The Synoptic Climatology Lab, with support from the U.S. Environmental Protection Agency Climate Change Division, developed the Philadelphia Hot Weather-Health Watch/Warning System (PWWS) prior to the summer of 1995.

First implemented by the City of Philadelphia Department of Public Health (CDPDH) in the summer of 1995, the WWS system uses National Weather Service forecast model data to successfully identify high-risk weather and climate conditions (i.e., conditions associated with unusually high numbers of deaths) up to two days prior to their occurence. Once a high-risk airmass is identified, the Center for Climatic Research coordinates with the Department of Public Health to issue watches, warnings and alerts via various media resources. In the case of an alert, special measures are taken by the CPDPH to mitigate the effects of the heat stress - including outreach programs to those who are most vulnerable (see articles below).  A flowchart of PWWS methodology may be viewed here.  An article on the development and application of this system follows.

The City Health Department has estimated that approximately 300 heat-related deaths were averted in 1995 as a result of this system.  The 1995 run was both a success and a learning experience, and the improved PWWS was utilized again in summer of 1996.

The Philadelphia Hot Weather-Health Watch/Warning System:
Development and Application, Summer 1995

Laurence S. Kalkstein
Center for Climatic Research
Dept. of Geography
University of Delaware
Newark, DE 19702

Paul F. Jamason
Center for Climatic Research
Dept. of Geography
University of Delaware
Newark, DE 19702

J. Scott Greene
Oklahoma Climatological Survey
University of Oklahoma
Norman, OK 73019

November, 1995

Figures and Tables not currently available for web distribution.

Abstract

Last summer, Philadelphia, PA instituted a new Hot Weather-Health Watch/Warning System (PHWHWWS), to alert the city's residents of potentially oppressive weather situations which could negatively affect health. In addition, the system was used by the Philadelphia Department of Public Health for guidance in the implementation of mitigation procedures during dangerous weather. The system is based on a synoptic climatological procedure which identifies "oppressive" air masses historically associated with increased human mortality. Air mass occurrence can be predicted up to 48 hours in advance with use of model output statistics (MOS) guidance forecast data. The development and statistical basis of the system are discussed, and an analysis of the procedure's ability to forecast weather situations associated with elevated mortality counts is presented. The PHWHWWS, through greater public awareness of excessive heat conditions, may have played an important role in reducing Philadelphia's total heat-related deaths during the summer of 1995.

1. Introduction

Interest in the impact of weather on human health has increased dramatically in recent years, especially in light of potential climate changes (IPCC, 1995; WHO/WMO/UNEP, in press). There is well-documented evidence that hot weather, in particular, contributes to increased morbidity and mortality in large urban areas (Pennsylvania Emergency Management Council, 1994), and numerous cities have established watch/warning systems to help the local health departments prepare for dangerous conditions. Most of the systems in place are based on a series of National Weather Service (NWS) guidelines, and rely on the computation of the "heat index" (better known as apparent temperature), which combines the impact of temperature and relative humidity (NOAA, 1994). Specifically, an excessive heat warning is issued by the NWS when daytime heat index values are expected to reach 40.5oC (105oF) or above for more than three hours a day for two consecutive days, or when the daytime heat index is expected to exceed 46oC (115oF) for any length of time.

Watch/warning systems based on the heat index are deficient for a number of reasons. First, they assume that people respond to a combination of two meteorological variables: temperature and relative humidity. It is quite clear from other studies that a number of other meteorological variables play a significant role. For example, cloud cover has been shown to be a statistically significant predictor of elevated human mortality during hot weather, as clear skies add considerably to the heat load of dwellings, especially those in impoverished urban areas (Kalkstein and Davis, 1989). In addition, wind speed is a desiccating factor, and adds heat load to the body when temperatures are excessive (Steadman, 1979). Second, the present NWS system does not take into account the negative impact of several consecutive days of oppressive weather (no changes are made beyond two consecutive days), nor does it account for the fact that heat waves earlier in the summer season seem to create more of a health danger than those late in the season (Kalkstein, 1993). Third, the heat index values used to define dangerous conditions have not been proven as estimators of either morbidity or mortality. Fourth, no estimates of morbidity or mortality can be derived from excessive heat warnings, as there is no empirical basis for the establishment of criteria. Finally, these same values are used at numerous locations without regard to human adaptation or acclimatization; a heat index of 40.5oC will have a much different impact on the population in Boston than in Dallas (Kalkstein and Valimont, 1986).

