Extreme floods in the Delaware River basin are examined through analyses of a sequence of record and near-record floods during September 2004, April 2005, and June 2006. The three flood episodes reflect three principal flood-generating mechanisms in the eastern United States: tropical cyclones (September 2004); late winter–early spring extratropical systems (April 2005); and warm-season convective systems (June 2006). Extreme flooding in the Delaware River basin is the product of heavy rainfall and runoff from high-gradient portions of the watershed. Orographic precipitation mechanisms play a central role in the extreme flood climatology of the Delaware River basin and, more generally, for the eastern United States. Extreme flooding for the 2004–06 events was produced in large measure from forested portions of the watershed. Analyses of flood frequency based on annual flood peak observations from U.S. Geological Survey (USGS) stream gauging stations with “long” records illustrate the striking heterogeneity of flood response over the region, the important role of landfalling tropical cyclones for the upper tail of flood peak distributions, and the prevalence of nonstationarities in flood peak records. Analyses show that changepoints are a more common source of nonstationarity than linear time trends. Regulation by dams and reservoirs plays an important role in determining changepoints, but the downstream effects of reservoirs on flood distributions are limited.
This study centers on analyses of three floods in the Delaware River basin (Fig. 1) that occurred in 2004–06 (Figs. 2 and 3). The three floods represent three major flood agents in the eastern United States (Miller 1990): landfalling tropical cyclones (September 2004; Hurricane Ivan); winter–spring extratropical systems (April 2005); and warm-season convective systems (June 2006). We combine analyses of the 2004–06 flood events with analyses of annual flood peak observations from U.S. Geological Survey (USGS) stream gauging stations to examine properties of the upper tail of flood peak distributions in the Delaware River basin [see Morrison and Smith (2002), Villarini and Smith (2010), and Villarini et al. (2009) for related analyses]. Important elements of regional flood peak distributions include the role of “mixtures” of differing flood-generating mechanisms, heterogeneities in rainfall forcing of extreme floods, heterogeneities of hydrologic response tied to both natural (terrain, forest cover, soil properties, drainage network structure) and anthropogenically altered properties (urbanization, agricultural practice, regulation through dams and reservoirs) of the land surface, time trends in climate and land use, and dependence of flood properties on basin scale.
The flood of record at many locations in the Delaware River basin is from Hurricane Diane in August 1955 (following days after Hurricane Connie). We focus particular attention in this study on the role of landfalling tropical cyclones, such as Hurricanes Ivan (2004) and Diane (1955), in controlling the upper tail of flood peak distributions in the Delaware River basin. Villarini and Smith (2010) show that there is pronounced spatial heterogeneity over the eastern United States in the frequency of flood peaks from tropical cyclones [see Colle (2003), Sturdevant-Rees et al. (2001), Atallah and Bosart (2003), and Hart and Evans (2001) for related analyses].
Orographic precipitation mechanisms play an important role in each of the three storms and more generally in the flood hydrology of the Delaware River basin. Distribution of heavy rainfall relative to complex terrain of the Appalachians is an important element of scale-dependent flood response [see, e.g., Barros and Kuligowski (1998), Nykanen (2008), Smith et al. (1996) and Pontrelli et al. (1999); related studies include Delrieu et al. (2005), Gaume et al. (2004), and Petersen et al. (1999)]. We use rainfall and discharge observations to examine the role of orographic precipitation mechanisms for the flood hydrology of the Delaware River basin.
The Delaware River basin is predominantly forest covered, with modest urban development occurring in the Bethlehem–Allentown, Pennsylvania, urban corridor in the Lehigh Valley. The role of forests in mitigating the effects of flooding is a topic that has received more than a century of scientific examination (Dunne 1978). Major changes to the Delaware River basin during the past century have been linked to the construction of dams and reservoirs. Like most major rivers in the United States, regulation by reservoirs is an important issue in the regional flood hydrology. Despite the extensive system of dams in the basin, the effects on flood hydrology are poorly understood (Williams and Wolman 1984; Batalla et al. 2004). We examine spatial heterogeneities in the flood response of the Delaware River basin, through both analyses of the 2004–06 flood events and through regional flood frequency analyses for the basin.
Wolock et al. (1993) have pointed to the role of climate change in altering the flood hydrology of the Delaware River basin, especially through changing seasonal patterns of snow accumulation and melt. Flood frequency in the Delaware River basin is also subject to changing climate, especially as it affects the principal flood-generating mechanisms. Changing frequency of tropical cyclones has received considerable attention, with little consensus on the potential for changes to landfalling tropical cyclones that affect the northeastern United States (e.g., Emanuel 2005; Landsea 2005; Holland and Webster 2007; Knutson et al. 2007, 2008; Vecchi and Knutson 2008). Similar issues hold for extratropical cyclones (e.g., Yin 2005; Leibensperger et al. 2008).
The contents of the sections are as follows. The study region, data, and analysis methodologies are introduced in section 2. Analyses of the three flood events are presented in section 3. A synthesis of regional flood frequency is presented in section 4. A summary and conclusions are presented in section 5.
2. Study region and methodology
The Delaware River basin above Trenton, New Jersey (Fig. 1), has a drainage area of 17 560 km2. Elevation ranges from more than 1200 m in the upper Delaware River basin in the northern portion of the basin to less than 50 m at the basin outlet at Trenton (Fig. 1). The northern and western boundaries of the watershed are mountainous, with the lower elevation and lower gradient portions of the watershed to the south and east. The eastern tributaries of the Delaware River in New Jersey between Port Jervis and Belvidere (Fig. 1) also include mountainous terrain.
