2. Data and Methodology

2. Data and Methodology

2.1 PRE Identification and Stratification

2.1.1 Use of Radar Imagery

The search for PREs started with a comprehensive inspection of Next Generation Weather Radar (NEXRAD) national mosaic reflectivity images between 1998 and 2006, available from the National Climatic Data Center (NCDC) at http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?WWNEXRAD~images2. Examination of the images began while a named TC of at least tropical depression strength was still well offshore according to the National Hurricane Center (NHC) best-track dataset (http://www.nhc.noaa.gov/pastall.shtml). The main criterion for initial PRE identification was the appearance of 35 dBZ or greater reflectivity values within a coherent area of rainfall persisting for at least 6 hours at a distinct separation distance (SD) from the TC. Radar depiction of one of the PREs ahead of Gaston (2004) is given in Fig. 2.1.

If the TC rain shield expanded to a great distance away from the center in a quasi-uniform manner as the storm underwent ET [e.g., Irene (1999)], the case was not included. If, instead, a separate area of rainfall developed ahead of a TC and then was ingested into the TC circulation [e.g., Floyd (1999)], the event was counted until the candidate PRE became indistinct from the rainfall directly associated with the TC. The synoptic-scale factors bringing about these respective scenarios were likely quite similar, but the distinction is made here because of the focus on significant areas of rainfall occurring as separate entities ahead of TCs.

2.1.2 Determination of Rainfall Amounts

A 24-h precipitation rate of 100 mm (24 h)−1 was the minimum threshold for inclusion in the PRE dataset. Rainfall amounts were compared to PRE duration (see section 2.2.1) and extrapolated to a 24-h rate, since the lifetime of many PREs was much shorter than 24 h. For example, a maximum of 25 mm falling during a PRE lasting 6 h would be extrapolated to 100 mm in 24 h, and thus exceed the threshold. There was no strict minimum areal coverage of this rainfall rate, but a significant attempt was made to limit PRE identification to areas of convection that would not be categorized subjectively as scattered or isolated by inspection of representative radar images.

The determination of rainfall amount for cases during 2001–06 was made using the National Precipitation Verification Unit (NPVU) quantitative precipitation estimate (QPE) archive. The data are a combination of radar-estimated and rain gauge precipitation amounts (McDonald and Baker 2001), available online at http://www.hpc.ncep.noaa.gov/npvu/archive/rfc.shtml. The NPVU QPE data obtained for this study were on a 10-km resolution grid encompassing the U.S. The NPVU plots agree well with manually prepared TC total rainfall plots obtained from http://www.hpc.ncep.noaa.gov/tropical/rain/tcrainfall.html. Starting in 2004, the NPVU data were supplemented by NWS cooperative observer reports noted in the public information statements of affected weather forecast offices when available. The public information statements sometimes reported point totals that were higher than corresponding values evident at the same point in the smoothed, gridded NPVU analyses.

For PREs occurring prior to 2001, the Unified Precipitation Dataset (UPD) was utilized on the basis of its applicability in prior related studies (e.g., DeLuca 2004; Srock 2005), as well as its completeness and consistency back to 1948 (see Higgins et al. 2000). The UPD gives a reasonably accurate spatial representation of precipitation occurring in 24-h intervals, but precipitation magnitude sometimes can be significantly underestimated (Atallah et al. 2007). Therefore, any events noted by the UPD as having 24-h rainfall rates greater than 100 mm would surely meet the minimum rainfall criterion.

2.1.3 Synoptic-Scale Environment

The final ingredient for inclusion in the PRE database was an indication that predecessor heavy rainfall was attributable to a TC. Twice-daily synoptic charts from the Storm Prediction Center archive at http://www.spc.noaa.gov/obswx/maps were examined to determine if the environmental flow transported moist, tropical air from the TC toward the candidate PRE through some tropospheric layer. For example, BC78 noted the poleward transport of tropical air in the surface–400 hPa layer with the Wellsville, NY, PRE ahead of Agnes (1972). If an area of heavy rainfall identified as a potential PRE demonstrated no signs of receiving tropical air from the vicinity of the TC, it likely formed independently of the TC and is not considered further.