Persons respond to the total effect of all weather variables interacting simultaneously on the body, rather than to individual meteorological elements. Therefore, an appropriate means to evaluate weather/health relationships is through the identification of high risk or "oppressive" air masses that, when present, could negatively impact human health. Studies funded by the U.S. EPA and the Southern Regional Climate Center indicate very strong relationships between particular excessively hot, oppressive air masses and increased human mortality (Scheraga and Sussman, in press). Based on results from these and other studies, the Philadelphia Hot Weather-Health Watch/Warning System (PHWHWWS) was developed. Through the identification of such oppressive air masses, the PHWHWWS provides information to the public and appropriate health agencies that weather situations which could be potentially hazardous are predicted or imminent.

A direct association has been noted between oppressively hot weather and increased human mortality in a number of studies (Smith and Tirpak, 1989; Kalkstein, 1993; Kunst et al., 1993; Touloumi et al., 1994). In particular, mortality increases sharply above specific weather thresholds, especially maximum temperature, in many domestic and international cities (Kalkstein and Davis, 1989; IPCC, 1995). Through the use of a synoptic climatological approach, Kalkstein (1991) found that one particular summer air mass in St. Louis possessed the highest mean mortality and occurred on many of the highest mortality days, even though it was climatologically infrequent. Although not all days within this air mass possessed high mortality totals, it was possible to determine which meteorological (e.g., temperature, cloud cover) and non-meteorological (e.g., consecutive day sequence, within-season timing of occurrence) parameters were associated with the highest daily mortality totals.

City location, the magnitude of the urban heat island effect, and housing conditions influence the magnitude of the negative health impacts associated with oppressive summer weather. Cities in the northeastern and midwestern U.S. demonstrate the strongest weather/mortality relationships, and this may be due in part to the irregularity of oppressive summer air masses. Such situations occur much more frequently in the southeastern U.S., permitting behavioral and possibly physiological acclimatization to these conditions. Housing type, especially in the inner cities, may also play a role. In cities such as Philadelphia and St. Louis, the prevalence of multi-family structures characterized by red brick walls, tar roofs, and poor air flow contributes to inferior ventilation and increased solar load on the building when compared to similar inner city housing in the Southeast and the West. It is easy to understand why the number of heat-related deaths in a city such as Chicago can top 500 in a few days during an intense heat wave (Kalkstein, 1995). Thus, it is these cities that require the establishment of a comprehensive weather-health watch/warning system, to permit city health departments to take mitigating action and to alert the public that dangerous weather is predicted. 2. Development of the Watch/Warning System Meteorological Data and Synoptic Category Development

An automated air-mass-based climatological index was developed for the PHWHWWS to categorize each day based on its meteorological character using a synoptic climatological approach. The synoptic procedure has been designed to group days that are meteorologically homogeneous. This is accomplished by defining each day in terms of six readily available meteorological elements (air temperature, dew point temperature, total cloud cover, sea-level pressure, wind speed, and wind direction). The elements are measured four times daily (0100, 0700, 1300, and 1900 local standard time), and the developed 24 variables represent the basis for categorization.

The PHWHWWS is based on the temporal synoptic index (TSI) (Kalkstein et al., 1987), which uses principal components analysis (PCA), a technique that rewrites the original 24-variable data matrix into a new set of components that are linearly independent and ordered by the amount of the variance they explain (Daultrey, 1976). Component loadings are calculated, which express the correlation between the original variables and the newly formed components. Each day is then expressed by its particular set of component scores, which are weighted summed values whose magnitudes are dependent on the weather observation for each day and the principal component loading. Thus, days with similar meteorological conditions will tend to exhibit proximate component scores. Refer to Kalkstein et al., 1987 for a detailed discussion of PCA in the development of the TSI.

