Chinese blue days: a novel index and spatio-temporal variations

Daily average meteorological observations, including precipitation, wind speed at 10 m above the ground, temperature, relative humidity, cloud cover, low cloud cover and visibility (observed four times per day at 2:00, 8:00, 14:00, 20:00 BJT) during 1980–2014 from 385 ground stations are used to identify the blue days in China. The dataset was obtained from the National Meteorological Information Center (http://data.cma.cn/), which has been under strict quality control (Husar et al 2000, An et al 2013, Lin et al 2014, Guo et al 2016). The observed monthly sea-level pressure (SLP) data, with a resolution of 2.5° × 2.5°, from the National Centers for Environmental Prediction Reanalysis 2 (Kanamitsu et al 2002) is used to provide atmospheric circulation background for the blue days. In addition, hourly data of the air quality index (AQI) in 2013 and 2014 from the China National Environmental Monitoring Center (Zheng et al 2014; http://www.cnemc.cn/) and the daily weather phenomena records in 2011–2014 in 32 provinces (see supplementary section 2 is available online at stacks.iop.org/ERL/14/074026/mmedia) are used to examine whether the CBDI defined in this study can accurately represent the blue days variations in China. Finally, a Community Emissions Data System for historical emissions (Hoesly et al 2018; http://www.globalchange.umd.edu/ceds/) is used to study the anthropogenic contribution.

Zoom In Zoom Out Reset image size

2.2.1. CBDI index definition and validation

There are two criteria to determine CBDI. First, the selected day should have a blue sky, with no rain, which represents a sunny day. Second, it should have clean air, which represents good air quality. The details for the definition are shown as follows:

• (1)  
  First, we identify a sunny day as a day with daily precipitation ≤ 0.1 mm and low cloud cover ≤ 75th percentile. The criteria are selected based on a statistical analysis of the daily weather phenomenon record from 2011–2014 (see supplementary, section 2) in 32 provinces. After selecting all sunny days (recorded as sunny and no rain) in this dataset and comparing their meteorological factors on that day, the results shown in figure 1 prove that the excursion of the total cloud cover is too large to be used in the definition, and the outliers of low cloud cover belonging to frequently cloudy and rainy areas (like Beihai, Guangzhou) show that a percentile index can better represent change in all stations rather than an absolute index. In all, the AX30 (low cloud cover ≤ 30%), TX70 (low cloud cover ≤ 70th percentile), and TX75 (low cloud cover ≤ 75th percentile) account for 93%, 95.5%, and 96.5% of low cloud cover on recorded sunny days, respectively. According to visibility ranks for distance in table 1 and the median visibility in figure 1, we choose 'great' visibility (≥20 km) as part of our definition. Taking cloudy and rainy areas into consideration, we also choose the lower quartiles visibility (≥15 km) in figure 1 as a comparison. Then, we obtain 6 alternative CBDI indexes (shown in table 2).
• (2)  
  Early studies find the relative humidity (RH) affects the visibility, indicative of an error in relation to human observation (Deng et al 2011, Bai et al 2014, Chen et al 2017, Liu et al 2017). Thus, dry visibility (equivalent visibility in dry conditions) is used to reduce the effect of RH. RH values greater than 40% on non-rainy days were converted by formula (1) (Rosenfeld et al 2007, Xu et al 2017):

  $$\begin{eqnarray}&&\begin{array}{l}\displaystyle \frac{{\rm{VIS}}}{{\rm{VIS}}\left({\rm{dry}}\right)}\,=\,\left\{\begin{array}{c}0.26+0.4285\times \,\mathrm{log}\,10(100-{\rm{R}}{\rm{H}}),\,40 \% \lt {\rm{R}}{\rm{H}}.428\,\\ 0.26,\,\,\,\,\,\,{\rm{R}}{\rm{H}}\gt 99 \% \,\,\,\,\,\,\,\end{array}\right.,\end{array}\end{eqnarray} \tag{ 1 }$$

  where RH is in percent and VIS is the observed visibility (km).
• (3)  
  Next, daily average AQI values of 8 stations (four in North China and four in South China) are calculated. The values in March and June in 2013 and September and December in 2014 are selected to test whether the CBDI can correspond to grade II air quality (AQI ≤100) and which definition can explain the dual effects of meteorology and pollution. We conduct a precision analysis by comparing with the forecast score (table 3) used in CMA (2007). The accuracy of prediction is calculated using formula (2)–(4):

$$\begin{eqnarray}&&{\rm{TS}}=\displaystyle \frac{{\rm{N}}{\rm{A}}+{\rm{N}}{\rm{D}}}{{\rm{N}}{\rm{A}}+{\rm{N}}{\rm{B}}+{\rm{N}}{\rm{C}}+{\rm{N}}{\rm{D}}}\times 100 \% \end{eqnarray} \tag{ 2 }$$