The rainfall associated with a few potential PREs occurred predominantly just offshore, where land-based impacts were minimal and accurate rainfall accumulation data were not available. Two such cases were retained because they exhibited radar reflectivities and lifetimes consistent with cases occurring over land that were counted as PREs based on rainfall amount. The inclusion of these two offshore cases in the PRE database only occurred after determining that they formed in synoptic-scale environments similar to those for which definitive rainfall data were available.

2.1.4 PRE Stratification

Table I presents the PRE database after classification into groups relative to the track of the rainfall of the parent TC: left of track (LOT), along track (AT), and right of track (ROT). This type of categorization is intended to link to previous studies (e.g., DeLuca 2004; Atallah et al. 2007) that made similar references to the location of heavy precipitation directly associated with a TC relative to the track of the TC center. The motivation for the stratification in the current study is to determine how common it was for TC rainfall to pass over the same area that was previously affected by a PRE.

In most cases, the combination of rainfall totals, radar imagery, and NHC best-track data was enough to determine the appropriate category of a given PRE. Fig. 2.2 gives a schematic depiction of the methodology used in ambiguous cases. The total rainfall encompassing the time period in which both the PRE and the TC rainfall affected an area was calculated and compared with the rainfall attributed only to the PRE, a partition made possible by the availability of reliable NPVU 6-h rainfall amounts. If the PRE rainfall comprised 25–75% of the total rainfall for the entire period, the PRE was classified as AT because the location received two episodes of heavy rainfall—one from the PRE and one from the TC. The PRE depicted ahead of the representative TC in Fig. 2.2 would be classified as AT because approximately 75 mm of the 150 mm (about 50%) of rain that fell in southern New York could be attributed to the PRE. If the PRE percentage was below (above) the 25–75% range, then most of the rainfall in the area was due to the TC (PRE), and the PRE was deemed LOT or ROT. Hypothetically, the rainfall associated with a TC (PRE) may be so excessive in a given area that the PRE contribution is small (large), even though the TC and PRE pass over the same region, but neither of these scenarios was observed in any of the ambiguous examples.

2.2 Climatology

2.2.1 Statistical Analysis

An extensive catalog of information on each PRE was assembled for statistical analysis. PREs were assigned a position every 3 h based on their rainfall centroid as seen on radar imagery. PRE start and end times were governed by the presence of a 35 dBZ or greater reflectivity contour within a coherent area of precipitation, and PRE lifetimes were calculated from these respective times. A SD was then calculated every 3 h based on the assigned PRE position and the NHC best-track position of the given parent TC (Fig. 2.3). The best-track position was linearly interpolated to 3-h intervals when necessary. The SD calculation was made using a U.S. Geological Survey straight-line distance calculator, available through ftp://kai.er.usgs.gov/pub. Figure 2.3 also illustrates the method used to calculate the time lag, defined as the time required for the center of the TC to reach the latitude of the PRE centroid. Finally, a maximum rainfall amount was assigned to each PRE based on the data and methods described in section 2.1.2.

Statistical analysis of the assembled quantities consisted of the calculation of means, medians, and standard deviations for PREs in the dataset as a whole, and for the LOT, AT, and ROT groups individually. Rainfall statistics were calculated only for post-2000 cases because these correspond to the PREs for which the NPVU data were available. Histograms and box plots were created from these statistics to illustrate physical properties of PREs and their variability.