A clustering procedure is then used to group those days with similar component scores into meteorologically homogeneous groups, which represent the air mass types. There are numerous clustering methods available, but previous studies have shown that an efficient clustering procedure in the development of a synoptic climatological index is the average linkage method (Kalkstein et al., 1987; Yarnal, 1993). Once the groups have been determined, average meteorological characteristics are determined for the 24 meteorological variables for all days within each particular group (air mass). Weather map evaluation is also performed to describe the general characteristics and similarities of each TSI group (Table 1). Mortality Data

The National Center for Health Statistics (NCHS) produces a detailed mortality database that contains a record for every person who died in the United States from 1964 to the present. The data include cause, place and date of death, age, and race (NCHS, 1978). These values were extracted for the Philadelphia Standard Metropolitan Statistical Area (SMSA) from 1964-1966, 1973-1976, 1978, and 1980-1988, years for which the date of death is included. A tabulation of total deaths is made for each day through the period of record.

All mortality data are adjusted to account for changes in the total population of the Philadelphia SMSA during the period of record. A direct standardization procedure is used, and a mortality trend line is constructed based on mean daily mortality for each year of record. Mortality is expressed as a deviation around this inter-annual trend line (Kalkstein, 1991; Lilienfield and Lilienfield, 1980). Relationship Between Synoptic Categories and Mortality

The mean daily mortality for each synoptic category, along with the standard deviation, is determined to ascertain whether particular categories exhibited distinctively high or low mortality values. Potential lag times are accounted for by evaluating the daily synoptic category on the day of the deaths, as well as 1, 2, and 3 days before the day of the deaths. Daily mortality is also sorted from highest to lowest during the period of record to determine whether certain synoptic categories are prevalent during the highest and lowest mortality days in Philadelphia. For many cities, it is apparent that one or two hot air masses possess a much higher mean mortality than the others, and these "oppressive" air masses contain an inordinately high percentage of days with the greatest mortality totals (Kalkstein, 1991). For Philadelphia, this offensive air mass is identified as Category 3 (maritime tropical, oppressive; Tables 1 and 2). This is the hottest air mass in Philadelphia during summer, and is also characterized by the highest dewpoint temperature, southwesterly winds, and partly cloudy conditions. Category 3 possesses the highest mean mortality for a lag of 0 days.

While this maritime tropical air mass has a daily mean mortality well above the overall mean, not all days within this air mass type possess elevated mortality totals; the standard deviation of daily mortality is particularly high for this air mass (Table 2). Thus, the PHWHWWS must not only identify the oppressive air mass, it must also identify which days within this air mass will have elevated mortality. Using a standard stepwise multiple regression analysis, it is possible to determine which factors within the oppressive air mass contribute to elevated mortality (Table 3). In Philadelphia, the factors contributing to elevated daily mortality when the oppressive air mass is present include:

the number of consecutive days the air mass has been present, maximum temperature, the time of season (e.g., whether the oppressive air mass occurs early or late within the summer season). The resulting algorithm satisfies the Box and Wetz (1973) criteria for being a statistically robust predictor, and can be used to estimate mortality for any given day. Format of the PHWHWWS

Previously, health warnings were issued by the Philadelphia Health Commissioner if the local National Weather Service issued an excessive heat warning based on the heat index. Unlike the NWS system, the PHWHWWS is based on the identification of oppressive air masses that are actually associated with elevated mortality in summer.

Using NWS forecast data for upcoming days, it is possible to predict the arrival of an oppressive air mass up to 2 days before it arrives. The Nested Grid Model (NGM) forecast issued by the NWS is used to predict the arrival (or continuance) of oppressive air masses 2 days in advance. The NGM is a 16-layer model with 80 km resolution. It generates 48-hour forecasts twice a day and is used for model output statistics (MOS) guidance. These MOS values include the standard meteorological variables necessary to classify each day into one of the pre-existing synoptic categories. Since the categories, and their respective means, have already been pre-determined, the post-TSI classification of the forecast data requires a separate statistical technique. The use of PCA and average-linkage clustering is restricted to identifying initially the air masses at a given locale. Since the goal here is to classify each forecast day into one of the synoptic categories listed in Table 1, the appropriate tool is discriminant function analysis (Klecka, 1980). Discriminant analysis is similar to the use of multiple regression. For each air mass type, a discriminant function is developed based on the means of the 24 variables. Then, for each forecast day, a discriminant score is calculated for each of the synoptic categories. The day is classified into the category possessing the highest score, which represents the most similar synoptic situation.

