4. Discussion and Interpretation
Reviewing Table 1 combined with an analysis of non-routine products, associated Doppler and ground-truth signatures, plus input from NWSFO staff lead to the following discussion which may be grouped into four general categories:
1) Generally, since Doppler became operable, there has been a significant increase in non-routine products issued by this office. This is especially noteworthy because January-March 1997 was very dry presenting few opportunities to issue non-routine products. Likewise, El Nino rendered the 1997 hurricane season unusually inactive across the northeast Caribbean.
Specifically, during DY1, the NWSFO San Juan registered 638 non-routine products compared to only 358 during PD. This represents an additional 280 products or an approximate 76 percent increase. Much of the difference occurred during two key months, August and November, when the local area experienced a significant increase in very active weather.
Compared with conventional radar, Doppler's most significant contribution was a markedly clearer identification of short-fuse warning signatures. Not surprisingly, the annual difference between short-fuse warnings, and especially thunderstorm warnings was very significant (i.e. SMW, SVR, with around eleven and eight fold increases respectively during DY1). This substantiates that Doppler provided better defined severe weather signatures and a greater variety of products than with the previous and conventional WSR-74S radar. This translated to a higher level of forecaster confidence in issuing short-fuse warnings. Many of the SMWs were for severe thunderstorms, previously a largely ignored product as any SMWs were mainly for sighted waterspouts and not for thunderstorms. Many of the SVRs were for severe thunderstorms based on mesocyclones, merging cells, VAD wind profiles and other signatures previously undetected by conventional radar. These clear-cut signatures left little forecaster discretion in product issuance accounting for such significant increases.
2) Although FFWs increased 71 percent from PD, this was comparatively small compared to other short-fuse warnings. Some of the FFWs were based strictly on Doppler rainfall estimates, whose quantitative accuracy is still being assessed. It should be noted these estimates were often significantly higher than conventional USGS/ALERT or observer reports and may have influenced forecasters to "pull the trigger" when they would have resisted given more traditional rainfall guidance. In particular, the one and three hour precipitation totals significantly influenced forecaster watch and warning decisions. This was particularly evident across the western third of Puerto Rico where initial investigations suggest a 2:1 ratio of Doppler estimates versus conventional amounts. Conversely, the ratio lowers to approximately 1:1 ratio elsewhere. However, because of uncertainties in the accuracy of Doppler estimates and discrepancies with Doppler versus conventional rainfall amounts, forecasters had greater latitude or discretion in issuing FFWs compared with the more clear-cut thunderstorm warnings. This may account for the comparatively modest increases in FFWs and remains a focus of NWSFO San Juan investigation.
Conversely, relatively little difference was noted between longer-term statements and advisories, For example, the MWS, FFA and FFS recorded only 09, 60 and 60 percent increases respectively while non-forecast products like PNS and RER actually declined. The greatest increases were registered with coastal marine products, like CFW, CFS and HSA but because Doppler signatures have little effect on such products, these were not addressed here.
The increase in non-routine products was significantly greater for the east half as compared to the west half of Puerto Rico. This was likely due to the elimination of blockage or ground clutter seen on the WSR-74S that greatly hindered an accurate interpretation, particularly in the San Juan metropolitan area and southeast sections as well as over the south half of the U.S. Virgin Islands.
Data suggest that forecast verification may be, in part, influenced by the location of the event. Given the same rainfall amounts, either in intensity or time, the likelihood of flash flooding across the northwest quarter of Puerto Rico is noticeably less than with other sections. This is attributed to the geology of the area where the soils can more readily absorb heavy rain, and to land use practices where considerably less construction and other alteration of natural surfaces occur along the relatively few flood prone areas. Additionally, civil defense, police and spotter reports seemingly decrease in proportion to the distance from the San Juan area. Across the more sparsely populated west and western highlands, noticeably fewer contacts are available. In addition, heavy rainfall and strong thunderstorm events are considerably more common across the west half of Puerto Rico. Therefore, such scenarios would be less likely to create any undue concern across this area decreasing the likelihood that individuals or local officials would feel a need to contact the NWSFO. Again, this skews any conclusions about verification and the degree of over forecasting.
d. Additional Factors
The significant increases in non-routine products were not entirely a byproduct of Doppler. They were also attributed to improved guidance available during DY1. This included, the advent of N-AWIPS and HP workstations including GARP and other higher resolution satellite data, SWAPS and other marine data. This guidance, independently and in conjunction with Doppler, revolutionized the model output and forecasting process and significantly increased forecaster confidence level.
