IDENTIFICATION OF FLORIDA COASTAL SHOWERS Brian C. Zachry and Steven M. Lazarus Florida Institute of Technology, Melbourne, Florida ABSTRACT Showers that occur along the east coast of Florida during the fall months (Sept., Oct. and Nov.) are often referred to as “coastal showers”. Four years of hourly surface data were sampled to construct two distinct data sets in an effort to identify these events. The first data set consisted of data in which hourly observations (events) were selected using specific criteria for time, wind speed/direction and precipitation amount. A second data set is produced using a linear regression of the along-coast pressure gradient to select days in which onshore flow is assumed strong enough to produce coastal showers. Results indicate that the number of ‘identified’ events are sensitive to the selection criteria (and station location). These data sets will be used to better understand the nature of the phenomenon, including their propensity for a nocturnal maximum, their inter-seasonal variability, and precipitation distribution and amount. 1. INTRODUCTION Relatively significant precipitation events that occur along the east coast of Florida during the fall months are often referred to as “coastal showers”. These easterly flow events are associated with the easterly low-level fetch from a surface high pressure system moving off or remaining anchored along the coast of the east/central United States mainland. Observations indicate that the coastal shower events range from ephemeral (i.e. on the order of a few hours) to extended periods of up to several days in some cases. In part, coastal showers appear to be a result of air mass modification as relatively cool/dry air is advected over the Atlantic Ocean (along the eastern/southern periphery of the surface high pressure) towards the east coast of Florida. The impact of the Gulf Stream, although not totally understood, appears to plays a role in coastal shower development along the Florida east coast. Observations suggest that Gulf Stream SSTs, being a few to several degrees warmer than surrounding SSTs during fall months, acts as an energy source to help initiate showers (provided sufficient moisture availability). Once formed, the coastal showers are typically advected towards the coast by relatively strong low-level easterly winds that accompany these events. The inland land mass, usually cooler than adjacent ocean, often causes nocturnal coastal showers to dissipate rapidly with increasing inland penetration. For the most part, coastal locations tend to observe more precipitation than do inland stations - the latter of which often observe little or no precipitation. The shower events appear to have a nocturnal maximum intensity prior to sunrise and have often been observed to propagate inland shortly after sunrise. Ultimately, little is known concerning the specific forcing (initiation) mechanisms responsible for coastal showers but given their relationship with the large-scale flow, coastline, SSTs, etc., they appear to involve complex interactions between the synoptic and mesoscale. 2. Data Integrated Surface Hourly Weather Observation data were downloaded from the National Climatic Data Center (NCDC) for 19 stations along the east coast of Florida (Table 1). The data currently ranges from 1 January 2000 to 9 October 2004 with anticipation that it will eventually be extended back to 1994. The data contains the typical variables included in all 1 hourly surface weather observations. However, in order to identify coastal shower events, the parameters of interest include: wind direction (degrees), wind speed (m/s) and precipitation total (mm). Integrated Surface Hourly Weather Observation data were mined to construct two distinct data sets in an effort to identify coastal shower events. The first (initial) data set consisted of hours in which hourly observations met specific criteria to denote a precipitation event as a coastal shower (see next section for details). This data set, clearly not completely robust, used the variables above (and time of day), to construct a data set of hourly coastal shower events. The second (alternative) data set is comprised of days in which the pressure gradient from JAX to PBI is greater than 3 mb. This set acts as a surrogate for the low level wind flow since light and variable winds are often observed during the morning hours at land stations. The idea is to identify potentially significant easterly flow wind events, for the initial data base, that might otherwise be overlooked due to nocturnal decoupling effects on the 10 m wind. The pressure gradient is an arbitrary minimum threshold chosen to represent an onshore (i.e. easterly) geostrophic wind, for which to identify coastal shower events. Methods used to construct the data sets are shown in the following section. 3. Methods a. Weather Observing Stations Weather observing stations nearest the east coast were selected for the initial data set. These coastline stations were chosen to ensure the highest probability that coastal showers would be identified by reported precipitation amounts. Ultimately, the study will be enhanced to include inland stations in order to quantify what appears to be the dissipation of showers with increasing inland penetration. Figure 1: Weather observation station map of Florida where station locations are depicted by a black dot with corresponding station IDs to the upper-right. 