The accuracy of the forecast data, and the performance of the discriminant function analysis, were verified through a "backcasting" technique, in which previously-issued MOS forecast data were applied. The same procedure outlined above was used; however, results from the discriminant analysis could then be compared to days already defined by the TSI, which used actual meteorological data for the same days. When 24-hour forecast data were used, the "backcasting" technique correctly identified the oppressive synoptic category on 32 out of 36 days (88.8%) in 1988; this rate decreased to 71.4% (25/35) when 48-hour forecast data were utilized. Of course, the reduction in accuracy when using the 48-hour forecast data is no reflection on the backcasting technique, but rather reflects the veracity of MOS forecasting ability.

The framework of the PHWHWWS is depicted in Figure 1, and consists of a three-tiered system which produces a health watch, health alert,or health warning. The system is coordinated with the local Philadelphia region National Weather Service office in Mount Holly, NJ. The information is transmitted to the Philadelphia Department of Public Health from the University of Delaware's Center for Climatic Research, and after consultation with the Center and the local National Weather Service office, the Health Commissioner makes the final decision on the issuance of health advisories.

The system is initiated with the analysis of MOS data, and air mass type is predicted for a 3-day period ("today", "tomorrow", and "day after tomorrow") using the discriminant function analysis described earlier. If the procedure forecasts the arrival of the oppressive air mass for the day after tomorrow, a health watch is issued by the Health Commissioner up to 48 hours prior to its predicted arrival. If the forecast arrival of the oppressive air mass is tomorrow, the Health Commissioner issues a health alert up to 24 hours prior to air mass arrival. Since not all oppressive air mass days produce elevated mortality, the next level of this system involves identification of those days predicted to be associated with high daily mortality. This is accomplished by using the algorithm developed from the evaluation of mortality variance within oppressive air mass days (Table 3). A health warning is issued by the Health Commissioner either the afternoon before, or the morning of, the forecast occurrence of an offensive air mass only if elevated mortality is predicted by the algorithm. In addition, the local NWS office must agree to issue a simultaneous excessive heat warning. Depending upon the magnitude of the excess deaths predicted, one of three levels of health warning is issued (Figure 1). For Philadelphia, a level one warning is issued if 1 to 4 heat-related deaths are predicted by the algorithm. A level two warning is issued if 5 to 14 deaths are predicted, and a level three warning is issued if 15 or more deaths are predicted.

A series of guidelines have been developed by the City of Philadelphia Department of Public Health, which indicate steps to be taken under each of the watch/warning system scenarios. These include:

activation of hot lines for people to obtain information during heat emergencies, initiation of the "buddy system", in which volunteers within the city make daily visits to an elderly person who may need assistance during hot weather, contacting the Philadelphia Corporation for Aging, which provides special services for elderly persons in need, contacting local utility companies, to make certain that electrical service is not terminated for any individual during the heat emergency, contacting "block leaders", to make certain that there are no problems within the neighborhood, contacting the Philadelphia Water Department, to make certain that adequate water supplies are maintained during the emergency, contacting radio and television stations, as well as newspapers, so the public is notified rapidly of emergency conditions. In addition, the Department of Public Health broadcasts advice on how to avoid heat-related illness during the oppressive weather; advising nursing homes in the area that an emergency situation exists, possibly opening air conditioned shelters, for those who don't have access to cooler environments. Does the System Work?

The meteorological summer (June 1 to August 31) of 1995 was the hottest on record for the city of Philadelphia, with a daily average maximum temperature near 32oC (90oF). Temperatures reached or exceeded 32oC on 22 days in July, and minimum temperatures averaged 22.5oC (72.5oF) for the month. The PHWHWWS was instituted on July 12, 1995 and continued through September 21, 1995. During this period, the oppressive synoptic category occurred on 16 days (Table 4), with most occurrences confined to the period between July 13 and August 14. Two particularly extreme heat episodes stand out: July 13 through July 15, and August 2 through August 5. Of the 72 heat-related deaths reported by the Philadelphia medical examiner for the summer of 1995, 32 were associated with these particular heat episodes.

