The objective of this paper is to characterize the precipitation amounts originating from tropical cyclones (TCs) in the southeastern United States during the tropical storm season from June to November. Using 12 years of precipitation data from the Tropical Rainfall Measurement Mission (TRMM), the authors estimate the TC contribution on the seasonal, interannual, and monthly precipitation budget using TC information derived from the International Best Track Archive for Climate Stewardship (IBTrACS). Results derived from the TRMM Multisatellite Precipitation Analysis (TMPA) 3B42 showed that TCs accounted for about 7% of the seasonal precipitation total from 1998 to 2009. Rainfall attributable to TCs was found to contribute as much as 8%–12% for inland areas located between 150 and 300 km from the coast and up to 15%–20% for coastal areas from Louisiana to the Florida Panhandle, southern Florida, and coastal Carolinas. The interannual contribution varied from 1.3% to 13.8% for the period 1998–2009 and depended on the TC seasonal activity, TC intensity, and TC paths as they traveled inland. For TCs making landfall, the rainfall contribution could be locally above 40% and, on a monthly basis, TCs contributed as much as 20% of September rainfall. The probability density functions of rainfall attributable to tropical cyclones showed that the percentage of rainfall associated with TC over land increased with increasing rain intensity and represent about 20% of heavy rainfall (>20 mm h−1), while TCs account for less than 5% of all seasonal precipitation events.
The precipitation climatology of the southeastern United States spans a very broad spectrum of precipitation regimes. The warm season is characterized by isolated thunderstorms, mesoscale convective systems, and tropical cyclones (TCs), and the winter season is characterized by widespread frontal rain, ice, and snowfall. Each of these types of precipitation systems impacts regional hydrology in very different ways, and are associated with a large variety of natural hazards. From the early 1950s to 2006, landfalling tropical cyclones and their remnants have been responsible for damages of over $370 billion (all amounts are in U.S. dollars; adjusted 2006) and for more than 3750 deaths in the United States with over 1500 deaths for Hurricane Katrina only (Blake et al. 2007). Moreover, Rappaport (2000) showed that freshwater flooding was responsible for more than one-half of the U.S. hurricane-related deaths from 1970 to 1999. In addition to the important human death toll caused by TCs, the economic impact is also important: affected coastal counties experience after an average hurricane a decrease of at least 0.45 percentage points in the county annual growth rate (Strobl 2011). The development of consistent and global hurricane record databases (Kossin et al. 2007; Knapp et al. 2010) is an important effort to better characterize the trends in tropical cyclone activity, which was the object of debate within the scientific community (Webster et al. 2005; Emanuel 2005a,b; Landsea 2005; Pielke 2005). Results have confirmed an upward trend for the Atlantic basin while no similar trends were found for the other TC basins (Kossin et al. 2007). Furthermore, model projections suggest that a warming world could result in more destructive tropical cyclones with conflictive trends for tropical cyclone frequency and different long-term sensitivity between all basins (Knutson et al. 2010). Regardless of scientific challenges regarding future projections, studies showed that the increase in national monetary losses (adjusted by inflation to 2006 dollars) related to hurricane catastrophes over the past century (Blake et al. 2007) were in large part due to an increase in population density of exposed areas, growing wealth, and insurance coverage in the Gulf and East Coast areas (Pielke et al. 2008; Schmidt et al. 2010; Changnon 2011).
The impact and contribution of tropical cyclones on the southeastern United States has been the focus of recent studies (Konrad et al. 2002; Knight and Davis 2007, 2009; Konrad and Perry 2010; Villarini and Smith 2010; Kunkel et al. 2011). These studies used rain gauge observations for which the geographical coverage is not spatially homogeneous. Therefore, one of the limitations of this approach is the use of extrapolation and smoothing techniques that can generate local artifacts when observations are scarce. Gridded precipitation products such as gridded fields from the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis or NCEP–Climate Prediction Center (NCEP–CPC) gridded rainfall estimates could partially alleviate this problem, but the grid resolution (2.5° × 2.5° for NCEP–NCAR and 1° × 1° for NCEP–CPC) and the temporal coverage (daily) are too coarse to allow the derivation of tropical cyclone contribution at a satisfying resolution (Atallah et al. 2007). However, because of their temporal coverage of over several decades, those datasets have been used successfully for longer-term climatological and continental-scale studies (Larson et al. 2005; Atallah et al. 2007). By contrast, studies based on satellite precipitation products or satellite-based composite precipitation products can provide rainfall estimates at a finer resolution both spatially and temporally. Lau et al. (2008) used data from the Global Precipitation Climatology Project (GPCP) at 2.5° × 2.5° (Xie et al. 2003) and the 3-hourly Tropical Rainfall Measurement Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) at 0.25° × 0.25° (Huffman et al. 2007) to investigate the relationship between tropical cyclones and extreme rainfall events. Jiang and collaborators used TRMM TMPA 3B42 and TRMM precipitation radar (PR) 2A25 to assess the global contribution of tropical cyclones for six tropical cyclone basins around the globe (Jiang and Zipser 2010; Jiang et al. 2011). However, the aforementioned studies were conducted for tropical cyclone basins as a whole and did not focus specifically on the climatological impacts of landfalling tropical cyclones and their contribution to the precipitation amount (Lau et al. 2008; Jiang and Zipser 2010; Jiang et al. 2011). Using TRMM TMPA 3B42, Shepherd et al. (2007) showed that tropical cyclones were responsible for an average of 12.8% of the precipitation amount for the coastal southeastern United States limited to selected minibasins and coastal counties from northern Maine to southwestern Texas.