the rate of missing reports (PO) and the empty reports (FAR) are:

$$\begin{eqnarray}&&{\rm{PO}}=\displaystyle \frac{{\rm{N}}{\rm{C}}}{{\rm{N}}{\rm{A}}+{\rm{N}}{\rm{C}}+{\rm{N}}{\rm{D}}}\times 100 \% ,\end{eqnarray} \tag{ 3 }$$

$$\begin{eqnarray}&&{\rm{FAR}}=\displaystyle \frac{{\rm{N}}{\rm{B}}}{{\rm{N}}{\rm{A}}+{\rm{N}}{\rm{B}}+{\rm{N}}{\rm{D}}}\times 100 \% .\end{eqnarray} \tag{ 4 }$$

Table 1.  Visibility ranges in each level.

Ranks	Range (km)	Grade
0	<0.05	Worst grade
1	0.05–2	Bad grade
2	2–10	Medium grade
3	10–20	Good grade
4	>20	Great grade

Table 2.  The alternative CBDI index definitions.

	AX30	TX70	TX75
Visibility>=20 km	Blue_1	Low70_1	Low75_1
Visibility>=15 km	Blue_2	Low70_2	Low75_2

Table 3.  Test classification list.

Definition
Observation	CBDI	NOT CBDI
AQI<=100	NA	NC
AQI > 100	NB	ND

The results (table 4) show that all 6 indexes fit into the category of good air quality in summer and autumn. After using the percentile index, the TS improves largely in Guangzhou and Kunming, almost to 75% (not shown). Furthermore, among all 6 indexes, the low75_2 index has the highest PC (65.18%) and lowest PO (22.79%). Additionally, the FAR is 19.29%. We must pot out that the frequent dust activities and consistently sufficient moisture lead to low TS in some cities, such as Xining and Hangzhou, making the average TS not exceed 70%.

Table 4.  The results of the prediction accuracy.

	Blue_1	Blue_2	Low70_1	Low70_2	Low75_1	Low75_2
TC	53.91	56.07	59.81	63.78	60.89	65.18
PO	39.86	33.65	33.39	24.76	31.97	22.79
FAR	16.13	21.65	14.58	19.27	14.70	19.29

Although there are some subjective parts of our index due to the lack of long-time actual observation records for sunny days and AQI, it is still the first time we try to combine pollution data with meteorology data to define a reasonable blue day, which demonstrates good agreement with reality. Therefore, finally, we put out the following CBDI definition: no rain (daily precipitation ≤ 0.1 mm), low cloud cover ≤ TX75, and the observed visibility in 14:00 ≥ 15 km.

2.2.2. Statistical methods

Figure 2(b) illustrates the spatial distribution of climatological annual mean CBD in China during 1980–2014. The climatological CBD is relatively large in northwest China but small in southeast China with three low-value centers in the southeast Sichuan Basin, southern Hebei, and northeast Guangxi. The largest value, above 170 d/y (day per year), is located in Yunnan Province, while the lowest value is recorded in Hebei Province, with only 15.06 CBD a year, likely due to its high anthropogenic emissions induced by heavy industries (Wu et al 2013, Li et al 2016, Kong et al 2017).

Zoom In Zoom Out Reset image size

Among 385 stations in China, approximately 42% of stations show a prominent, increasing trend and approximately 23.4% of stations show a decreasing CBD trend (figure 2(c)). The averaged CBD for all stations in China is increasing at a rate of 1.6 d/10 y, above the 95% confidence level. Most stations in western China (west of 107 °E) are dominated by an increasing trend exceeding 4 d/10 y. A distinct decreasing CBD trend is found in most stations in the Yangtze–Huai River Valley (YHRV), southern North China Plain, and part of Northeast China, with a rapid decreasing trend below −2 d/10 y. Moreover, we use the revised Mann–Kendal method to test whether the annual CBD in China has experienced interdecadal change during 1980–2014 (figure 2(d)). The result shows that there is a significant interdecadal shift in CBD near the mid-1990s.

To analyze the regional characteristics of CBD variations, two typical subregions, which share the most significant variations, are selected based on topographic distribution and results of EOF (figure S2). One subregion is the YHRV (30°N–35°N, 115°E–122°E), including 35 stations, and the other is southwestern China (SWC; 26°N–34°N, 98°E–107°E), including 40 stations. Their scopes are shown in figure 2(c) in black.