2 b. Identifying Potential Coastal Shower Events 1) Initial Data Set Specific criteria were chosen so as to maximize the probability of identifying a coastal shower event using the NCDC data. Events were identified based on the following: (1) events are limited to occurrence in September, October and November; (2) stations are limited to DAB, VRB, MLB, TTS and COF (3) precipitation totals must be reported (i.e. not 0 mm or ‘missing’); (4) a wind speeds must be reported (i.e. not 0 mph or ‘missing’); (5) wind directions are limited to 345° to 180° and can be ‘missing’; (6) times are limited to 2340 to 1220 UTC (roughly 8PM – 8AM EDT or 7PM – 7AM EST). An observation in which precipitation total was ‘missing’ was discarded as it’s the fundamental metric that defines an event. An event in which a ‘missing’ wind speed and/or direction occurred was retained since discarding such data could result in the exclusion of a coastal shower event where the weather observing station failed. It is possible that the missing wind speed and/or direction can be inferred from nearby stations, an earlier observation or the low-level synoptic conditions. 2) Alternative Data Set Florida coastal showers are related to easterly low-level synoptic-scale wind flow events over the region. The data that constitute this portion of the project reflect cases where a strong onshore geostrophic wind component, as determined from the pressure gradient over Florida, exists. A sequence of steps was taken in order to ensure the quality and consistency of the selected data. The first of these was to reduce the data such that only one observation per hour per station remained (provided the data was not ‘missing’). The most recent (i.e. latest) observation from 30 minutes before the hour to on the hour (e.g., the latest observation from 1230 and 1300 was used for the 1300 observation time) was taken for each observation time (i.e. 00 UTC – 23 UTC). Secondly, an alongshore n̂ coordinate axis was defined as positive north (Figure 2, left). This natural coordinate yields a relatively clear delineation between an onshore and offshore wind, with its orientation determined so as to compensate for Florida’s offset from true north. The coordinate origin, Jacksonville (JAX) was defined to be located at 0 km with all remaining stations (south of JAX) defined in terms of a negative distance (i.e. in the - n̂ direction). Since not all the stations used in the calculation lie directly on n̂ , the distances to these stations from JAX were determined by the method shown in Figure 2. The negative distance along n̂ (red), where the normal (blue) line intersected n̂ , was used to determine the station distance from the coordinate origin (see distance determination for station SGJ shown in Figure 2, right). 3 Figure 2: (Left) Defined n̂ axis (red) used to determine the distance from origin (JAX) (i.e. along n̂ ) to each station via the intersection of the normal line (blue) drawn from the station location and the n̂ axis. Weather observing station locations are depicted by a black dot with corresponding station IDs to the upper-right. (Right) Modified zoom image from (Left) showing the determined distance (green) along n̂ to station SGJ. The distance determination was used in combination with reported surface pressure values to obtain an average pressure gradient across the stations for each day. To accomplish this, a Perl statistics regression module was used to calculate a ‘best-fit’ line (the straight line that best represents the trend that the points in a scatter plot follow) to pressure using the method of “least-squares”. The least-squares method minimizes the difference between the ‘estimate’ y (here, assumed to be linear), i.e. y mx b (1) and observations yo (here, surface pressure). Here, ‘x’ represents the station distance as measured from the coordinate origin (JAX), the coefficient b (intercept) is the ‘estimated’ pressure at zero distance (i.e., JAX) and m (slope) is the (best-fit) pressure gradient ( dp dn ) along the normal. Minimization of the quadratic [y – yo]2 yields the slope (m) and intercept (b), respectively m n xy x y (2) n x 2 x 2 4 b n m x (3) n where n is number of stations. A correlation coefficient r is defined as r n xy x y (4) n x 2 x * n y 2 y 2 2 The statistics regression module returns R² (coefficient of determination), dp dn , n and b (although not important here) for each hour. Estimates of the pressure gradient are restricted to cases where more than 8 stations (n) were used and R² was greater than or equal to 0.65. Here, a ‘significant’ correlation indicates a value greater than 0.8 (e.g., Tyrrell et al. 2004, MBJ 2004, Mathbits 2004). This corresponds to an R² value of 0.65 or greater. The coefficient of determination represents the percent of the data that is closest to the line of best fit or, in other words, a measure of how well the independent variable (here, distance) predicts the dependant variable (here, pressure). The cutoff-value selected for R² means that 65% (at minimum) of the total variation in y can be explained by the linear relationship between x and y (as described by the regression equation). The other 35% (at maximum) of the total variation in y remains unexplained by the regression. An average pressure gradient (and coefficient of determination) was calculated by averaging the relevant regression outputs over each hour. The average pressure gradient and coefficient of determination were output only if dp dn was greater than a pre-determined cutoff value (3 mb). The distance and orientation of the n̂ axis indicates that the pressure gradient must be positive (i.e., pressure increasing in the +n direction, a surface high pressure to the north) in order to result in an onshore geostrophic wind. A back-of-the-envelope calculation of the geostrophic wind using a pressure gradient of 3 mb yields a significant onshore geostrophic wind component (on the order of 19 kts, assuming air density of 1 kg/m³, dn of -442.4 km and Φ of 28.52°N). 