Fifteen of the 16 days within the oppressive synoptic category during the period were predicted to be associated with excess mortality based on the algorithm. Thus, the Center for Climatic Research (CCR) suggested to the National Weather Service and the Philadelphia Department of Public Health that warnings should be issued on these 15 days (Table 4). According to system criteria, level two and level three warnings should have been issued on twelve and three days, respectively (no level one warnings were predicted). The maximum excess mortality value was forecast to occur on July 15th (23 deaths), the third consecutive day of the oppressive air mass. A comparison of the daily numbers of predicted excess mortality versus actual heat-related deaths as reported by the medical examiner indicates that the system significantly overestimated the number of deaths that were to occur. It should be noted, however, that the system predicted deaths for the entire SMSA, while heat-related deaths represent those that occurred in the city only. In most cases, the system predicted heat-related deaths at times when they actually did occur. System overestimation appeared to be greatest near the end of the summer season.

Actual warnings were not issued by the Philadelphia Department of Public Health on every day the PHWHWWS suggested that warnings be issued. Of the 15 warning days suggested by the system, actual warnings were issued on 9 days. An actual warning can only be issued with National Weather Service concurrence, and such concurrence was not offered by the NWS on 6 days. Thus, only health alerts were issued on these days; however, it should be noted that significant numbers of heat-related deaths occurred during these alerts.

Two notable observations are apparent relating to the application of the PHWHWWS. First, it appears that the NWS should have concurred with the issuance of a health warning on those 6 days. Five of those 6 days were associated with excess mortality, yet the Philadelphia Department of Public Health did not institute many of their mitigating measures on those days due to lack of concurrence. The NWS did not concur on those days as they did not meet the standard heat advisory criteria (based on the heat index) which had traditionally been used by NOAA. It is suggested here that local NWS forecasters should be more flexible in their use of established criteria to issue heat advisories and warnings. Such latitude is clearly granted to forecasters in the NWS Regional Operations Manual, where it is stated that, "... at the forecaster's discretion, excessive heat warnings may be issued at a lower (heat index) threshold...to account for possible differences between the official temperatures reported at exposed observation sites and those within the more congested city areas" (NOAA, 1994). Second, it is noteworthy that the PHWHWWS overpredicted mortality by a considerable margin during the hot weather of early and mid-August (Table 4). Overprediction was relatively minimal earlier in the summer season. Two possibilities are suggested. First, many susceptible individuals died during the earlier heat episodes, leaving a smaller number of vulnerable individuals later in the season. This notion of "mortality displacement" is well documented in the literature (e.g. Kalkstein, 1993). However, the predictive mortality algorithm used in the PHWHWWS allows for mortality displacement as it contains a "time of season" variable, which proved to be statistically significant and inversely related to mortality (i.e. as the season proceeds, less mortality is expected given similar weather conditions). Second, it is possible that the system was effective in saving lives as the season progressed, as evidenced by the increasing overprediction of the algorithm through the period. Virtually all of the Philadelphia media broadcast the health warnings and alerts as they were issued by the Department of Public Health. Thus, public awareness of vulnerabilities to health problems related to oppressive weather was probably higher in Philadelphia than ever before.

Evaluation of the PHWHWWS will continue, and the Climate Change Division of the U.S. Environmental Protection Agency is developing a comprehensive plan to determine its potential effectiveness during the summer of 1995. In addition, a NOAA scientific team has evaluated the response of various cities (including Philadelphia) to cope with health problems during the hot summer of 1995, and their report suggests continued testing of the PHWHWWS, and expansion to additional cities. The system will be initiated again for the summer of 1996 in Philadelphia, and similar systems are planned to be on line for Chicago and Atlanta.