2.2.2 Composite Analysis

All TCs during the study period, regardless of whether PREs developed, were divided into track categories based on similarity of TC track, as determined from NHC best-track data. The motivation for this division was to determine if certain TC tracks were more likely to favor PRE formation than others. Of all the TC track categories, the three containing the highest fractions of PRE-producing TCs were: 1) Southeast recurvature (SR) TCs, which all made landfall on the Florida coast and exhibited an eastward component to their poleward motion by 32°N so that their centers passed along or east of the spine of the Appalachians; 2) Atlantic recurvature (AR) TCs, which originated over the Atlantic Ocean and recurved near or off the eastern U.S. coastline such that the center passed northwest of 35°N, 68°W, regardless of the occurrence of landfall; and 3) central Gulf (CG) TCs, which made landfall between the Texas/Louisiana border and the Alabama/Florida border and then moved north or northeastward subsequent to landfall. A complete breakdown of the TCs in each of the three categories is given in Table II. TCs making landfall southwest of the Texas/Louisiana border were placed into a separate category to distinguish them from CG TCs, which made landfall further east. TCs exhibiting an eastward component to their motion in the Gulf of Mexico south of 28°N and passing southeast of 30°N, 80°W also were placed into a separate category to distinguish them from SR TCs, which moved further north.

National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis 2.5° × 2.5° gridded datasets (Kalnay et al. 1996; Kistler 2001) were obtained for each PRE case exhibiting a common synoptic-scale signature in the SR group. These datasets were used for compositing because they give complete spatial and temporal coverage of the study period. Composites were constructed for a 24-h period centered on the initial time of PRE formation and displayed using the General Meteorology Package (GEMPAK) (Koch et al. 1983). The motivation for compositing cases in groups based on similarity of TC track is to derive a representative schematic of synoptic-scale features important in PRE formation when the approximate track the parent TC will take may be anticipated in real-time forecasting scenarios.

The compositing approach used in this study differs from that used by Atallah et al. (2007), who composited the TCs in their LOT and ROT categories relative to a representative storm track so that they could account for the positions of pertinent large-scale features relative to the TCs. A consequence of the Atallah et al. (2007) method is that the geography upon which the features are overlaid is artificial, but their method likely produces a cleaner signal of the synoptic-scale features than the method employed in the current study. However, since the members of the current composites were chosen based on similarity of TC track, these composites should still isolate the dominant synoptic-scale signatures present during PREs. Nevertheless, differences in the absolute positions, sizes, strengths, and orientations of the TCs and the associated synoptic-scale features may cause some signal distortion. Composite TC center and PRE centroid positions were calculated and plotted on synoptic-scale maps every 6 h so their positions could be compared with the positions of synoptic-scale features.

2.3 Case Studies

Case studies were selected for their ability to illustrate synoptic-scale patterns for LOT, AT, and ROT PREs and to demonstrate the variety of mesoscale features that can contribute to PRE formation. Specifically, two LOT PREs will be examined in conjunction with Gaston (2004), and one AT and one ROT PRE will be documented in conjunction with Katrina (2005). The events surrounding the predecessor rain activity well ahead of Fran (1996) will also be examined. Finally, the null case of Cindy (2005), chosen because of its similar track to Katrina (2005), will be contrasted with the PRE cases to suggest possible factors determining PRE formation and location.

TC tracks were overlaid on storm total and 24-h rainfall maps for each case using the NHC best-track and NPVU datasets [except for the use of the UPD for Fran (1996)] to show the locations and magnitudes of rainfall associated with the TCs and the PREs. These maps use total QPE during the period of TC and PRE rainfall, and thus can be used to visualize whether and where predecessor rainfall activity occurred in relation to the TC track. Additionally, Weather Services International (WSI) Corporation NOWrad national radar composites were utilized to depict PRE evolution because of their flexibility in display area and 15-min temporal resolution. The WSI data files are available starting in 2002 at http://locust.mmm.ucar.edu/WSI/access_WSI_data.html. For a complete description, see Grassotti et al. (2003). Since the WSI NOWrad composites were not readily available for Fran (1996), reflectivity images for selected WSR-88D sites were downloaded from the Hierarchical data storage system Access System (http://has.ncdc.noaa.gov) and viewed in the NCDC Java NEXRAD Viewer and Data Exporter, which is available for free download at http://www.ncdc.noaa.gov/oa/radar/radardata.html#DOWNLOAD.