The focus of this study is to be found somewhere in between large-scale studies (Lonfat et al. 2004; Larson et al. 2005; Atallah et al. 2007; Lau et al. 2008; Jiang and Zipser 2010; Jiang et al. 2011) and more localized coastal studies (Shepherd et al. 2007). In this work we are interested in the contribution of tropical cyclones to the overland precipitation amount for the southeastern United States (24°–40°N, 72°–104°W). This impact of tropical cyclones on a regional scale and on a year-to-year basis will be performed using all available satellite precipitation estimates from TRMM TMPA 3B42 from 1998 to 2009. This paper is organized as follows: In section 2 we present the precipitation data and the methodology used in this work. In section 3 we focus on precipitation associated with TCs for the domain of study and the contribution of TCs to the total precipitation budget. In section 4 we will discuss interannual and monthly variability in terms of precipitation amount. In section 5 we will look at the link between tropical cyclone characteristics and precipitation extremes. Finally the paper will wrap up by summarizing the major findings of this work.
2. Data and methodology
a. Precipitation dataset TRMM TMPA 3B42
In this study we focus on the southeastern United States, a geographical domain delineated by 24°–40°N, 72°–104°W. The precipitation data used in this study are taken from TRMM Multisatellite Precipitation Analysis precipitation product TRMM TMPA 3B42 version 6 (hereafter TRMM 3B42). Launched in 1997, the TRMM satellite was the first entirely devoted to the measurement of precipitation and provides several precipitation estimates for a band (40°S–40°N) centered on the equator. TRMM 3B42 is a combination of different remotely sensed microwave [TRMM Microwave Imager (TMI), Special Sensor Microwave Imager (SSM/I), Advanced Microwave Scanning Radiometer (AMSR), Advanced Microwave Sounding Unit (AMSU)] and calibrated IR estimates with rain gauge corrected monthly accumulation (Huffman et al. 2007). TRMM 3B42 provides 3-hourly/0.25° precipitation estimates for the domain (40°S–40°N), which allows for deriving seasonal, monthly, daily, and subdaily precipitation characteristics. Although TRMM 3B42 is available for 50°S–50°N starting in 2001, we limit our study to the upper limit of 40°N in order to maintain a better consistency between datasets. The main caveat of the TRMM 3B42 is that it incorporates different satellite products for which algorithms have been modified over the years. For instance, the incorporation of precipitation estimates from AMSU-B introduced in 2001 generated low biases due to modification in the AMSU-B algorithm in 2003 (Huffman et al. 2007). Precipitation estimates from 3B42 were greatly improved following a modification of the algorithm in 2007 (Huffman et al. 2007). In this work we cover the period 1998–2009 including 12 TC seasons [June–November (JJASON)].
b. The tropical cyclones characteristics and the IBTrACS database
Information on the Tropical Cyclones is found in the International Best Track Archive for Climate Stewardship (IBTrACS). The database (http://www.ncdc.noaa.gov/oa/ibtracs/) gathers historical records of tropical cyclone characteristics for all hurricane basins (Knapp et al. 2010). IBTrACS is similar to the commonly used Atlantic basin hurricane database (HURDAT). It includes additional track information provided by the other Regional Specialized Meteorological Centers (RSMCs) around the world for the other TC basins. The IBTrACS database provides the location of the center of the TC, the TC intensity, and other important characteristic parameters of the TC every 6 h (Knapp et al. 2010). For the Atlantic basin, the different types of tropical cyclones are defined based on the 1-min maximum sustained wind (MSW) according to the Saffir–Simpson hurricane scale (SSHS). In this work, tropical cyclone is a generic term that includes tropical depressions (TDs; 0 kt ≤ MSW ≤ 34 kt), tropical storms (TSs; 35 kt ≤ MSW ≤ 63 kt), and hurricanes (MSW ≥ 64 kt). In addition, storms continue to be tracked in IBTrACS after they become cold-core, and any precipitation from these extratropical systems is also included in the totals.
To assess the tropical cyclone contribution on the precipitation total, we identify all TRMM 3B42 precipitation events occurring within a 500-km radius from the TC center as rainfall attributable to TCs (Larson et al. 2005; Lau et al. 2008; Jiang and Zipser 2010; Schreck et al. 2011). Therefore, all precipitation events related to tropical cyclones within the domain of study (24°–40°N, 72°–104°W) correspond to TCs whose center crosses the geographical area 19°–45°N, 67°–109°W. The 500-km radius criterion is consistent with the extent of the TC primary wind circulation domain (i.e., 80–400-km radius from TC center) and with the extent of the curved TC cloud shield (i.e., 550–600-km radius) described elsewhere (Cry 1967; Englehart and Douglas 2001). Furthermore, sensitivity tests showed little change in TC rainfall with respect to distance from the TC center for distances greater than 500 km and up to 1000 km (not shown). One of the caveats of the 500-km radius is that precipitation arising with existing troughs/fronts might be included in the totals. Since TC center locations from IBTrACS are available every 6 h, we perform a simple linear interpolation to determine the position of the TC center every 3 h in order to match the temporal resolution (3 h) of the TRMM 3B42 dataset. Because storms are tracked independently, the rain associated with each tropical storm is accounted separately. The seasonal, annual, or monthly rain attributable to TCs is the sum of all the 3-h TRMM 3B42 rainfall pixels for which the distance to the TC center location at the same time is less than 500 km.
