An Exploration of Temperature Metrics for Further Developing the Heat-Health Weather Warning System in Hong Kong

Background . e current weather warning system aims to reduce mortality from heat and cold stress but still has room to be improved in terms of incorporating other temperature metrics. e aim of this study is to determine how extreme temperature aﬀects mortality in Hong Kong. Methods . An ecological study was used; daily weather data were subdivided into seven temperature metrics.Dailydetrendedmortalitydatawerestrati�edbydiseasegroupsandanalysedusingsevendiﬀerentmetricsfortemperature.etemperaturemetricswerethencompared. Results . A diurnal temperature range (DTR) of ≥ 8 ∘ C leading to an increase in median mortality of up to 16 % and a mean temperature change between neighbouring days of ≥ 4 ∘ C leading to an increase in median mortality of up to 6 % were the critical thresholds for excess mortality in Hong Kong. Conclusions . is study reveals that mean net eﬀective temperature, DTR, and temperature change between neighbouring days are eﬀective to predict excess mortality in Hong Kong.


Background
ere has been a growing interest in the impact of extreme heat and cold events on health globally. In subtropical Asia, this issue has been investigated in countries including Japan, China, and South Korea. Most heat-related mortality research in Hong Kong has been conducted in the last decade [1][2][3][4][5]. A summary of previous �ndings in Hong Kong is outlined in Table 1. A U-or J-shaped relationship is observed between temperature and mortality [5]. is is consistent with extant studies [6][7][8]. is paper will explore seven temperature metrics in relation to mortality in Hong Kong. In particular, the effects on mortality of diurnal temperature range (DTR) and temperature change between neighbouring days are not well understood in the Hong Kong literature. Understanding these effects will enhance the current Hong Kong heat-health weather warning system. Diurnal temperature range (DTR) is the difference between the highest and lowest temperature within a single day [9]. DTR is shown to in�uence heatrelated mortality in Hong Kong [10]. DTR has been decreasing from about 5.5 ∘ C in 1947 to about 4 ∘ C in 2002 [11]. Traditionally, a large DTR relieves heat stress: it assumes a cool night happens aer a hot day. is effect is believed to be important in Melbourne, Australia [12]. Conversely, a large DTR might increase mortality during hot days, particularly with cardiovascular diseases [9] and strokes [13]. A large DTR might also increase blood pressure in the elderly leading to cardiovascular disease [10].
e Hong Kong Observatory (HKO) has issued very hot and cold weather warnings since 2000; these warnings are derived from a weather stress index (WSI) [14]. In Hong Kong, WSI is calculated from net effective temperature (NET), which incorporates ambient temperature, relative humidity, and wind speed [14]. It is assumed that Hong Kong people generally adapt better to hot weather than cold weather. Indeed, cold-related mortality is higher than heat-related mortality in Hong Kong [2,14]. In addition, Li and Chan [14] observed that there is a skewed Ushaped relationship between WSI and mortality, in which the mortality is greater in winter months (November to March) than summer months (May to September . e data were partitioned into all-cause mortality, respiratory, and cardiovascular mortality according to the International Classi�cation of Diseases (ICD) 9 and 10 ( Table 2). RD refers to respiratory diseases and CVD refers to cardiovascular diseases.

Weather Data.
Daily time series, weather data were obtained from the Hong Kong Observatory (HKO) from 1999 to 2009. e temperature metrics chosen were the following: daily maximum ( max ) and daily minimum ( min ) temperature, daily mean apparent temperature (AT), daily net effective temperature (NET), diurnal temperature range (DTR), and temperature change between neighbouring days (temp change) in terms of mean temperature ( mean ), max and min . max and min were found on the HKO website database. mean is de�ned as the average of today's minimum temperature ( min ) and the previous day's maximum temperature ( max ), as this approach is demonstrated to predict excess mortality [12]. In order to determine the effects of season on mortality, this paper examines summer and winter separately: max and max NET for summer, min and min NET for winter, as well as DTR in summer and winter. Summer is de�ned as June to August, whereas winter is de�ned as December to February. �ox plots are produced to visualise the temperature-mortality relationships and identify temperature threshold for excess mortality. DTR = daily maximum temperature − daily minimum temperature on the same day (1) Temp change (See [19]) = today's mean temperature − previous day's mean temperature.
e equation of Apparent Temperature incorporates wind speed and moisture characteristics to calculate the human perceived air temperature in terms of discomfort. It is calculated as follows [16,17]: where is air temperature and is dew point temperature. �ind increases heat �ow a�er mean daily temperatures are above 34 ∘ C, wind-speed correction is not necessary when temperatures are below this [17]. e following is the equation for net effective temperature (NET): where is the ambient temperature (in ∘ C), the wind speed (in m s −1 ), and R H the relative humidity (in %).
is study uses seasonal decomposition to remove short and long-term trends from the data, such as population ageing and seasonal highs or lows. e multiplicative seasonal decomposition model is as follows (deaths = trend-cycle * seasonal factor * anomaly) [12,18]: where TCt is the "trend-cycle" component, St is the "seasonal" component, and It is the "irregular" or "random" component.