Acknowledgements

The development and implementation of the PHWHWWS was possible through the support of the Climate Change Division, US EPA (CR-824404), and the Southern Regional Climate Center (R-108542), and we are greatly appreciative. We are most grateful to many individuals at the Philadelphia Department of Public Health for permitting us to implement the system, and having the confidence in our work to use the system as the basis for their health mitigation measures. We thank Ms. Estelle Richman, Health Commissioner, Dr. Allen Chandler, Medical Director, and Dr. Lawrence Robinson, Deputy Health Commissioner, for their support. We give special thanks to Mr. Jerry Libby, Planning Supervisor, for the number of hours he contributed to the development and implementation of the system, and for his constant attention to detail. David Rodenhuis, Robert Bermowitz, and Jim Laver of the National Weather Service's Climate Prediction Center deserve special thanks for their suggestions and efforts to heighten NOAA awareness about the system. Finally, we thank numerous students and staff at the Center for Climatic Research who worked long hours to keep the system operational during this very hot summer. 

References

Box, G.E.P., and J. Wetz, 1973: Criteria for judging adequacy of estimation by an approximating response function, University of Wisconsin Statistics Department Technical Report,

#9, Madison, WI.

Daultrey, S., 1976: Principal components analysis. Concepts and Techniques in Modern

Geography, Institute of British Geographers, 8, 1-51.

Intergovernmental Panel on Climate Change (IPCC) Impacts Assessment, 1995: Assessing the

Health Impacts of Climate Change, Geneva.

Kalkstein, L.S., 1991: A new approach to evaluate the impact of climate on human mortality. Environmental Health Perspectives, 96, 145-150.

_____, 1993: Health and climate change - direct impacts in cities. The Lancet, 342, 1397-1399.

_____, 1995: Lessons from a very hot summer. The Lancet, 346(8979), 857-859.

Kalkstein, L.S., and R.E. Davis, 1989: Weather and human mortality: an evaluation of demographic and inter-regional responses in the United States. Annals of the Association of American Geographers, 79(1), 44-64.

Kalkstein, L.S., G. Tan, and J.A. Skindlov, 1987: An evaluation of three clustering procedures for use in synoptic climatological classification. Journal of Climate and Applied Meteorology, 26, 717-730.

Kalkstein, L.S., and K.M. Valimont, 1986: An evaluation of summer discomfort in the United States using a relative climatological index. Bulletin of the American Meteorological Society, 67(7), 842-848.

Klecka, W.R., 1980: Discriminant Analysis, Sage University Pres, Newbury Park, CA, 71 pp.

Kunst, A.E., C.W.N. Looman, and J.P. Mackenbach, 1993: Air pollution, lagged effects of temperature and mortality: The Netherlands 1979-87. Journal of Epidemiology and Community Health, 47, 121-126.

Lilienfield, A.M., and D.E. Lilienfield, 1980: Foundations of Epidemiology, Oxford University Press, New York.

National Center for Health Statistics, 1978: Standardized Micro-Data Tape Transcripts,

Department of Health, Education, and Welfare, Washington, DC.

National Oceanic and Atmospheric Administration (NOAA), 1994: Eastern Region National Weather Service, Bohemia, New York, Regional Operations Manual Letter.

Pennsylvania Emergency Management Council, 1994: Heat Wave Preparedness Task Force Status Report, 55 pp.

Scheraga, J.S., and F. Sussman. Preliminary Assessment of the Benefits to the U.S. of Avoiding or Adapting to Climate Change. Washington, DC: EPA Climate Change Division, in press.

Smith, J.B., and D. Tirpak, Eds., 1989: The Potential Effects of Global Climate change on the United States, Appendix G: Health. Report to Congress, U.S. Environmental Protection Agency, Office of Policy, Planning and Evaluation, Office of Research and Development, Washington, DC, 105 pp.

Steadman, R.G., 1979: The assessment of sultriness: Part I: A temperature-humidity index based on human physiology and clothing science. Journal of Applied Meteorology, 18, 861-873.

Touloumi, G., S.J. Pocock, K. Katsouyanni, and D. Trichopoulos, 1994: Short-term effects of air pollution on daily mortality in Athens: a time-series analysis. International Journal of Epidemiology, 23(5), 957-967.

WHO/WMO/UNEP: Climate and Health: The Potential Impacts of Climate Change.

WHO/WMO/UNEP, Geneva, Switzerland, in press.

Yarnal, B., 1993: Synoptic Climatology in Environmental Analysis, London, Belhaven Press, pp. 195