3. Contribution of tropical cyclones to the total precipitation budget
a. Tropical cyclone activity in the southeastern United States 1998–2009
For the period 1998–2009 (12 TC seasons), a total of 113 TCs crossed the area during the TC season (i.e., from June to November) with 60 TCs making landfall in the southeastern United States (Fig. 1a). Please note that the three TCs that occurred in May (1) and December (2) are not included in this study. From this total there were 47 hurricanes of categories 1–5 (CAT15 with MSW ≥ 64 kt) and 21 classified as major hurricanes of categories 3–5 (CAT35 with MSW ≥ 96 kt) (Fig. 1a). During those 12 years (1998–2009), the strongest year for TC activity was 2005, with 18 TCs crossing the area including seven hurricanes, which was followed by the weakest year for the 1998–2009 period (2006) with only five TCs (Fig. 1a). On a monthly basis, the TC frequency displays the characteristic bell-shaped curve in the North Atlantic (NA) hurricane basin (Jiang and Zipser 2010). The maximum TC activity (60%) is equally spread between the months of August and September (Fig. 1b). This distribution for the domain 24°–40°N, 72°–104°W contrasts slightly with the TC monthly distribution for the greater North Atlantic hurricane basin for which the TC frequency is about 40% higher in September than in August (Jiang and Zipser 2010).
Figure 2a displays the annual number of TC events for the 12-yr period derived from IBTrACS. The spatial distribution represents the density of possible rain events (3-h duration) attributable to TCs considering a 500-km spread centered over the TC track. The storm density was maximum over ocean in the Atlantic and the Gulf and decreased when moving inland in the northwest direction. Along the coasts a larger number of tropical cyclones spanned an area from Louisiana to North Carolina (Fig. 2a). Regardless of TC intensity, we observe a similar pattern for tropical depression and tropical storms (TD/TS: MSW ≤ 63 kt: Fig. 2b), minor hurricanes [categories 1 and 2 (CAT12): 64 kt ≤ MSW ≤ 95 kt: Fig. 2c], and major hurricanes (CAT35: MSW ≥ 96 kt: Fig. 2d). Tropical storms and depressions (TD/TS) were equally spread along the Gulf and Atlantic coasts and the Florida peninsula (Fig. 2b). For hurricanes we note that weaker hurricanes (CAT12) and associated precipitation were more important in the Atlantic (Fig. 2c) while major hurricanes (CAT35) are dominant over the Gulf (Fig. 2d). This pattern is consistent regardless of whether the period considered was 1950–2009 or 1980–2009 (not shown). The presence of a warm core eddy (WCE) and a warm loop current (LP) in the Gulf of Mexico have been found to have an intensifying effect on tropical cyclones (Bender and Ginis 2000; Bosart et al. 2000; Hong et al. 2000) that could explain the higher number of intense hurricanes. In the Atlantic, the interaction of TCs with the Gulf Stream tends to strengthen less intense storms (CAT12 or weaker) or early TC season storms whose tracks are parallel to the Gulf Stream (Bright et al. 2002).
b. Spatial distribution of tropical cyclone seasonal contribution
Figure 3a displays the seasonal (JJASON) precipitation for the southeastern United States. Over land, important rainfall (>4 mm day−1) was found for the coasts surrounding the Gulf of Mexico and for the Florida peninsula. Precipitation ranged from less than 1 mm day−1 on the western part of the domain to above 6 mm day−1 in southern Florida. Along the Atlantic coast, the average precipitation was between 3 and 4 mm day−1 (Fig. 3a). Precipitation over ocean was more intense than over land, ranging from 2 to above 7 mm day−1. Significant rainfall (3–4 mm day−1) was found over the Gulf of Mexico and more intense precipitation (>7 mm day−1) was observed offshore of the Atlantic coast corresponding to the location of the Gulf Stream. Recent studies have shown that during summertime [June–August (JJA)] deep convection develops along the Gulf Stream that creates a narrow band of cumulus precipitation following the Gulf Stream (Minobe et al. 2008; Kuwano-Yoshida et al. 2010). When rainfall associated with TCs was removed from the total, we observed a similar pattern with a local maximum along the southeastern coast over land (Fig. 3b). The most significant differences in rainfall were observed over the ocean with maximum differences (>1 mm day−1) over the Gulf of Mexico, over the Caribbean, and off the eastern coast of the Florida peninsula (Fig. 3b). Figure 3c displays the seasonal (JJASON) variance of precipitation for the period 1998–2009 for all rain events. Higher variability was found over ocean with maximum variance along the Atlantic coast, over the Caribbean, and over the Gulf of Mexico. Over land, the maximum variance was observed along the Louisiana–Texas coast. Once TC rainfall was removed (Fig. 3d), the precipitation variance was reduced throughout most of the domain. Precipitation variance remains high over land along the Louisiana–Texas coast, and locally around 27°N between the Florida coast and the Bahamas. Moderate seasonal variability was observed in the portion of the Gulf of Mexico corresponding to the loop current. The precipitation variance remained high along the Gulf Stream above 32°N, thus illustrating a strong year-to-year variability of the rainfall activity along the Gulf Stream.