Effects of Temperature Change between Neighbouring Days on Mortality.
Temperature change between neighbouring days is de�ned as today's mean temperature minus yesterday's mean temperature [19]. It is a different temperature measure from DTR. is study's mean temperature is de�ned as the average of yesterday's maximum temperature and today's minimum temperature [12]. e mean of temperature change ( mean ) is 0 (SD = 1.31), and for temperature change ( max ), the mean is 0 (SD = 1.946). Figures 6 and 7 demonstrate that excess mortality (all-cause, RD) occurred when temperature change (mean temperature) exceeded 4.1 ∘ C. Excess mortality (all-cause, RD) also occurs when temperature change (maximum temperature) is greater than 5.6 ∘ C (see Figures 8 and 9). Apart from that, there appears to be some signi�cant reduction below baseline mortality when temperature change is greater than −4.0 ∘ C. e changes vary from approximately 2-9% below baseline mortality (see  Figures 6, 8, and 9). When the temperature cools off, the mortality rate drops. Table 4 summarises temperature-mortality relationships across different temperature metrics. e table was produced based on the graphical relationship in the box plots produced in SPSS [20]. reshold temperatures are identi�ed when the median mortality anomaly increases above baseline (mortality anomaly > 1). e number of days exceeding the thresholds refers to the number of days between 1999 and 2009. In short, mean NET, DTR, and temperature change between neighbouring days are demonstrated to be effective in predicting excess mortality.

Temperature Metrics and Mortality in Hong Kong.
Net effective temperature (NET) incorporates ambient temperature, wind speed, and relative humidity and is used in both hot and cold situations [14]. A high positive value indicates high heat load whereas a high negative value indicates substantial heat loss [14]. In order to establish a hot and cold weather warning system, the HKO uses the extreme values of NET as a gauge. It appears that the results support the current HKO's standard for issuing very hot weather warnings, with excess mortality occurring when daily maximum temperature exceeds 35 ∘ C (Table 4). ere is no cold excess mortality during winter on lag day 0 (Table 4). However, excess mortality occurs when mean daily apparent temperature falls below 8 ∘ C (Table 4). Generally, excess mortality occurs in the upper and lower 2.5% of NET as indicated by Li and Chan [14]. It is interesting to note that there are many nil values when max NET and min NET are examined. is indicates that no excess mortality occurs using these two temperature metrics. Based on the above results, the current weather warning system in Hong Kong might not accurately re�ect people's experience in urban area for two reasons. First, the system depends on max NET and min NET to issue very hot and cold weather warning. Second, the warning system does not incorporate diurnal temperature range (DTR) and temperature change between neighbouring days. A large DTR increases health risks (Figures 1, 2, 3, 4, and 5). ere were 37 days from 1999 to 2009 with DTR exceeding 8 ∘ C, associated with an increased median mortality ranging from 3-16%. During summer (June to August), a large DTR indicates a very hot day (above 95th percentile) and warm night on the same day. For example, on 19th July 2005, the maximum temperature was 35.4 ∘ C and the minimum temperature was 26.9 ∘ C. A lack of heat relief is associated with excess mortality because people do not have time to recover from heat exposure. is is important as hot nights are expected to increase in the future in Hong Kong [21]. e large DTR observed in Hong Kong is different from situations such as Melbourne when a large DTR indicates a cool change in the aernoon; in other words, a hot day and cool night represent a large DTR [12]. During winter (December to February), a large DTR might indicate a cool day with unusually cold night in Hong Kong. For instance, on 26th December, 2002, the maximum temperature was 16 ∘ C and the minimum temperature was 7.2 ∘ C. Exposure to such cold weather can trigger deaths caused by cardiovascular disease [22]. Temperature between neighbouring days is a temperature measure that is seldom investigated in the Hong Kong literature because rapid day-to-day temperature changes are perceived to be rare in Hong Kong [15]. A large temperature change indicates unstable weather systems and can increase the risk of mortality [19]. On the days with mean temperature change between neighbouring days greater than 4 ∘ C (13 days from 1999 to 2009), excess mortality (all-cause and RD) increases by at least 7% (see Figures 6 and 7). A large change in maximum temperature (>5.5 ∘ C) occurred 8 times between 1999 and 2009, this results in about a 3 to 4% increase in excess mortality for all-cause and RD, respectively (see Figures 8 and 9). In summary, the HKO should incorporate DTR and temperature change between neighbouring days in both hot and cold weather warnings. In re�ning the current warning system, it can prevent avoidable deaths and reduce the health risks of vulnerable populations.