The high-elevation regions of the Delaware River basin are largely forest covered. In the East Branch Delaware (E. Br.), for example (Fig. 1), 82% of the basin is in forest cover, 12% is grassland and brush, 3% is wetland and water, and less than 3% is urban. Agricultural land use is concentrated in the Lehigh Valley area, including the lower Lehigh River basin and the northeastern extension of the valley into New Jersey (Fig. 1).
The Delaware River basin includes a number of major dams and reservoirs (Fig. 1). Reservoirs in the upper Delaware River basin in New York [East Branch, West Branch (W. Br.), and Neversink subbasins; see Fig. 1] are important elements of the New York City water supply system. These dams are not systematically operated for flood control purposes. Dams in the Lackawaxen and Lehigh River basins provide both flood control and water supply functions. River basins in the United States of comparable size to the Delaware River basin are typically characterized by extensive regulation by dams and reservoirs.
“Unit values” discharge observations from U.S. Geological Survey (USGS) stream gauging stations at time intervals ranging from 15 min to 1 h provide the principal data resources for examining flood hydrology. Annual peak data from stations with records longer than 50 yr are used in section 4 for flood frequency analyses. Rainfall accumulation products at 15-min intervals were developed from the Hydro–Next Generation Weather Radar (Hydro-NEXRAD) system (Krajewski et al. 2007, 2008; Krajewski and Smith 2002). Rain gauge datasets include observations from the operational network used by the National Weather Service (NWS), daily observations from the Cooperative Observer Program (COOP), and 15–60-min observations from USGS and Integrated Flood Observing and Warning System (IFLOWS) networks. Lightning data from the National Lightning Detection Network (NLDN; Cummins et al. 1998) are used for examining convective intensity and evolution of the three storms (see Orville and Huffines 2001).
3. Hydrology and hydrometeorology of the “three floods”
The September 2004, April 2005, and June 2006 storms produced record and near-record flooding in the Delaware River basin (Figs. 2 and 3; Table 1). From the 97-yr record of the Delaware River at Trenton (Fig. 3), the 4 April 2004 peak of 6853 m3 s−1 (unit discharge of 0.39 m3 s−1 km−2) is the second largest flood peak. The 29 June 2006 peak of 6711 m3 s−1 (0.38 m3 s−1 km−2) is a close third, and the 19 September 2004 peak of 5692 m3 s−1 (0.32 m3 s−1 km−2) ranks fifth (Fig. 2). The largest flood peak for the Delaware River at Trenton of 9311 m3 s−1 (0.53 m3 s−1 km−2) on 20 August 1955 was produced by the back-to-back passage of Hurricanes Connie and Diane (U.S. Weather Bureau 1955).
The June 2006, April 2005, and September 2004 flood peaks, ranked first, second, and fourth, respectively, for the 96-yr record of flood peaks in Beaver Kill (Fig. 3; Table 1), a 624-km2 drainage basin in the mountainous upper Delaware River basin (Fig. 1). The record peak of 1767 m3 s−1 (2.83 m3 s−1 km−2) on 28 June 2006 in Beaver Kill culminated an 11-yr period in which the five largest flood peaks in the 97-yr record occurred (Fig. 3). The third largest peak in Beaver Kill occurred on 19 January 1996 and was produced by a combination of rapid snowmelt and extreme short-term rainfall rates from a squall line in mountainous terrain [see Barros and Kuligowski (1998) for detailed analyses]. The flood response in Beaver Kill receives particular attention in the following sections.
Flood response along the main stem of the Delaware River for the 2004–06 flood events is characterized by increasing flood peak magnitudes in the mountainous upper Delaware from Callicoon to Montague and flood wave translation in the reach from Montague to Trenton (Fig. 2; see Fig. 1 for locations). The travel time of flood peaks from Montague to Belvidere is approximately 12 h, and the travel time from Belvidere to Trenton is 10 h (Fig. 2). Travel time from Callicoon to Montague is 8–9 h (Fig. 2; see also Fig. 1). The Delaware River basin above Montague (4714 km2) is largely composed of the East Branch and West Branch Delaware subbasins. Beaver Kill (Figs. 1 and 2) is a subbasin of the East Branch Delaware. The drainage area of the Delaware River basin increases to 7951 km2 at the Port Jervis gauging station, with the Lackawaxen basin the largest of the tributaries between Callicoon and Port Jervis (Fig. 1). The drainage area of the Delaware increases to 9013 km2 at Montague, 11 746 km2 at Belvidere, and 17 560 km2 at Trenton, with the Lehigh River the largest tributary in the lower basin (Fig. 1).
The June 2006 flood peaks were larger than the April 2005 flood peaks at Callicoon, Port Jervis, and Montague (Table 1; Fig. 2). The two peaks are virtually identical at Belvidere, 6400 m3 s−1 in April 2005 versus 6371 m3 s−1 for June 2006. The April 2005 peak edges ahead of the June 2006 peak at Trenton.