Figure 4a displays the seasonal tropical cyclone rainfall—that is, the accumulation of rain events derived from TRMM 3B42 occurring within a 500-km radius centered over the TC track. The TC rain accumulation (Fig. 4a) corresponds to the difference between the rainfall accumulation from all precipitation events (Fig. 3a) and rainfall accumulation unrelated to TC activity (Fig. 3b). Higher accumulation was found over ocean, over the Gulf of Mexico, and over the Atlantic below 36°N with a maximum TC rain amount greater than 200 mm yr−1. TC precipitation over ocean was generally higher than 120 mm yr−1 over the Gulf of Mexico (81°–93°W) and the Atlantic (25°–34°N). Rodgers et al. (2001) indicated comparable TC precipitation using rain rates derived from the SSM/I. Maximum TC rainfall over land is significantly lower (about 30% lower) than over ocean. The southern part of the Florida peninsula and the Florida Keys (below 28°N), the Florida Panhandle, and a small portion of the North Carolina coast (34.25°–35.25°N) experience a higher TC rainfall amount, greater than 120 mm yr−1. The high accumulation over the North Carolina coast was mainly due to two major hurricanes that made landfall in 1999 (Dennis and Floyd). Inland areas located within a distance of 150–300 km from the coast of the Gulf of Mexico (east of 96°W) and along the Atlantic coast (south of 38°N) experienced a TC rainfall amount greater than 60 mm yr−1. Figure 4b displays the number of TC-related events (i.e., the number of nonzero 3-h precipitation pixels derived from TRMM 3B42 and falling within 500 km of the TC center). The patterns in Fig. 4b were comparable to Fig. 2a, which showed the total number of pixels (raining or not) that fell within 500 km of a TC. Overall, the fraction of rainy pixels (i.e., with TC rainfall greater than zero) over the total number of possible rain events (Fig. 2a) was between 15% and 35%. Over land this ratio was lower and between 15% and 25% while over ocean this ratio was higher than 30% over the Gulf of Mexico and in the Atlantic south of 36°N (not shown). The differences in pattern observed over land and in particular over the Florida peninsula illustrate the changes in TC structure and weakening during landfall.
Figure 4c displays the ratio of rain attributable to TCs over the total JJASON precipitation for the period 1998–2009. As expected TC rainfall contribution was the highest along the coasts (Gulf of Mexico, Atlantic coast) and over southern Florida. Over land the percentage of rainfall contributed by TCs was about 8–12% for areas located within a distance of 150–300 km from the Gulf and Atlantic coasts. This contribution decreased northwestward down to 2% and less as we progress inland for the upper part of the domain corresponding to western Texas and the Midwestern states. Directly along the coasts, the TC contribution was higher and between 15% and 20% over an area that corresponds to the coastal plain of southern/mid-Atlantic states (North Carolina/South Carolina) and over the coastlines along the Gulf of Mexico. This value is consistent with Shepherd et al. (2007), who found a TC rainfall contribution between 8% and 17% (12.8% average) along the coastal southeastern United States using TRMM 3B42. Similarly, using gridded daily precipitation datasets, Larson et al. (2005) found a similar contribution of 15–20% along the U.S. Gulf Coast for the period 1950–98. From a broader perspective, results are comparable with other studies using similar precipitation datasets. Jiang and Zipser (2010) found that the overall contribution for the entire NA hurricane basin was about 8–9% for the period 1998–2006, while we found about 7% for the entire domain (24°–40°N, 72°–104°W). In addition to the fact that we are studying a smaller geographical domain, years 2007–09 were characterized by a relatively low TC activity (Fig. 1a). However, this ratio (7%) is about twofold higher than that of Rodgers et al. (2001), who reported a TC rainfall contribution of about 4% for the NA basin. Differences in sensors (SSM/I vs TRMM) and time periods (1987–98 vs 1998–2009) could explain the differences observed. In addition, surface observation stations have shown a comparable contribution of TCs with as much as 15% of the rainfall amount for the hurricane season from 1980 to 2004 (Knight and Davis 2007). The pattern for the rainfall contribution of TCs derived from TRMM 3B42 (Fig. 4c) agrees surprisingly well with surface stations [see Fig. 4 of Knight and Davis (2007)] regardless of the time periods considered (1998–2009 vs 1980–2004). The good agreement between this work and previous studies (Knight and Davis 2007) allows us to be fairly confident regarding the robustness of the seasonal results despite the limited 12-yr time period considered here. Because hydrological applications often consider an annual basis for precipitation totals, Fig. 4d displays the annual contribution of TCs (including winter and spring precipitation). For the annual TC contribution (Fig. 4d), we observe a comparable spatial partition as for the seasonal TC contribution (Fig. 4c) with a maximum contribution (9–11%) located along the coasts. Next we examine the interannual and monthly variability in the TC rainfall contribution.