Policy
Implications. e HKO currently uses a �xed threshold temperature when issuing very hot and cold weather warnings. Sometimes excess mortality occurs before the warning is issued. Excess mortality occurs when DTR exceeds 8 ∘ C (Figures 1 and 2) and temperature change (mean temperature) of neighbouring days exceeds 4.1 ∘ C ( Figure 5). erefore, it is necessary to incorporate the above two temperature measures in a heat-health weather warning system. With temperature forecasts available for a week in advance, the government can issue advisories a few days ahead and then issue warnings as the threat increases. Our paper is one of the few that examine the impact of cold weather on mortality in Hong Kong. Understanding the cold threshold temperature will improve the current cold weather warning system. However, it is suggested that the elderly might have a lower heat threshold or a higher cold threshold for excess mortality [23]. It is because they have impaired thermoregulation, and the use of drugs can also affect normal homeostasis. In future research, a cohort study design with focus groups and questionnaires can help to target the more vulnerable elderly population.
ere are several limitations of this study. e �ndings of this study might not necessarily be generalized to cities from temperate regions or even other subtropical regions, due to the varying demography and socioeconomic factors. Additionally, an ecological study design is susceptible to ecological fallacy, so it is not able to explain individuallevel responses from aggregate data. It is acknowledged that air pollution is an important confounder, but it was not included in the study due to the scope of this project. In terms of mortality data quality, there might be misclassi�cation of diseases. Furthermore, air temperature fails to consider indoor temperature which is in�uenced by the housing design and the use of air conditioning and heating. It is beyond the scope of the current study design to analyse the differences in exposure based on the time spent either indoors or outdoors. Finally, preliminary �ndings of lag day 1-3 do not reveal any signi�cant excess mortality. However, with longer lag days excess mortality might result.

Conclusions
is study aimed to determine the impact of extreme temperature on mortality in Hong Kong. It identi�ed various temperature thresholds for excess mortality. Speci�cally, mean net effective temperature (NET), diurnal temperature range (DTR), and temperature change (mean temperature) are shown to be effective temperature metrics in predicting excess mortality in Hong Kong. To the best of our knowledge, temperature change between neighbouring days is a temperature metric that has not been studied in Hong Kong before. e Hong Kong Observatory should consider incorporating DTR and temperature change in the heat-health weather warning system, rather than using a �xed weather stress index that is only based on NET. is study demonstrates that the current heat threshold for very hot weather warnings (daily maximum temperature ≥33 ∘ C) predict excess summer mortality in Hong Kong. In comparison, the threshold for very cold weather warning (daily minimum temperature ≤12 ∘ C) does not appear to predict excess cold mortality on the same day as exposure. Moreover, winter mortality is higher than summer mortality but is seldom studied in Hong Kong. A greater emphasis on cold-related mortality research can re�ne the cold weather warning systems in Hong Kong and reduce winter mortality.

Con�ict of �nterests
e authors declare no potential con�ict of interests with respect to the research, authorship, and/or publication of this paper.
Disclosure is research received no speci�c grant from any funding agency in the public, commercial, or not-for-pro�t sectors.