Runoff for each of the three events is preferentially generated in the high-elevation portions of the Delaware basin (Table 1; runoff is computed by integrating discharge over the duration of the event and converting volume to depth using the drainage area of the basin). For each of the three events, runoff depth decreases from the upper Delaware basin to the lower Delaware basin (as reflected in runoff depths for the Trenton station). The runoff depths for the Delaware at Callicoon (upper Delaware) and Trenton (lower Delaware) are 105 and 88 mm for the September 2004 event, 204 and 183 mm for the March–April (MA) 2005 event, and 193 and 125 mm for the June 2006 event, respectively.
a. September 2004
Hurricane Ivan was a long-lived hurricane that formed off the west coast of Africa on 31 August 2004 and reached category 5 strength on the Saffir–Simpson hurricane scale (see Fig. 4 for a partial track of Hurricane Ivan). Ivan made landfall as a category 3 hurricane along the Gulf Coast of Alabama around 0700 UTC 16 September (Fig. 4). Ivan moved northeastward over Alabama, weakening to a tropical storm by 2000 UTC and a tropical depression by 0000 UTC 17 September. Northeastward motion continued until approximately 1200 UTC 18 September, when the interaction of Ivan’s circulation with a cold front resulted in the low pressure center curving south (Fig. 4). Ivan ultimately restrengthened and made a second landfall along the east coast of Florida (Fig. 4). Extratropical transition of Ivan, as with many tropical cyclone flood events in the northeastern United States (see Bosart and Dean 1991; Hart and Evans 2001; Atallah and Bosart 2003; Colle 2003), played a central role in heavy rainfall and flooding in the Delaware River basin from Hurricane Ivan.
Storm total rainfall analyses from rain gauge observations for Ivan (Fig. 5) show that heavy rainfall was concentrated along the eastern margin of the central Appalachians. In the Delaware River basin, maximum rainfall accumulations were located in high-elevation portions of the Lehigh River basin, which included an area of accumulations greater than 160 mm. An axis of rainfall accumulations exceeding 140 mm extended along the main southwest–northeast strike of the mountains. The largest accumulations were in the southern end of the mountain region, with accumulations exceeding 180 mm (Fig. 5).
Heavy rainfall over the Delaware River basin was organized into convectively intense rainbands. Cloud-to-ground (CG) lightning observations (figure not shown) show that thunderstorms in complex terrain contribute to heavy rainfall and flooding from Ivan. Notable examples of tropical systems that produce extreme rainfall along the eastern margin of the Appalachians from intense thunderstorm systems include Hurricane Camille in Nelson County, Virginia (see Schwarz 1970), and Tropical Storm Allison, which produced local accumulations of more than 400 mm in June 2001 in eastern Pennsylvania (see Javier et al. 2010).
Beaver Kill (Fig. 1) produced the largest runoff value—143 mm—for the September 2004 storm (see Table 1). Contoured rain gauge accumulations for Beaver Kill ranged from more than 110 to 150 mm (Fig. 5), with a basin-averaged accumulation that is slightly smaller than 143 mm. Beaver Kill produced anomalously large peak discharge and runoff values, compared with other Delaware River basin stream gauging stations, for each of the three flood events (Table 1). Discharge estimates for large floods are subject to significant errors (e.g., Potter and Walker 1981; House and Pearthree 1995), and the question arises as to whether the Beaver Kill peak discharge and runoff values are simply in error. The alternative explanation requires extreme runoff values for Beaver Kill with runoff ratios (i.e., the ratio of runoff to rainfall) that are close to 1 and pronounced contrasts in flood response properties between Beaver Kill and surrounding watersheds. Analyses presented next for the September 2004 storm and in subsequent sections suggest that the latter holds: that the Beaver Kill peaks and runoff values are accurate and that they are markedly larger than other regions of the Delaware River basin.
The 143-mm runoff value for Beaver Kill is consistent with runoff values from the East Branch Delaware at Harvard and Fishs Eddy (Table 1; see also Fig. 1 in which Harvard is marked by a “b” and Fishs Eddy by a “c”). The Harvard stream gauge, which is upstream of the confluence with Beaver Kill and downstream from the Pepacton Reservoir, measures runoff from 1186 km2 of the East Branch Delaware. The Fishs Eddy stream gauge is downstream of the confluence with Beaver Kill and has a drainage area of 2030 km2. The runoff from the East Branch Delaware above Harvard for the September 2004 flood was 89 mm; runoff from the basin above Fishs Eddy was 111 mm. Weighting the runoff values for Beaver Kill (143 mm) and Harvard (89 mm) by their drainage areas yields an average runoff of 107 mm, slightly less than the observed value of 111 mm for Fishs Eddy. Runoff values less than 143 mm in Beaver Kill for the September 2004 are, therefore, not consistent with the basinwide water balance and analogous results hold for the other events (Table 1).
The 89 mm of runoff at Fishs Eddy, which is largely controlled by the Pepacton Reservoir (see Fig. 1), was slightly larger than runoff at Margaretville (denoted “a” in Fig. 1), which was 88 mm. Operation of the Pepacton Reservoir did not markedly alter the water balance of the September 2004 flood.
The second and third largest runoff values from USGS gauging stations were located in the area of maximum rainfall accumulations (Fig. 5) in the Lehigh River basin. Runoff from Tobyhanna Creek (denoted “4” in Fig. 1) was 139 mm, and runoff from the Lehigh River at Stoddardsville (denoted “3” in Fig. 1) was 129 mm. Rainfall analyses (Fig. 5) suggest that storm total rainfall accumulations were 120–160 mm in the Tobyhanna and Lehigh Stoddardsville basins.
The Delaware River basin above Belvidere is largely contained within the 100-mm rainfall contour for the September 2004 storm (Fig. 5). Rainfall accumulations drop sharply in the lower basin, with accumulations less than 20 mm at Trenton. Runoff in the mainstem Delaware is approximately 100 mm for the Port Jervis, Montague, and Belvidere stations, with a slightly larger value in the upper Delaware above Callicoon (Table 1). Lower rainfall in the low-elevation portion of the basin leads to a drop in runoff to 88 mm at Trenton. These analyses show that high-elevation watersheds in the Delaware River basin have runoff ratios for the September 2004 flood that are in the 70%–100% range.
b. April 2005
Extreme flooding in the Delaware River basin on 2 April 2005 was associated with an extratropical cyclone that tracked up the East Coast. The surface low was located along the Gulf Coast near New Orleans at 1200 UTC 1 April with a minimum sea level pressure of 1009.7 hPa; 12 h later, central pressure had decreased to 1001.1 hPa and the low was centered in the southern Appalachians (figure not shown). From 1200 UTC 2 April to 0600 UTC 3 April, the cyclone tracked from northern Virginia to the northwestern corner of the Delaware River basin in Pennsylvania with minimum sea level pressure decreasing from 995.5 to 981 hPa during the 18-h period. Heavy rainfall in the Delaware River basin on 2 April was associated with a rapidly deepening cyclone, frontogenesis, and strong flow of moist air at 850 hPa.