4. Tropical cyclones rainfall: Interannual variability and monthly trends
a. Interannual variability and yearly contribution of tropical cyclones
Figure 5 displays the interannual TC rainfall percentage from 1998–2009, which is very different from the 12-yr mean contribution (Fig. 4c) as it includes the influence of isolated TCs. For major hurricanes making landfall, the amount of TC rainfall was locally greater than 300 mm, such as over coastal Alabama and the Florida Panhandle in 1998 (George), coastal Carolinas in 1999 (Dennis, Floyd), the Florida peninsula in 2004 (Frances, Jeanne), or coastal Louisiana and southern Florida in 2005 (Cindy, Katrina, Rita). The TC contribution associated with the abovementioned events (1998, 1999, 2005) was found locally greater than 40% for impacted coastal areas. In addition, particular situations were able to produce locally important TC contribution inland. In September 2004, the passage of two tropical cyclones (Frances and Ivan) dropped a combined amount of rain greater than 200 mm, which contributed for more than 30% of the seasonal rainfall in northern Georgia and western North Carolina and caused landslides and debris flows in the Appalachians (Wooten et al. 2008). Figure 6 summarizes the year-to-year overland rainfall accumulation for all rain events, non-TC events, and TCs only (Fig. 6a), along with the tropical cyclone contribution to the seasonal overland precipitation and percentage of the domain impacted at least by one TC (Fig. 6b). For the period 1998–2009, the total overland rainfall accumulation showed a 58% difference from a minimum of 432.2 mm (0.25° × 0.25°; 2005) to a maximum of 681.8 mm (0.25° × 0.25°; 2004). Similarly, rainfall accumulation from non-TC events displayed a 65% variation between low (2005) and high (2004) years (Fig. 6a). Interannual TC rainfall presented a ninefold difference between 2009 (7.0 mm; 0.25° × 0.25°) and 2004 (66.8 mm; 0.25° × 0.25°) (Fig. 6a). A better sense of the local TC contribution can be understood by normalizing TC rainfall by the area impacted (AI) by TCs. The area impacted was defined as the overland domain that displayed rain attributable to TCs for a given year or, in other words, the annual geographical extent of the TRMM 3B42 rainy pixels associated with TC rainfall. The year-to-year variation was about fivefold between 2007 (26.8 mm per AI) and 2004 (125.5 mm per AI) (Fig. 6a). The years 2004 and 2005 have the highest TC-related rainfall, while 2000, 2006, 2007, and 2009 have the lowest precipitation amount arising from TCs. Years with major hurricanes [Georges (1998), Floyd (1999), Frances (2004), Ivan (2004), Katrina (2005), Rita (2005), Gustav (2008)] displayed a normalized accumulation above 80 mm per AI (1998, 2005, 2008) or even greater than 110 mm per AI (1999, 2004) due to the localized impact of major hurricanes making landfall [Floyd (1999)] or a particular configuration with back-to-back hurricanes crossing over the same area [Frances (2004), Ivan (2004)].
Overall for 1998–2009, the percentage of overland rainfall related to TCs was found to be 7% with the yearly percentage varying within a tenfold range from 1.3% for 2009 to 13.8% for 2005, the record year for the Atlantic hurricane basin (Fig. 6b). For 2009, the low percentage is explained by the fact that only two TCs of the six in the domain made landfall, giving the lowest precipitation amount originating from TCs over land (7 mm; 0.25° × 0.25°). In 2005, 7 TCs of 18 made landfall and, despite a higher percentage contribution (Fig. 6b), the year 2004 (7 of 9) presents a higher accumulated total (Fig. 6a). Locally, the TC contribution is highly dependent on the extent of the area impacted. Figure 6b shows that the area impacted by TCs was generally above 50% of the domain of study with the exception of 1999, 2000, 2006, and 2009. Typically for years with major landfalling hurricanes (1998, 1999, 2004, 2005), rainfall associated with TCs accounts for more than 15% of the seasonal precipitation over the area impacted (Fig. 6b). We note that even for years with low overall TC contribution (<4%: 2000, 2006, 2007, 2009), TC rainfall could contribute as much as 25% locally as seen for the years 2000 and 2007 (Fig. 5). There has been speculation as to whether a relatively low cyclonic activity could increase drought probability, and although the length of record does not allow us to confirm this hypothesis, it worth noticing that the record-setting drought year 2007 corresponds to a succession of two years (2006 and 2007) of low TC activity (Figs. 5 and 6). However, because of the short period of observation no conclusions can be drawn.
