2. Data and Methods

2.1. Population
Population data were obtained from the demographic database of the fifth and sixth national population census of China, carried out in 2000 and 2010, respectively (China Statistics/Beijing Info 2000, 2010) [17,18]. We used data from 345 Chinese cities in our analyses. These data included the number of individuals aged 85 years and older (the oldest available age group; 85+), and the number of individuals aged 65 years and older (65+). We used the 85+/65+ ratio as a city-level measure of longevity, as it is less sensitive to factors such as migration and birth rates which can obscure other measures of longevity.
It is acknowledged that the current population structure includes 65+ years old people who might have moved prior to where they lived in 2010. Since no data are available to take into account the migration of 65+ years old people prior to 2010, the assumption is that only the climate and local food of their present location affects their health status, although it is understood that where people lived in the past plays a role in their current health status. The geographical distribution of 85+/65+ ratios in Chinese cities in 2000 and 2010 are illustrated in Figure 1 and Figure 2.
As the population census data in south Xinjiang was not accurate [19] because investigation found the centenarians and elders count to be inaccurate as there was large gap between self-reported age and verified (actual) age [20,21]. Therefore, we deleted the data for south Xinjiang in 2000.
Two kinds of factors, potentially related to longevity, were collected for analysis: environmental and nutritional variables.

2.2. Environmental Factors
We collected the standard deviation of the monthly mean temperature to exam potential relationships between extreme temperatures and longevity. Additional variables of interest included average humidity and altitude. We acquired annual average precipitation data for each city from the public meteorological service center website of the Chinese meteorological administration (www.weather.com.cn) and (www.tianqi.com). Yearly mean temperature is illustrated in Figure 3. The standard deviation of monthly mean temperature is illustrated in Figure 4. Humidity was calculated according to a classic equation formulated by de Martonne in 1926 (Equation (1)):I = P/(T + 10)(1)
In this equation, I represents humidity; P is annual average precipitation (mm); T is annual average temperature (Centigrade). Each city’s humidity is illustrated in Figure 5 with higher values indicating higher humidity.
We downloaded the digital elevation model (DEM) of China (resolution: 1 km × 1 km) from the data cloud of the Chinese Academy of Sciences (http://www.csdb.cn/), and subsequently calculated the average altitude of each city using zonal statistics analysis tools and ArcGIS software with the DEM data and map of each city. Average city altitudes are illustrated in Figure 6.
Environmental pollution is currently a serious issue in China and has been for the past 10 to 15 years. It is well known that pollution from industry, automobiles and power plants that burn coal exerts cumulative and long-term effects on human health. Most of the older people (85+) included in this analysis lived most of their lives in rural areas prior to the current pollution problems. It is also problematic to associate sources of pollution with older people at the geographic scale used in this analysis. For these reasons, no pollution variables are included in the analysis.

2.3. Nutritional Factors
Studies found that the Se content in soil has a significant positive correlation with longevity. In contrast, barium (Ba) and nickel (Ni) have significant negative correlations longevity, while distributions of cadmium (Cd), cobalt (Co), chromium (Cr), copper (Cu), manganese (Mn), vanadium (V), zinc (Zn), lithium (Li), and iron (Fe) showed no significant correlations with longevity [16]. Of all the trace elements, Se appears most closely linked with longevity [22,23,24,25]. Food contributes a greater proportion of daily elemental intake than drinking water [26], and the Se content in food is positively correlates with the Se content in soil [27]. Because of this, we selected the Se content in soil to represent the relationship between longevity and trace elements. Soil Se can be considered as total Se (T-Se) and water-soluble (WS) Se (WS-Se). Soil WS-Se is a better indicator of environmental effects than T-Se [28]. In this study, both T-Se and WS-Se content of soil were considered. We collected background concentrations of T-Se and WS-Se from China’s Soil Environment Background Concentration Research (Ministry of Environmental Protection of the People’s Republic of China, China National Environmental Monitoring Centre 1990) [29] and other studies related to soil environmental background values in China [11,30]. We calculated soil T-Se and WS-Se in each city using the union and statistic tools in ArcGIS software. T-Se and WS-Se, for each city, is illustrated in Figure 7 and Figure 8.
We also noted per capita meat production (Figure 9), per capita freshwater-fish production (Figure 10), and per capita seawater fish production (Figure 11 (Shanghai and Zhoushan are regarded as one fishing ground)) to examine potential relationships between omega-3 intake and longevity. These data were collected from the statistical bulletins of each city, and the 1990 (Fishery Administration Bureau of Ministry of Agriculture 1990) and 2016 (Fishery Administration Bureau of Ministry of Agriculture 2016) China Fisheries Statistics Yearbook [31,32]. Water food production in 1990 and 2015 are illustrated in Table 1.
Table 1 shows that fishing production increased slowly from 1990 to 2015, and seawater fishing and freshwater fishing increased 2.39 and 2.89 times, respectively. Because of the limitations of natural water food resources, aquaculture production increased sharply due to recent developments in aquaculture and biological technologies, seawater aquaculture and freshwater aquaculture increased 11.58 and 6.86 times, respectively. Individuals who were aged 85+ years had mainly eaten fishing productions rather than aquaculture productions throughout their lives, particularly at a young age. Consequently, seawater aquaculture production was not included in the seawater food production data in this research.
In the past, especially 20 years ago when China’s highway network was not built, and food freezing technology was not developed, sea fish were seldom transported inland. Because sea fish will quickly die and deteriorate after leaving seawater, they were almost exclusively eaten by coastal-area residents. Even today, most sea fish in China are consumed by coastal area residents. There is no accurate per capita sea fish consumption data for Chinese cities, so we launched an investigation of the frequency of sea fish consumption, per month, for the residents of 139 cities in China, as a representative sample of the whole country. It would be too difficult to obtain this data from all cities, so the 139 selected cities account for 40.3% of all cities in China, and are distributed throughout every part of the country, as illustrated in Figure 12. Frequency of sea fish consumption is related to distance from the sea. In coastal cities, residents eat sea fish 3–10 times per month, while in inland areas most consume sea fish fewer than one time per month. We evaluated the consumption of sea fish for each city according the city’s per capita sea fish consumption and seawater food production in coastal regions (Equation (2)): (2) Ci=∑i=1nFi×Ti×Pi∑i=1n(Ti×Pi)
In this equation, Ci is per capita sea fish consumption in city i, Fi is sea fish production in city i, Ti is the monthly frequency of sea fish consumption in city i, and Pi is the population of city i.

2.4. Methods
Kolmogorov-Smirnov tests were performed to determine whether the distribution for each of the factors above was normal. For those factors that were non-normal, logarithmic transformations were used to normalize the distributions. Spearman’s rank correlation was used to evaluate the relationship between each factor and the longevity ratio. A stepwise multiple linear regression (MLR) analysis and geographically weighted regression (GWR) were used to identify the factors significantly associated with the longevity ratio.