3. RESULTS Preliminary Florida coastal shower data sets have been identified in two ways via the methods discussed above. The first (initial) data set identifies hourly coastal shower events based on the criteria specified above. The number of hourly ‘hits’ for the stations of focus are reported in Table 2 which indicates that the number of ‘identified’ events are sensitive to the station location (and to the selection criteria used to pinpoint an event). With hourly ‘hits’ it is true that more than one event can occur at each station for each day. However, coastal showers are defined on a daily scale. Hence, the data set will eventually be broken into event days with multiple hours in each. The second (alternative) data set identifies daily easterly flow events across the designated region of Florida in Figure 2 (i.e. length of n̂ ). This data set was produced using a linear regression of the along-coast pressure gradient to select days in which onshore flow is assumed strong enough to produce coastal showers. A pressure gradient equal to or greater than 3 mb over the area resulted in 94 identified events. As with the first data set, the number of potential events are sensitive to the selection criteria. An example regression plot of a costal shower event using Microsoft Excel is shown in Figure 3 from the 11:00 PM EST observation on 27 September 2000 which includes: the line of ‘best-fit’ and its corresponding regression 5 equation and the coefficient of determination (R²). The corresponding data and the regression parameters output from the statistics module are shown in Table 3 – the latter of which are in good agreement with the statistical parameters determined using Microsoft Excel. 4. COMMENTS Florida coastal showers have been identified using two methods. The robust methods above are believed to have produced two data sets which capture coastal showers based on: 1) nighttime showers generating precipitation and an onshore component to the wind and 2) a situation where a strong pressure gradient exists over the region with high pressure to the north and low pressure to the south. These data sets will be used to better understand the nature (both dynamic and thermodynamic) of the phenomenon, including their propensity for a nocturnal maximum, their inter-seasonal variability, and precipitation distribution and amount. 5. REFERENCES Mathbits, cited 2004. Correlation Coefficient. [Available online at http://mathbits.com/MathBits/TISection/Statistics2/correlation.htm.] Tyrrell, Sidney and Nicholson, James, cited 2004. Correlation Coefficient. [Available online at http://www.mis.coventry.ac.uk/~nhunt/regress/good4.html.] British Medical Journal (BMJ), cited 2004. Statistics at Square One. [Available online at http://bmj.bmjjournals.com/collections/statsbk/11.shtml.] 6 Table 1: Station IDs and corresponding names used for the initial data set of Integrated Surface Hourly Weather Observation data from NCDC. Station ID COF CRG DAB FLL FXE HST HWO JAX MFL MIA MLB NIP PBI PMP SFB SGJ TMB TTS VRB Station Name Patrick Air Force Base Craig Municipal Airport Daytona Beach International Airport Fort Lauderdale/Hollywood International Airport Fort Lauderdale Executive Airport Homestead Air Force Base Hollywood/North Perry Airport Jacksonville International Airport Miami Miami International Airport Melbourne International Airport Jacksonville Naval Air Station West Palm Beach International Airport Pompano Beach Airpark Sanford/Orlando International Airport St. Augustine Airport Miami Kendall-Tamiami Executive Airport NASA Shuttle Facility Vero Beach Municipal Airport Table 2: Station IDs and the number of hourly ‘hits’ for stations of focus for the selection criteria specified in section 3b. Station ID DAB VRB TTS MLB COF # of 'hits' 378 359 207 204 171 Table 3: (Left) Regression parameters output from the regression module in Perl. (Right) Data used to calculate the regression parameters using Microsoft Excel (and in regression module). Regression Module intercept 1017.0243 dp/dn 0.0071 R² 0.93 n 9 Excel Calculation Distance (km) Pressure (mb) -265.8 1015 -16 1017 -152.4 1016.1 0 1017.4 -22 1016.8 -442.4 1014.3 -185.7 1015.2 -218.3 1015.2 -333.6 1014.6 7 PresGrad Regression for 11:00 PM EST Observation on 09/27/2000 1018 y = 0.0071x + 1017 1017.5 2 R = 0.9302 Pressure (mb) 1017 1016.5 1016 1015.5 1015 1014.5 1014 1013.5 -500 -400 -300 -200 -100 0 Distance (km) Figure 3: Example regression plot of a costal shower event from the 11:00 PM EST observation on 27 September 2000 – showing the line of ‘best-fit’ and its corresponding regression equation and the coefficient of determination (R²). 8