Previous studies have reported the influence of the La Niña and El Niño/–Southern Oscillation (ENSO) events on the TC formation. During La Niña (cold) and neutral years, there is evidence of higher TC activity in the NA hurricane basin in contrast with El Niño years (warm) that correspond to a lesser TC activity and a lower probability of strikes from tropical cyclones (Gray 1984a,b; Goldenberg and Shapiro 1996). Figure 7 displays the number of storms (Fig. 7a) and the TC seasonal overland contribution (Fig. 7b) as a function of ENSO activity with error bars representing one standard deviation. The 12 seasons of TC activity corresponded to four El Niño years (2002, 2004, 2006, 2009), three La Niña years (1998, 1999, 2007), and five neutral years (2000, 2001, 2003, 2005, 2008) (http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml). During El Niño years, the average number of TCs is 7.7 TCs per year, whereas it is slightly higher during La Niña years (8.3 TCs yr−1: +8%) and noticeably higher for neutral years (11.4 TCs yr−1: +47%) (Fig. 7a), which is consistent with long-term studies (Gray 1984a). Globally the difference between El Niño and La Niña or neutral years is +33%. For hurricanes (i.e., storms category 1 and higher), the differences between warm years and cold years were +45% (2.75 CAT15 yr−1 vs 4.0 CAT15 yr−1) and up to +75% (4.8 yr−1) for neutral years (warm versus cold/neutral is +64%) (Fig. 7a). Similarly, the TC seasonal contribution was lower for warm years (+5.6%) than for neutral years (+7.6%), cold/neutral years (+7.7%), and cold years (+7.9%) (Fig. 7b). On average, the TC contribution was 35% (40%) higher during cold (neutral) years than during warm years. Those results are comparable with longer-term studies for the NA hurricane basin (Gray 1984a,b; Goldenberg and Shapiro 1996), although more TRMM observation years would be necessary to derive more robust statistics and longer-term trends.
b. Monthly rainfall contribution of tropical cyclones
Figure 8 displays the tropical cyclone monthly rainfall and contribution. Important variability was observed locally for the monthly precipitation contribution due to the limited period of observation. Monthly spatial patterns of TC rainfall were particularly dependent on isolated TCs. For months that displayed limited TC activity (e.g., June, July, November, and, to a lesser extent, October), TCs provided limited overall rainfall below 10 mm month−1. The locations with the highest TC rainfall corresponded to the passage of particular TCs over the 12-yr period (Fig. 8). For instance, June and October displayed local accumulation above 30 mm month−1 over southern Louisiana and the Florida peninsula, respectively, which corresponded roughly to areas where two TCs tracks crossed paths over the 12-yr period. However, even a limited number of TCs can contribute locally to a significant amount of the total rainfall. In November, the month with the lowest TC rainfall, the maximum of about 15 mm month−1 over southern Florida was associated with the passage of Hurricane Mitch (1998). This isolated event represented locally an important contribution of about 20%–30% of the monthly rainfall due to the relatively low non-TC precipitation in November. For August and September, months corresponding to the peak of TC activity, maximum TC rainfall was located along the coasts from Texas to the Carolinas. In August, maximum accumulation greater than 25 mm month−1 was found over areas of southern Florida and coastal North Carolina impacted by four or more TCs occurring over the 12-yr period. The corresponding TC contribution represented about 20% of the monthly rainfall in southern Florida and 25%–30% over coastal North Carolina. In September, a higher number of landfalling TCs (Fig. 1b) together with a more homogenous partition of TC tracks over the coastal domain made TC rainfall less sensitive to particular TC tracks (Fig. 8). TC rainfall above 50 mm month−1 spanned an area from east Texas to central Georgia and from northern South Carolina to North Carolina. For inland areas more than 300 km from the coasts, TC rainfall accounted locally up to 30 mm month−1. Overall, the TC rainfall contribution was above 20% for a large portion of the domain with local contribution above 40% over the Florida Panhandle, southern Alabama, southern Georgia, and coastal North Carolina. In southern Florida TC monthly rainfall accounted for about 20% because of important non-TC precipitation (>6 mm day−1) induced by sea-breeze rainfall (Prat and Nelson 2012, manuscript submitted to Atmos. Res.).
Figure 9 displays the average monthly rainfall over land for all events, non-TC events, and TCs (Fig. 9a), along with the monthly TC contribution, the extent of the area impacted by at least one tropical cyclone (i.e., monthly geographical extent of TRMM 3B42 rainy pixels associated with TC rain for the period 1998–2009), and the TC contribution over the domain experiencing TCs (Fig. 9b). For the entire overland domain, there was a 60% difference for total (non-TC) rainfall between June and November with 103.0 mm/0.25° × 0.25° (99.8 mm/0.25° × 0.25°) and 64 mm/0.25° × 0.25° (63 mm/0.25° × 0.25°), respectively (Fig. 9a). Because of the short period of observation, months with limited TC activity [June–July and October–November (JJ-ON)] displayed large variability for the TC rainfall for the entire overland domain, which decreased when reported to the area impacted (Fig. 9a). During the months corresponding to the peak of the TC activity [August–September (AS)], the average TC rainfall was found to be 7.5 and 18.8 mm in August and September, respectively. The higher TC rainfall observed in September is explained by the fact that more TCs made landfall in September (27 of 33) than in August (15 of 35: Fig. 1b). On average over the 12-yr period, TC rainfall accounted for 1.5%–3.5% (JJ-ON), 8.5% (August), and 20% (September) of the monthly precipitation (Fig. 9b). Again, a large year-to-year variability was observed for the TC contribution over the entire domain, for the TC contribution related to the AI, and for the extent of the area impacted by at least one TC in 12 years. For the entire period of observation, the geographical extent of TC contribution ranged from 20% of the domain in November to about 80% during the peak of TC activity (AS). We note that the TC contribution per AI, which is less dependent on year-to-year TC variability, displayed a higher and more homogenous average contribution than the overall domain contribution with 4%–7% (JJ-ON), 9.5% (August), and 22% (September). The limited number of TCs, especially during the months of low TC activity (JJ-ON), did not allow us to derive robust patterns for the monthly TC contribution. However, the monthly results displayed spatial patterns for the monthly TC contribution over the southeastern United States that are comparable to similar longer-term studies (Cry 1967; Larson et al. 2005; Knight and Davis 2007, 2009; Nogueira and Keim 2011). The local differences with the aforementioned studies can be explained by different observation periods and by the important interannual variations in TCs frequency and intensity (Fig. 1).