Storm total rainfall accumulation from the operational rain gauge network for 2 April 2005 (Fig. 6) shows a 75–120-mm swath of maximum rainfall oriented southwest to northeast along the main axis of the central Appalachian Mountains. The largest rainfall accumulations were located in high-elevation regions of the upper Lehigh River basin. Low-elevation regions in the southeastern portion of the Delaware River basins received rainfall accumulations ranging from 20 to 60 mm. No lightning was detected over the Delaware River basin for the April 2005 storm.
Flooding in the Delaware River basin was affected by both rainfall on April 2 and snowmelt from precipitation during the end of March (see Fig. 2b). Separating the snowmelt contribution to runoff, particularly for drainage basins affected by reservoir storage (as discussed below), and for large drainage areas, is difficult. Runoff computations in Table 1 for the April 2005 flood extend from March 27 through April 6 and include the period of rain and snowmelt in late March.
Flood response in the upper Delaware River basin for the April 2005 flood, like the September 2004 and June 2006 floods, was inordinately influenced by Beaver Kill (Table 1 and Fig. 7). The peak discharge in Beaver Kill of 1439 m3 s−1 results in a unit discharge peak—2.31 m3 s−1 km−2—that was more than twice the unit discharge peak from the smaller and adjacent East Branch Delaware River at Margaretville (location is denoted “a” in Fig. 1). The peak discharge for the East Branch Delaware River at Fishs Eddy (denoted “c” in Fig. 1) of 1842 m3 s−1 reflects a dominant contribution from the Beaver Kill tributary (1439 m3 s−1) and a much smaller contribution—603 m3 s−1—from the larger East Branch at Harvard (denoted “b” in Fig. 1). The Fishs Eddy peak discharge was 90% of the sum of peak discharge from Beaver Kill and Harvard—2042 m3 s−1—suggesting that the Beaver Kill and Harvard peaks contribute nearly synchronously at the downstream station and that there is only modest flood wave attenuation. The 195-mm runoff for the East Branch at Fishs Eddy reflects a much larger contribution—287 mm—from Beaver Kill and a much smaller runoff—142 mm—from Harvard (Table 1).
The total runoff for Beaver Kill (including the snowmelt runoff) of 288 mm was 107 mm larger than the runoff from Margaretville (Table 1). The timing of runoff production from the two basins was not, however, markedly different (Fig. 7). In response to the 20-h period of rainfall on 2 April, Beaver Kill and Margaretville both peak between 0500 and 0600 UTC 3 April. The East Branch Delaware gauging station at Harvard (denoted “b” in Fig. 1) is downstream of the Pepacton Dam and reservoir (which is downstream of the Margaretville gauging station). The reservoir was near capacity due to the preceding snowmelt runoff, limiting its effects on downstream flood response (Fig. 7). The runoff of 141 mm from Harvard, which is smaller than the upstream Margaretville runoff, as well as runoff values from other stations in the region, reflects the fact that a net increase in reservoir storage occurred over the March snowmelt and 2 April rain period.
Runoff for the Lackawaxen River basin at the downstream station at Hawley—267 mm—was larger than the upstream station at Honesdale (243 mm) and smaller than the 311 mm at the headwater, high-elevation station at Aldenville (Table 1; Fig. 7). Streamflow in the Lackawaxen River below the two major dams (Prompton and Wallenpaupack) reflected the storage of snowmelt runoff and release for dam safety during and following the 2 April rain event (Fig. 7).
c. June 2006
Extreme flooding in the Delaware River basin was the product of heavy rainfall from warm-season convective systems from 1200 UTC 25 June to 1200 UTC 28 June 2006. Storm elements during the 3-day period produced pulses of runoff that are reflected in hydrograph variability at the upper Delaware stream gauging stations (Callicoon and Port Jervis; Fig. 2, bottom). The synoptic setting of the June 2006 flood was characterized by a stationary midtropospheric trough over the Ohio Valley (figure not shown), with a series of weak frontal boundaries organizing rainfall during the 3-day period. The storm environment was characterized by a deep layer of humidity, with large precipitable water values and strong moisture transport in the lower atmosphere to the east of cold fronts. There was virtually no cloud-to-ground lightning from the mesoscale convective systems during the 26–29 June period, reflecting the nearly tropical moisture conditions [see Petersen et al. (1999) for a related discussion].
The storm total accumulation for the 3-day period (Fig. 8) exhibited an axis of rainfall accumulations greater than 300 mm extending in a southwest to northeast orientation along the easternmost ridge of the central Appalachians. Maxima in the rain gauge analysis exceeded 400 mm.
Hydro-NEXRAD rainfall analyses for the upper Delaware region (Fig. 9) provide additional detail on the sharp gradients of storm total rainfall over complex terrain. An axis of maximum rainfall extends from the high-elevation headwaters of the Lackawaxen basin through the confluence of the East Branch and West Branch into the downstream portion of the West Branch. The gradient in storm total rainfall over the Lackawaxen basin is large, with peak accumulations in the basin above the Aldenville stream gauge exceeding 350 mm and accumulations at the basin outlet at Hawley of only 100 mm.