Doppler signatures are in part, a function of the reflectivity Z-R relationship. Most research establishing these relationships was conducted in the United States employing mid-latitude criteria. Applying these standards to an untested tropical environment like Puerto Rico raises many questions and additional research is necessary before optimum relationships are established. Initially, Z=300R1.4 was used before permission was granted to switch to the more applicable tropical equation Z=200R1.6 . Certainly the former and likely the latter Z-R relationships have generated values and helped determine alarm criteria that may still not be representative of specific weather scenarios or seasons. This would also adversely affect the relationship between precipitation versus range, culminating in a skewed the forecaster decision making process.
Given that all forecasters-in-charge work a comparable number of shifts, there was a great disparity in the number of products each issued, ranging from 154 to 48 with a five person average of around 80 products. (The disparity between aviation forecasters who issued fewer non-routine products is, not surprisingly, smaller). This may infer that some forecasters are more aggressive than others when diagnosing and reacting to severe weather situations that require the issuance of non-routine products. These aggressive forecasters place more weight on Doppler signatures. However, since some forecasters delegate product issuance with greater frequency than others, the results presented must be viewed with caution.
Based on ground-truth, other timely guidance and post-event analysis, some of the severe weather/flash flood signatures and alarms were false echoes. This led to the issuance of some unnecessary products; consequently, the false alarm rate was noticeably higher than in previous years.
Despite the concern for false alarms, it is clear that during DY1 forecasters have issued more non-routine products, and especially more short-fuse warnings than during PD. Doppler, as the primary guidance tool, is largely responsible for this change. This may, in part, be attributed to a more veteran staff and the advent of a dedicated Doppler focal point during non-routine events. However, great variation in interpretation and product issuance exists between the respective NWSFO San Juan forecasters.
The large increase compared to PD implies a degree of over forecasting, however it remains very difficult to quantify that degree. Doppler is still in its first year of uninterrupted operation, being commissioned only since July 1997. No objective data yet exists to irrefutably determine what weight to assign to each signature and by extension what algorithms, if any, need to be adjusted. Additionally, there has been a relative increase in product issuance since the commissioning. This may, in part, be attributed to a lack of radar down-time but also to a self-imposed pressure on each forecaster given the legal liabilities involved in an official radar.
The degree of over forecasting may also be influenced by the lack of on-site familiarity that some forecasters have with the topography of Puerto Rico and the U.S. Virgin Islands. Site visits and an increased familiarity with the topography of the islands should improve their knowledge of local terrain; thus, decreasing the degree of over forecasting.
Because forecasters are driven by NWS regulations, as well as guidance obtained in WSR-88D school, they issue non-routine products based on stated criteria. This criteria may not be entirely accurate when applied to the northeast Caribbean. Additionally, the spotter network is neither sufficiently dense nor reliable to provide, confirm or refute the existence of severe weather or flooding. For example, lack of confirmation of an event should not necessarily be interpreted as a busted product. Until we can refine the Doppler algorithms and verification programs, including the Z-R relationship, the degree of over forecasting will remain unresolved.
DY1 marked a significant increase in new technology at the NWSFO San Juan. This was highlighted by the commissioning of Doppler radar which introduced a wide array of previously unavailable guidance. San Juan Forecasters enthusiastically responded with a significant increase in non-routine products, and especially with short-fuse thunderstorm warnings. As Doppler algorithms become more refined, more ground-truth becomes available and forecasters acquire additional knowledge of the local area and the vagaries of this system, non-routine products issued will best reflect the given weather scenarios.
6. Future Trends
Initial mid-latitude Doppler studies have demonstrated that significant over forecasting occurred across the NWS (personal conversations). This is not surprising as forecasters suddenly have access to a large amount of data previously unavailable with conventional radar. However, as initial data is reviewed, algorithms adjusted, and forecasters better understand what weight to apply to each signature, the number of products have generally leveled off and even decreased during Doppler year two.
Current NWSFO San Juan evaluations indicate the need for another year of data to accurately access both spatial and temporal forecaster interaction and reaction to their radar. Further study will also determine if the trend of increased product issuance will continue or level off as occurs in most other offices. Surprisingly, October 1997 to March 1998 data from Doppler year two reveal a persistent and noticeable increase in issuance compared to DY1. (446 products versus 265 during the same six month period or about a 70 percent increase). Part, but certainly not all, of this change may be attributed to an approximately 20 percent increase in rainfall. This trend forms the basis for further investigation. Because of the dearth of studies conducted in tropical environments, any conclusions gleaned from this office are especially valuable region wide and can be used to determine optimum algorithms necessary for forecasters to take correct watch and warning decisions.