5. Link between tropical cyclones and heavy precipitation
Table 1 summarizes the seasonal contribution as a function of the precipitation type and TC intensity. While accounting for about 7% of the total rainfall, tropical cyclones represented only 4.64% of all precipitation events (Table 1). We note significant differences between tropical storms (MSW ≤ 63 kt) and hurricanes (MSW ≥ 64 kt) that represented 5.05% (3.69%) and 1.87% (0.95%) of rainfall total (precipitation events), respectively (Table 1). Although TCs often create intense rainfall over a short period of time and large areas, they are not necessarily linked with extreme precipitation. For instance, Konrad et al. (2002) showed that 67% of TCs that made landfall from 1950 to 1996 were not associated with extreme precipitation (2-day accumulation > 400 mm) for geographical areas ranging from 2500 to 500 000 km2. A look at the probability density function (PDF) derived from TRMM 3B42 (Fig. 10a) showed that PDFs for all precipitation events were very similar to non-TC rainfall events given the limited number of TCs that represented 4.64% of the rainfall events. Similarly, PDFs for all TCs were comparable to that of TD/TS that represented 80% of all TC-related precipitation events. While over-ocean PDFs (not shown) presented heavier tails with increasing TC intensity (i.e., average rain rate of 3.88 mm h−1 for CAT35 versus 3.12 mm h−1 for CAT12), overland PDFs associated with major hurricanes displayed narrower distribution and lighter tails (i.e., average rain rate of 3.12 mm h−1 for CAT35 versus 3.65 mm h−1 for CAT12). Another illustration is given by the cumulative density function (CDF) for TRMM 3B42 (Fig. 10b). Rainfall events above 10 mm h−1 represented about 1% of all the precipitation events and 11% of the total rainfall for TRMM 3B42 (Fig. 10b). Similar trends were observed for TC rainfall with 2.5% of TC rain above 10 mm h−1, corresponding to 17% of the total rainfall (Fig. 10b). For low to moderate rain intensity (R < 10 mm h−1), the ratio of rainfall attributable to TCs gradually increased (Fig. 10b). A sharp increase was observed for higher rain intensity (>20 mm h−1) with a TC rain ratio around 20% for extreme precipitation >50 mm h−1 (Fig. 10b). Present results are comparable with other studies that investigated the link between rainfall associated with TCs and extreme precipitation. Kunkel et al. (2011) found that TCs account for about 6% of extreme events (1 in 5 yr recurrence event) for the geographical domain corresponding to the eastern United States, which is consistent with the present study (Fig. 10b). The TC contribution to intense rainfall could be 30% or more of all extreme precipitation for coastal areas corresponding to the present domain (Kunkel et al. 2011) and up to 90% over the coastal region of the Carolinas (Konrad and Perry 2010). Similarly, TCs were responsible for 20%–25% of precipitation days above 2 in. day−1 (50.8 mm day−1; Knight and Davis 2009) and between 33% and 66% of precipitation days above 4 in. day−1 (101.6 mm day−1; Barlow 2011) along the East Coast. Globally, we found that the geographical distribution of precipitation extremes matched the storm track densities regardless of the TC intensity (Fig. 2a). Although TRMM 3B42 performs relatively well for 24-h rainfall distribution as a result of a rain gauge adjustment when compared to similar satellite datasets products, moderate and heavy rainfall (≥25 mm h−1) tend to be considerably underestimated (Yu et al. 2009) and consistent with the negative biases observed for instantaneous rain rate in the case of tropical cyclones (Prat and Barros 2010).