Rainfall was organized into lines of convection (Fig. 10) during a critical period of the flood episode. At 0600 UTC 28 June, a line of extreme rainfall extended from south of the Lackawaxen, through the western margin of the Lackawaxen, and into the downstream reach of the East Branch Delaware. This rain period produced a second peak in the upper Lackawaxen and was responsible for the final rise to peak discharge in the lower Lackawaxen (Fig. 10, top, and Fig. 11). By 0730 UTC 28 June (Fig. 10, bottom, and Fig. 11), the line had moved to the east of the Lackawaxen and was producing record flooding in Beaver Kill. This storm element was also responsible for the extreme flooding in the East Branch Delaware below Beaver Kill and the East Branch at Harvard (Table 1).
Rain gauge time series (Fig. 12) from three stations along a transect of decreasing elevation in the Lackawaxen River basin highlight the systematic decrease in rainfall with elevation throughout the period. Analyses are consistent with the “seeder–feeder” mechanism of orographic enhancement (Houze 1993). For the Lackawaxen River basin, these analyses suggest that convective initiation was not as important as mechanisms that enhanced preexisting convection.
Storm total runoff for the Lackawaxen River at Aldenville (Table 1; denoted “1” in Fig. 1) of 328 mm is larger than rainfall accumulations from rain gauge, radar, or model analyses. It is, however, consistent with downstream runoff measurements. The Lackawaxen at Hawley (outlet of the basin highlighted in Fig. 1) had runoff of 272 mm. The southern tributary had runoff of 240 mm. High-elevation headwater catchments produced the largest runoff values, consistent with the down-gradient rainfall observations (Figs. 10 and 12).
As with the 2004 and 2005 events, flood peaks and runoff values in Beaver Kill are anomalously large. The 188-mm runoff for Beaver Kill was more than 50 mm greater than the runoff in the adjacent East Branch Margaretville basin (Table 1). Runoff increases downstream in the East Branch from 132 mm at Margaretville to 144 mm at Harvard and 183 mm at Fishs Eddy (only 5 mm less than Beaver Kill). The large runoff values in the lower portions of the East Branch are largely due to extreme rainfall from the line of convection illustrated in Fig. 10 (refer to the earlier discussion). The largest runoff values in the Delaware are in the high-elevation Lehigh basin watersheds that received the largest storm total rainfall (Table 1).
4. Regional flood frequency in the Delaware River basin
In this section, we examine regional flood frequency of the Delaware River basin by synthesizing analyses presented in the preceding section with analyses of annual flood peak distributions in the basin. Regional flood frequency is summarized through analyses of 1) nonstationarity of flood peaks, as represented by changepoints (CPs) and slowly varying trends in annual flood peak observations; 2) variation in flood quantiles for annual peak observations from stations with “long” records; and 3) the frequency of tropical cyclone floods (such as the September 2004 flood), winter–spring extratropical floods (such as the April 2005 flood), and floods produced from warm-season convective systems (such as the June 2006 flood).
Analyses of changepoints and trends in flood peaks utilize the procedures presented in Villarini et al. (2009) to examine stationarity of flood peak distributions for gauging stations in the Delaware River basin with records longer than 50 yr. In this study, we use the nonparametric Pettitt (1979) test as in Villarini et al. (2009) for changepoint analyses. It is a rank-based test that uses a version of the Mann–Whitney statistic to test whether two samples—(X1, … , Xm) and (Xm+1, … , Xn)—come from the same population. It allows detection of changes in the mean when the changepoint time is unknown. In this study, changes in variance are tested by applying the Pettitt test on the squared residuals (Villarini et al. 2009; Villarini and Smith 2010). Evaluation of the presence of temporal trends in annual maximum peak discharge is based on the Mann–Kendall test (e.g., Helsel and Hirsch 1993; McCuen 2003; Kundzewicz and Robson 2004).
There are pronounced nonstationarities in the annual peak records of the Delaware River basin and changepoints dominate long-term trends as a mode of nonstationarity (Table 2). Stations with changepoints are principally linked with reservoir regulation upstream of and close to the gauging station.
Three of the four Lackawaxen gauging stations with long records exhibit changepoints (Table 2). Changepoints occur in the early 1960s, immediately following the completion of Prompton Reservoir and Jadwin Reservoir in 1959 and 1960, respectively. Similarly, changepoints for gauging stations in the East Branch and West Branch of the Delaware River basin and in the Neversink River basin (USGS 01436500 and 01437500; Table 2; Fig. 1) are tied to completion of the major New York City water supply dams in the upper Delaware River basin. Completion of the Papacton in the East Branch in 1954, Cannonsville in the West Branch in 1964, and Neversink in 1955 coincide with detected changepoints at downstream stations. For the Lackawaxen, East Branch, West Branch, and Neversink gauging stations, changepoints are linked to upstream reservoir regulation.
The effects of reservoir regulation on changepoints and linear trends does not extend far downstream of the source of regulation. There are no significant trends or changepoints for annual flood peaks in the mainstem Delaware River (Table 2). Notably, there is no changepoint or trend at the Port Jervis station downstream of the upper Delaware and Lackawaxen Dams.
The conclusions are similar for the Lehigh River basin. The Francis E. Walter Dam regulates the Lehigh River downstream of Tobyhanna Creek and the Lehigh River at Stoddardsville (Table 2) at 821 km2. It is upstream of the Lehigh River gauging station at Walnutport (2303 km2). The Walter Dam controls approximately a third of the basin above Walnutport and was completed in 1961. There is, however, no changepoint or trend in the 61-yr annual flood series at Walnutport or the 102-yr record at the downstream gauging station at Bethlehem (3312 km2).
Nonstationarities in Delaware flood peak distributions are also associated with land use change. As noted in section 2, pockets of urbanization have developed in the Lehigh Valley. Annual flood peak records for Little Lehigh Creek at Allentown (USGS 01451500; Table 2), Jordan Creek at Allentown (USGS 01452000), and Monocacy Creek at Bethlehem (USGS 01452500) exhibit nonstationarities. These stations are in the Bethlehem–Allentown urban corridor, which is the principal area of urbanization in the Delaware River basin. The Little Lehigh and Jordan Creek stations exhibit changepoints in the late 1960s (Table 2) that are tied to rapid urban development on the periphery of Allentown predating stormwater management regulations imposed by the Clean Water Act and its revisions in the 1970s. The Monocacy Creek gauging station in Bethlehem exhibits a long-term trend associated with urban development in the watershed. Despite being one of the largest urban regions in the Delaware River basin, the effects of urbanization in the Bethlehem–Allentown corridor do not lead to changepoints or trends in flood peaks in the downstream Lehigh River gauging station at Bethlehem (Table 2). The effects of urbanization are not detectable in flood peaks for the main stem of the Delaware.
There are striking nonstationarities in flood peak time series for the East Branch Delaware River. The East Branch Delaware station at Fishs Eddy (2030 km2; 01421000 in Table 2) does not exhibit a changepoint in mean, but it does have a changepoint in variance in 1991. The nonstationarity of flood peaks is clearly illustrated through the ratio of the Beaver Kill annual peak to the downstream peak at Fishs Eddy (Fig. 13, top); also shown in Fig. 13 (bottom) is the annual peak time series for Fishs Eddy (see Fig. 3 for the Beaver Kill time series; see also the discussion of the Beaver Kill and Fishs Eddy response to 2004–06 flood events in the previous section). A changepoint in mean for Beaver Kill (01420500 in Table 2) is detected in 1968 with a changepoint in variance in 1984 (Table 2), both preceding the 11-yr period with the five largest peaks in the record (see preceding section; Fig. 3). Gauging errors are unlikely to be responsible for the large Beaver Kill peaks, as shown in the preceding section through analyses of the 2004–06 events. The increasing ratio of Beaver Kill peaks to Fishs Eddy peaks (Fig. 13) is linked to the pronounced increase in Beaver Kill peaks after the 1960s and the decrease in flood peaks associated with upstream reservoir regulation, as reflected in changepoints during the mid-1950s for the East Branch at Harvard (1186 km2) and Downsville (964 km2; see Table 2). These analyses point to the utility of examining nonstationarities in variability of flood peak series, in addition to nonstationarities in measures of central tendency of the flood series.
Nonstationarity in the Beaver Kill flood record presents an interesting challenge for flood frequency analyses. As noted earlier, Beaver Kill contributes inordinately to flood magnitudes in the upper Delaware River basin and its contributions have increased markedly over time (Fig. 13). Land use change in the Beaver Kill watershed has been modest compared to the changes associated with urbanization in the Bethlehem–Allentown region described earlier. The watershed remains largely forest covered, as described in section 2. Trends in hydrologic response in Beaver Kill were examined by Baldigo (1999), specifically to address the consequences of highway construction along the lower section of Beaver Kill. On the basis of observations through 1994, it was concluded that “peak flows from storms recurring at 2-year and longer intervals after 1965 are significantly larger than those that recur at the same frequencies before 1965” (Baldigo 1999). This conclusion is based on observations that end prior to the 11-yr period of record flood peaks in Beaver Kill (Fig. 3)! The timing of changepoints identified both by the USGS study and in the present study follow the period of highway construction in Beaver Kill. There is not, however, a clear physical mechanism that links highway construction over a small portion of the watershed to striking increases in flood peaks and volumes.
Forest cover clearly does not prevent floods in Beaver Kill. More generally, analyses of the three floods in the previous section show that runoff production for large floods is inordinately concentrated in high elevation, forested portions of the Delaware River basin. Analyses suggest that runoff ratios close to 1 occur over large portions of the Delaware River basin for extreme floods.
The dependence of flood magnitudes on basin scale in the Delaware River basin is summarized in Fig. 14 through log–log plots of drainage area versus the maximum flood peak, the 10-yr flood peak, and median annual flood peak. Scaling properties of flood peaks exhibit distinctive features for basins smaller than 1000 km2, basins between 1000 and 10 000 km2, and basins larger than 10 000 km2. Flood quantiles for basins smaller than 1000 km2 are characterized by markedly larger variability than for basins larger than 1000 km2. As discussed in additional detail later, this property is tied to spatial heterogenereity of hydrologic and hydraulic processes. Flood quantiles exhibit a log–log linear increase with drainage area over the range from 1000 to 10 000 km2. For basins larger than 10 000 km2, the slope of the flood peak–drainage area relationship flattens.
The 10-yr flood magnitude for the Delaware River at 9013 km2 (Montague) has a unit discharge of 0.44 m3 s−1 km−2, which is virtually identical to the 10-yr flood magnitude of 0.45 m3 s−1 km−2 at 5232 km2 (Callicoon; Table 2). The 10-yr flood magnitude of the Lehigh River at 3313 km2 is also quite similar, taking a value of 0.41 m3 s−1 km−2 (Table 2). The 10-yr unit discharge decreases rapidly downstream from Montague, from 0.32 m3 s−1 km−2 at 11 746 km2 (Belvidere) to 0.26 m3 s−1 km−2 at 16 389 km2 (Riegelsville) and 0.25 m3 s−1 km−2 at 17 560 km2 (Trenton). Scale-dependent flood response for the 2003–05 flood events (section 3) mimics the scaling properties of flood peaks through flood generation in high-elevation portions of the watershed and flood propagation in the lower basin at basin scales larger than 10 000 km2.