Acknowledgments. I express my appreciation to the staff of the NWSFO San Juan for all their feedback on this paper. I am especially indebted to Shawn Bennett, SOO for his editorial input and overall support in publishing this paper and to Israel Matos, MIC for valuable comments on a variety of related topics. Forecasters Rachel Gross provided excellent feedback and technical support and Edward Tirado provided valuable suggestions on staff perception of Doppler that helped shape this presentation.
Alberty, R., 1993: An Update on the NEXRAD Program and Examples of Recent WSR-88D Successes. 26th International Conference on Radar Meteorology. Norman, Oklahoma. American Meteorological Society. 5-7.
Bennett, S., 1997: An Overview of Hurricane Hortense and its Aftermath. 22nd Conference of Hurricanes and Tropical Meteorology. Ft Collins, Colorado. American Meteorological Society.
Black, M, and P. Dodge, 1993: Time Lapse Radar Images of Hurricanes Hugo (1989) and Andrew (1992). 26th International Conference on Radar Meteorology. Norman, Oklahoma. American Meteorological Society. 97-99.
Buller, M.T., 1995: Performance of the Goodland WSR-88D During a Heavy Rainfall Event in Northwest Kansas.27th Conference on Radar Meteorology. Vail, Colorado. American Meteorological Society. 392-393.
Gould, K., C. Herbster and J. Korotky, 1996: WSR-88D, GOES-8 and MM5 Mesoscale Model Observations of the Florida Panhandle Sea Breeze Circulation Under Different Synoptic Flow Patterns. American Meteorological Society. 76th Annual Meeting. Atlanta, Georgia
Hodanish, S, S. Spratt and D. Sharp, 1997: WSR-88D Characteristics of Tornado-Producing Convective Cells associated with Tropical Cyclones. 22nd Conference on Hurricanes and Tropical Meteorology. Ft Collins, Colorado. American Meteorological Society.
Johnson, J.T., M.D. Ellis and D. Green, 1995: Operational testing of Enhanced WSR-88D Algorithms and Display Concepts in National Weather Service Offices. 27th Conference on Radar Meteorology. Vail, Colorado. American Meteorological Society. 170-172.
Lee, R.L., M.A Magsig., G.J. Stumpf and E.D. Mitchell, 1998: Performance of Radar Circulation Detection Algorithms: Texas Tornado 27 May 1997. 16th Conference on Weather Analysis and Forecasting. Phoenix, Arizona, American Meteorology Society. 240-242.
Sharp, D., S.Spratt and S.Hodanish, 1997: Doppler Radar Observations of a Spectrum of Outer Spiral Rainband Mesocyclones Associated with Tropical Cyclones. 22nd Conference on Hurricanes and Tropical Meteorology. Ft Collins, Colorado. American Meteorological Society.
Smalley, D.J. and F.I. Harris. 1993: Wind Field Characteristics of Hurricane Bob (1991). 26th International Conference on Radar Meteorology. Norman Oklahoma. American Meteorological Society. 77-79
Snell, W.L. E.W. McCaul Jr.. 1993: Doppler Signatures of Tornados Spawned by Hurricane Andrew near Montgomery, Alabama. 26th International Conference on Radar Meteorology. Norman, Oklahoma. American Meteorological Society. 80-82.
Tabata, A, and H Sakakibara, 1993: Structure of the Inner Core Region of Typhoon 9119 (Mireille) Observed by a Doppler Radar. 27th Conference on Radar Meteorology. Vail, Colorado. American Meteorological Society. 88-90.
Wood, L., 1997: Using the Tropical Z-R Relation to Improve Radar Precipitation Estimates during a Heavy Rain Event in Southeast Texas. 28th International Conference on Radar Meteorology. Austin, Texas. American Meteorological Society.
Wright, J.E. and S. Bennett, 1997: WSR-88D Observations of Bow Echos Embedded in Outer Rainbands of Hurricane Bertha. 22nd Conference on Hurricanes and Tropical Meteorology. Ft Collins, Colorado. American Meteorological Society.
Wright, J.E. and S. Bennett, 1997: Meso-vortices Observed in the Eye of Hurricane Bertha by the Puerto Rico WSR-88D. 28th International Conference on Radar Meteorology. Austin, Texas. American Meteorological Society.
* Corresponding author address: