Statistical test
Pollutant measurements and meteorological parameters were analysed by Multiple Correspondence Analysis (MCA) using SPAD
® software as previously described [
15].
MCA is a descriptive multivariate (multifactorial) method used to visualize a large number of variables on a same graph and to estimate the statistical relationship between variables. A descriptive multivariate analysis is particularly appropriate to assemble complex parameters with respect to multiple variables.
MCA was performed using the whole table of data with the 4,392 observations in rows (24 hours per day × 30 or 31 days per month × 6 months) and 8 variables in columns. These qualitative variables were sorted into categories which were noted in each column for each hour of a day of a month (statistical unit):
- Temperatures (T), expressed in degrees Celsius, were sorted into 7 categories: T1 = [- 4; 6[; T2 = [6; 11[; T3 = [11; 16[; T4 = [16; 21[; T5 = [21; 26[; T6 = [26; 31[; T7 = [31; 41[
- Relative Humidity (RH) values, expressed as a percentage of maximum moisture in the air, were sorted into 4 categories: RH1 = [0; 40[; RH2 = [40; 60[; RH3 = [60; 80[; RH4 = [80; 100[
- Sunshine (S), expressed in minutes of sunshine per hour, was sorted into 7 categories: S1 = 0; S2 = [1; 10]; S3 = [11; 20]; S4 = [21; 30]; S5 = [31; 40]; S6 = [41; 50]; S7 = [51; 60]
- Wind Directions were the different points of the compass card: North (N); North East (NE); East (E); South East (SE); South (S); South West (SW); West (W); North West (NW); plus Indetectable Wind (IW)
- Wind Speeds (WS), expressed in ms-1, were ranged from 0 to 13 ms-1 and classified using the Beaufort scale: WS1 = Calm wind : 0 to 3.9 ms-1; WS2 = Moderate wind : 4 to 7.9 ms-1; WS3 = Rather strong wind : 8 to 13.9 ms-1
- Atmospheric pressures adjusted to sea level and expressed in hectopascal were sorted into 3 categories: P1 = Depression : [990; 1005[; P2 = Normal atmospheric pressure : [1005; 1020[; P3 = Anticyclone : [1020; 1035[
- 1- and 8-hour ozone concentrations, both expressed in μg/m3, were sorted into 6 categories: 1h-O31 and 8h-O31 = [0; 50[; 1h-O32 and 8h-O32 = [50; 80[; 1h-O33 and 8h-O33 = [80; 100[; 1h-O34 and 8h-O34 = [100; 120[; 1h-O35 and 8h-O35 = [120; 150[; 1h-O36 and 8h-O36 = [150; 260[.
The 6 active variables of the analysis were meteorological parameters (temperature, relative humidity, sunshine, atmospheric pressure, wind speed and direction) and the 2 illustrative variables were pollution levels (1- and 8-hour ozone concentrations).
MCA allows to construct a space in which 2 different categories of active variables are close together if they simultaneously appear several times in the whole table of data. For example, if temperatures comprise between 31 and 40°C (T7) are frequently associated with relative humidity values comprise between 0 and 40 % (RH1), these 2 categories will be close to each other in the space. The different categories of active variables are positioned in a multidimensional space, the number of dimensions of the space being proportional to the number of categories of actives variables. To observe the distance between categories on a 2-dimensional graph, the cloud of categories is projected on a plane of 2 orthogonal factors, F1 and F2. These factors were chosen as the most representative of the global variance of the cloud. In a second step, the supplementary (illustrative) variables are placed on the graph according to their proximity with the different categories of active variables.
We checked that each category of each variable was sufficiently represented inside the variable to apply the statistical tests, i.e. above √3n (n = 4392).
The relationship between meteorological parameters and pollutant measurements was evaluated using the DEMOD (DEscription of MODalities) procedure of SPAD which statistically characterizes each category of 1- and 8-hour ozone concentrations in relation to each climate parameter category. A chi square was calculated between the proportion of a given category of a given variable inside another category of another variable and the proportion of the given category of the given variable inside the totality of observations. A P value of 0.05 or less was considered statistically significant.
Relationship between ozone concentration and temperature
For a more detailed analysis of the relationship between ozone concentration and temperature, we constructed a contingency table cross-classifying ozone concentration and temperature. Maximum values of ozone in each city were used.
Firstly, 1- and 8-hour ozone concentrations were cross-classified taking the value of 110 μg/m3/8h and the public information level of 180 μg/m3/h. In this way, the following situations were considered:
- Ozone concentrations below 110 μg/m3/8h corresponding to background pollution.
- Ozone concentrations above 110 μg/m3/8h but never reaching the 180 μg/m3/h level during the 8 hours. This situation is recognized as having potentially harmful effects on health, but the general public is non-informed in France.
- Ozone concentrations above 180 μg/m3/h, recognized in France as a pollution peak; the general public is informed.
For temperature, three different situations were considered:
- Temperatures below 26°C with no significant consequence on health.
- Temperatures above 26°C, demonstrated to be harmful to the health of patients with respiratory problems [
16].
- Temperatures above 30°C, defined as a heat wave by meteorologists in France [
16].