Magnitudes of the 10-yr flood vary by an order of magnitude for basin scales in the 100–500-km2 range (Table 2; Fig. 14). The smallest values of flood quantiles (Table 2) are in the New Jersey Highlands (high-elevation areas of New Jersey between Montague and Belvidere; see Fig. 1). The 10-yr flood values range from 0.15 to 0.30 m3 s−1 km−2 for basin scales ranging from 80 to 326 km2. The largest values of unit discharge for the 10-yr flood are clustered in high-gradient watersheds of the upper Delaware. The Neversink River at 173 km2 (USGS 01436000) has a 10-yr unit discharge value of 2.0 m3 s−1 km−2. Three stations in the East Branch of the Delaware, including two in Beaver Kill, have 10-yr unit discharge values ranging from 1.21 to 1.48 m3 s−1 km−2. The Beaver Kill unit discharge value of 1.21 m3 s−1 km−2 at 624 km2 stands out on the high end of 10-yr flood magnitudes (Fig. 14).
Analyses of the three floods in the previous section point to the dependence of flood frequency distributions on differing flood-generating mechanisms. To conclude this section, we examine regional flood frequency in the Delaware River basin through analyses of the mixture of tropical cyclone floods, like the September 2004 event; flooding from winter–spring extratropical systems, like the April 2005 event; and warm-season convective systems, like the June 2006 event. Mixture distribution models have stimulated important insights to regional flood frequency (e.g., Rossi et al. 1984; Miller 1990; Villarini and Smith 2010) and provided useful models for characterizing and analyzing flood frequency.
Tropical cyclones account for less than 10% of annual flood peaks in much of the Delaware River basin (Table 2). An annual flood peak is identified as a tropical cyclone peak if the track of the cyclone, as represented by the National Hurricane Center’s Hurricane Database (HURDAT; Jarvinen et al. 1984; Neumann et al. 1993; see Fig. 4), passes within 500 km of the stream gauging station in a time window of 14 days centered on the time of occurrence of the flood peak. On the basis of this criterion, 8.7% of annual peaks for the Delaware River at Trenton (see Fig. 3; Table 2) are tropical cyclone events and 7.5% of Beaver Kill peaks (see Fig. 3; Table 2) are tropical cyclone events. The frequency of tropical cyclone flood peaks increases systematically down the mainstem Delaware: 4.5% at Callicoon, 5.8% at Port Jervis, 6.9% at Montague, 8.1% at Belevdere, and 8.7% at Trenton, reflecting the dependence of tropical cyclone flood frequency on basin scale.
Despite the low frequency of annual peaks, tropical cyclone floods play an important role in determining the upper-tail properties of flood peak distributions in the Delaware basin. The August 1955 flood peak from Hurricane Diane is the flood of record for 40% of the stations in Table 2 with observations from 1955. The August 1955 peak ranks second for 48% and in the top five for 60% of the stations with observations from 1955. There is significant diversity in the tropical cyclones that produce flooding in the Delaware River basin. Ivan (2004) and Agnes (1972) are the only two tropical cyclone flood events for the Delaware River at Trenton that make landfall along the Gulf Coast; the remaining seven tropical cyclone peaks make landfall along the Atlantic Coast (Fig. 4). The track of Ivan (Fig. 4) and rainfall distribution relative to the track of the storm (Fig. 5) reflect the importance of extratropical transition (see Atallah and Bosart 2003; Hart and Evans 2001) for flood frequency in the Delaware basin.
Winter–spring extratropical floods are the largest contributor to annual flood peaks in the Delaware River basin (Table 2), accounting for 40%–50% of annual flood peaks along the main stem of the Delaware. The percentage of floods that occurs during March and April (Table 2) is used as a surrogate for winter–spring extratropical floods. The fourth largest flood peak at Trenton—6424 m3 s−1 (0.37 m3 s−1 km−2) on 19 March 1936—was produced by the extratropical system that produced record flooding in major rivers throughout the Ohio River basin and eastern drainages of the United States and stimulated the development of systems of multipurpose dams and reservoirs in the eastern United States through the Flood Control Act of 1936 (Billington and Jackson 2006). Winter–spring extratropical systems are important elements of the upper tail of flood peak distributions for drainage basins—like the Delaware River at Trenton—that are larger than 10 000 km2.
Warm-season convective systems, like tropical cyclones, generally account for less that 10% of annual flood peaks in the Delaware River basin. Floods from warm-season convective systems are represented in Table 2 through the percentage of floods that occur during the June–August (JJA) time window, excluding those that are identified as tropical cyclone floods. The frequency of warm-season convective system floods is scale dependent, with values less than 8% for mainstem Delaware and Lehigh stations. The highest frequencies of warm-season flood peaks are in watersheds affected by urbanization in the Lehigh Valley.
The largest unit discharge flood peaks in the eastern United States and some of the largest rainfall accumulations in the world for periods shorter than 6 h have been produced by orographic convective systems, like the June 2006 storms. These systems dominate the envelope curve of flood peaks for the eastern United States at basin scales smaller than 1000 km2. The density of stream gauging stations in the Delaware River basin, and elsewhere, is inadequate to provide a reliable characterization of their frequency (both in time and space).
