. However, there are some interesting exceptions in the data, whereby 1985 stands out as an exceptional volume year though the area affected is not as impressive. In that year, storms that tracked close to the coast (e.g., Hurricane Juan persisted along Louisiana’s coast for four days) and those that penetrated inland were quickly downgraded to tropical depression status (Hurricanes Danny and Elena) and were then no longer included in the analysis. In some years TCs traveled greater distances, thereby covering larger areas, increasing rainfall potential. In other years, even with many storms covering a large area, the total volume of rainfall was relatively low. The year 1979 is an example when TCs covered a large area with a corresponding small rainfall volume, in part because of to the rapid forward movement of the storms over the eastern United States, for example, Hurricanes David and Frederic. After 1995, TCs have become more numerous, producing larger volumes and affecting larger areas.
This research was partial funded by NOAA Grants NA080AR4320886 and EA133E-07-CN-0084. We also acknowledge the anonymous reviewers, whose comments improved this manuscript.
Corresponding author address: Ricardo Nogueira, Louisiana State University, E348 Howe-Russell, Baton Rouge, LA 70803. Email: firstname.lastname@example.org