In addition to the type of precipitation events, important differences exist over land and over ocean. For the area of study (24°–40°N, 72°–104°W), the ratio of overland/over-ocean precipitation events decreased with increasing TC intensity and was relatively consistent between satellite products (not shown). While land (ocean) represented 65% (35%) of the domain, the overland (over-ocean) precipitation events represented 57% (43%) regardless of the precipitation type, and 58% (42%) for non-TC rainfall. TC-related overland events decreased to 38% when compared to occurrence over ocean (62%). This trend was emphasized when we considered the category of the TC with a ratio 45/55 for TD/TS, 26/74 for CAT12, and 19/81 for major hurricanes (CAT35). Figure 11 displays the ratio of precipitation type for four rainfall classes. For each figure (Fig. 11a) the impacts of the geographical distribution are reported using vertical bars indicating ratio over land (upper limit) and over ocean (lower limit) while the median delineation indicates results for the entire domain. As seen previously, the proportion of TC rain increased with increasing rain intensity. Over land, the proportion of TC rain was 3% for low rain rates (0–1 mm h−1), 6% for moderate rain rates (1–10 mm h−1), 9% for high rain rates (10–20 mm h−1), and went up to 12% for rain rates above 20 mm h−1 (Fig. 11a). Over ocean, this ratio was more than twofold for each rain rate categories: 6% (0–1 mm h−1), 13% (1–10 mm h−1), 25% (10–20 mm h−1), and 31% (>20 mm h−1) (Fig. 11a). For TC rain only, the ratio of hurricanes (CAT12 + CAT35) over tropical storms/depressions (TD/TS) increased as a function of the rain intensity with hurricanes representing about 50% of high rain rate (>20 mm h−1) TC rain events (Fig. 11b).
6. Summary and conclusions
In this work we investigated the contribution of tropical cyclones on the seasonal precipitation totals for the southeastern United States from 1998 to 2009. For the entire period of study, we found that the tropical cyclones contributed on average 7% of the seasonal (JJASON) precipitation totals for the overland domain. This value is comparable to the 8%–9% reported for the entire Atlantic basin without distinguishing between over ocean and over land (Jiang and Zipser 2010). Depending on TC activity and characteristics, the interannual contribution varied between 1.3% (2009) and 13.8% (2005) for the period of study (Fig. 6). Tropical cyclone contribution was the highest near the coast and represents about 15%–20% of seasonal precipitation (Fig. 4), which is comparable with other studies using either satellite observations or rain gauge point measurements (Larson et al. 2005; Knight and Davis 2007; Shepherd et al. 2007). However, we would like to emphasize that caution is needed in generalizing our results for a longer-period climate record. This is particularly relevant considering the fact that the period 1998–2009 coincides globally with the intense Atlantic hurricane seasons of the decade 1995–2005 (Goldenberg et al. 2001; Smith et al. 2007) and the ongoing debate regarding the influence of different climate modes including ENSO (Gray 1984a,b; Goldenberg and Shapiro 1996; Smith et al. 2007), the Atlantic multidecadal oscillation (Goldenberg et al. 2001), and the North Atlantic Oscillation (Xie et al. 2005), along with the possible impact of greenhouse gases.
On a year-to-year basis, the amount of rainfall attributed to TCs was found greater than 300 mm with TC contribution locally above 40% for years characterized by major hurricanes such as Floyd (1999) and Katrina (2005) (Fig. 5). The TC interannual contribution decreased rapidly when moving inland as the TC loses strength when making landfall. The TC seasonal contribution decreased to 8%–12% for a domain located within 150–300 km from the coasts and down to 2% or less beyond that distance as few TCs traveled that far inland (Fig. 4). Particular situations can lead to substantial contributions to seasonal precipitation away from coastal areas typically prone to tropical cyclone activity. A seasonal contribution above 30% was observed in 2004 when two TCs crossed back-to-back over the western Carolinas and dropped a total accumulation above 200 mm (Fig. 5). On a monthly basis, tropical cyclones contributed from less than 2% in November to up to 20% in September (Fig. 9), with a strong month-to-month variability observed for the spatial patterns of TC rainfall and TC contribution (Fig. 8).
Even though the period of study is only slightly longer than a decade and therefore long-term trends of the TC contribution to precipitation extremes cannot be assessed, rainfall attributable to TCs was found to account for an important part of heavy rainfall events. While the ratio of heavy rain events (>10 mm h−1) represented 1% of the total number and 11% of the accumulated rainfall regardless of precipitation type, this ratio was 2.5% (number) and 17% (accumulation) for TC rainfall only. The ratio of TC rainfall over all precipitation events over land increased with increasing rain intensity and represented about 20% of heavy rainfall (>20 mm h−1) while TCs accounted for less than 5% of all precipitation events (Fig. 10). Among the limitations of the work are the fact that TC rainfall totals might be overestimated when systems are no longer tropical (TC remnants) or merge with existing troughs and fronts. Furthermore, the use of a constant 500-km radius to determine TC rain instead of tracking the rain shield individually for each storm might results in a bias.
The work currently in progress proposes to extend this study by using quantitative precipitation estimates (QPEs) at 1-km and 5-min resolution issued from the National Mosaic QPE (NMQ/Q2) reanalysis project for the same domain and period of observation. Higher spatial and temporal resolution will allow us to assess TC contributions at a more local scale and quantify the differences in precipitation estimates between sensors. The work is also being extended to the characterization of other precipitation regimes that are often accompanied by intense precipitation such as mesoscale convective systems or localized thunderstorms for which duration and geographical extent often occur at a finer scale than the temporal and spatial resolution of satellites.
This work was supported by the NOAA/NCDC Climate Data Records and Science Stewardship Program through the Cooperative Institute for Climate and Satellites-North Carolina under the Agreement NA09NES4400006. The authors are grateful to Carl Schreck, Ken Kunkel, Scott Stevens, and three anonymous reviewers for valuable comments and suggestions.