Mixture distributions provide useful avenues for examining the effects of climate change on flood peak distributions. Changing frequencies, tracks, and intensities of tropical cyclones and extratropical cyclones will have different signatures in the mixture of flood peaks from tropical cyclone, winter–spring extratropical systems, and warm-season convective systems.
5. Summary and conclusions
In summary, the major findings of this paper are as follows:
Record and near-record flooding occurred for three consecutive years from 2004 through 2006 in the Delaware River basin. The three flood events were produced by three flood-generating mechanisms in the eastern United States (as described by Miller 1990), landfalling tropical cyclones (September 2004), winter/spring extratropical cyclones (April 2005), and warm-season convective systems (June 2006). Analyses of the 2004–06 storms, along with flood frequency analyses for stream gauging stations with long records, highlight the role of “mixtures” of flood-generating mechanisms in determining regional flood frequency.
Flooding for each of the three events was disproportionately generated in high-elevation regions of the Delaware River basin. Orographic precipitation mechanisms play a central role in heavy rainfall and flooding for each of the three events and, more generally, for the flood hydrology of the Delaware River basin.
Landfalling tropical cyclones represent an important component of the flood hydrology of the Delaware River basin. Tropical cyclones account for approximately 8% of annual flood peaks in the Delaware River basin, but they play an important role in determining the upper tail of regional flood peak distributions. The flood of record at many gauging stations was produced by Hurricane Diane in August 1955. Hurricane Ivan ranks in the upper 10% of flood peaks at many sites. Extreme rainfall from Ivan, like most flood-producing tropical systems in the Delaware basin, was heavily influenced by the “extratropical transition” of the storm.
Winter–spring extratropical systems are responsible for record “large river” floods in the eastern United States, including the March 1936 flood. In the Delaware River basin, approximately 45% of annual flood peaks occur during March and April. The April 2005 storm combined snowmelt and heavy rainfall over a period of approximately 24 hours to produce the second largest flood peak in the Delaware River at Trenton from a record of 110 yr. Extreme rainfall in the Delaware River basin from the April 2005 storm was associated with a rapidly intensifying low pressure system that tracked up the East Coast during the period 1–2 April 2005 and produced rainfall maxima in the high-elevation portions of the basin.
Extreme flooding in the Delaware River Basin during June 2006 was the product of a series of mesoscale convective systems that were associated with a stationary trough over the upper Ohio River basin. Rainfall accumulations during the period 25–28 June 2006 exceeded 250 mm over a large portion of mountainous eastern Pennsylvania. Rainfall was concentrated in organized convective systems, including a line of intense rainfall rates that was responsible for record flood peaks in high-elevation headwater basins. Rain gauge observations along a sharp topographic gradient in the Lackawaxen River basin highlight the orographic amplification of rainfall that occurred systematically over the 4-day period. The convective systems that produced extreme rainfall and flooding in the mountainous terrain of eastern Pennsylvania and New York produced virtually no cloud-to-ground lightning, reflecting the dominance of warm-rain processes over the course of the flood episode. A persistent deep layer of moisture and strong low-level transport of moisture were characteristic features of the storm environment. The importance of warm-season convective systems as flood agents in the Delaware River basin depends strongly on basin scale. The frequency of warm-season flood peaks is highest—20% for smaller basins in the limited area of urban development. For large basins, the frequency of warm-season flood peaks decreases to approximately 7%.
Water balance analyses for the three flood events in the Delaware River basin suggest that runoff ratios close to 1 occur over portions of the basin for extreme floods. Forest cover clearly does not “prevent” floods in the Delaware River basin. Analyses of the three floods show that runoff production for large floods is inordinately concentrated in high elevation, forested portions of the Delaware River basin. There is, however, uncertainty in discharge estimates that determine the upper tails of flood peaks and storm total runoff for high-gradient watersheds in the Delaware River basin. The same conclusion holds for rainfall estimates associated with these peaks and runoff values. Measurement of extreme flood peaks remains an important challenge in characterizing regional flood frequency (e.g., Potter and Walker 1981; House and Pearthree 1995).
Analyses of flood frequency based on annual flood peak observations from USGS stream gauging stations with long records illustrate the striking heterogeneity of flood response over the region and the prevalence of nonstationarities in flood peak records. Analyses show that changepoints are a more common source of nonstationarity than linear time trends. Regulation by dams and reservoirs plays an important role in determining changepoints, but the downstream effects of reservoirs on flood distributions are limited.
Scale-dependent flood response in the Delaware River basin is linked to the interplay of the spatial and temporal distribution of rainfall for the principal flood-generating mechanisms, heterogeneities in runoff production, and drainage network structure of the basin. Future studies will explore modeling approaches to scaling analyses of flood peaks in the region, building on the procedures of Gupta et al. (2010).
The authors would like to acknowledge Joe Ostrowski (NOAA/NWS) for providing rain gauge data and helpful comments. NLDN data were provided by the NASA Lightning Imaging Sensor (LIS) instrument team and the LIS data center via the Global Hydrology Resource Center (GHRC) located at the Global Hydrology and Climate Center (GHCC), Huntsville, Alabama, through a license agreement with Global Atmospherics, Inc (GAI). The data available from the GHRC are restricted to LIS science team collaborators and to NASA EOS and TRMM investigators. This research was supported by NASA, the Willis Research Network, the NOAA Cooperative Institute for Climate Sciences, and the National Science Foundation (NSF Grants EAR-0409501, EAR-084734, and ITR-0427325).
Corresponding author address: James A. Smith, Department of Civil and Environmental Engineering, Princeton University, Engineering Quadrangle, Princeton, NJ 08540. Email: firstname.lastname@example.org