The time series of average CBD during 1980–2014 in the two areas are shown in figure 3(a). In YHRV, the averaged value of CBD changes from 148.44 d/y in 1980–1996 to 134.41 d/y in 1997–2014, suggesting that CBD in this region likely has experienced a notable interdecadal change around 1996. In SWC, the annual CBD increases gradually from 1980 to 2014 at the rate of 9.5 d/10 y, which is likely related to the severe and sustained droughts in SWC during these years (Niu et al 2014, Wang et al 2015, Wang et al 2018). With more rainless days and lower relative humidity, the low visibility events caused by meteorological conditions decrease obviously, which may increase CBD.

Zoom In Zoom Out Reset image size

Figure 3(b) shows the month-year evolutions of CBD in YHRC. The highest occurrence of CBD happens in autumn, with an occurrence rate of 44.92%, and it shows an increasing trend in the 1980s (8.41 d/y) and then turns to a decreasing trend after that (−5.81 d/y). CBD in winter has the lowest occurrence (36.12%), and it increases from 24.85% to 41.33% in the 1990s and then reduces to 20.70% by 2014; the average trend is −3.04 d/y. CBD in summer slightly increases in the 1980s and then maintains a decreasing trend of −3.74 d/y. In spring, CBD in YHRV is stable before the 2000s; it starts to increase at the rate of 9.29 d/y and then decreases rapidly after 2010 (−28.8 d/y). Overall, the annual average CBD in YHRV exhibits a decreasing trend of −7.5 d/10 y.

For SWC (figure 3(c)), where it is perennially wet, a steadily increasing trend is observed. CBD mainly occurs in winter and autumn, with an average occurrence of 51.02% and 37.98%. In winter it exhibits an increasing change at 10.69 d/y in the 1980s, 7.795 d/y in the 1990s, and 3.62 d/y from 2000 to 2014; the annual average trend is 3.23 d/y. In autumn and spring, CBD grows stably in 2.82 d/y and 4.29 d/y, and it increases rapidly at the rate of more than 20 d/y after 2000. However, in summer, it exhibits a decadal trend, that is, it increases in the 1980s (6.85 d/y) and after the 2000s (5.10 d/y) but decreases in the 1990s (−6.06 d/y).

Four climate factors, including RH, wind speed, windless days, and gale days are chosen to analyze the correlations with CBD. The summary of the Pearson correlation results is shown in table 5. In the annual mean, correlation coefficients with all these factors pass the 95% confidence level at national scales, showing a close relationship with CBD in China. The correlations of wind speed and gale days with CBD are −0.43 and −0.41, respectively. The negative correlation is somewhat counterintuitive, as large wind speed is generally considered to be favorable for blue sky (Fu and Dan 2014). This means that under adverse wind conditions, some other factors play a role, such as wind direction in a specific region. RH is negatively correlated with CBD at the national scale, with a correlation coefficient of −0.40. Reducing RH may decrease the growth of hygroscopic condensation nucleus (CNN), which is a key for the formation of aerosol particles, thereby increasing the possibility of sunny days. In summer and winter, CBD is weakly correlated with wind conditions but is obviously negatively correlated with the RH.

In YHRV, CBD is significantly and positively correlated with wind speed and number of gale days, and negatively correlated with windless days all year round, with correlation coefficients of 0.5, 0.39, and −0.52. This means that weakened wind conditions does lead to decreasing CBD. However, RH is not associated with the change of CBD annually. In summer, CBD in YHRV is strongly correlated with all factors. Whereas in winter, it is only negatively correlated with RH with a correlation coefficient of −0.59.

In SWC, there is a drop in RH yearly and seasonally (summer and winter), with high correlation coefficients of −0.78, −0.81, and −0.87, respectively. Thus, increasing RH leads to increasing CBD. Wind conditions show a completely inverse relationship between the two regions at the annual scale, which agrees with the results of Cheng et al (2013). They found an 'east plus west minus' distribution between visibility and wind speed in China, and obviously, visibility is positively correlated with CBD. In summer and winter, wind speed shows almost no correlation with CBD, and windless days and gale days show weak correlation coefficients. Considering the special basin topography and mountainous terrain, CBD in SWC should be studied further.

Previous studies (Chen and Wang 2015, Jian et al 2016, Cai et al 2017) show that the change of haze days in China is significantly modulated by atmospheric circulation. Does atmospheric circulation contribute to the change of blue days in China? To address the question, we apply a dynamic adjustment method developed by Wallace et al (2012) to isolate the role of atmospheric circulation on the trend in CBD according to the following procedure.

Following Hu et al (2018), SLP is seen as an indicator of atmospheric circulation. First, a PLS regression is performed by correlating the time series of seasonally averaged CBD T(n) at each station with standardized SLP in the domain of East Asia (20°N–60°N, 70°E–150°E) to obtain a one-station regression map, and the regression map is used as an SLP predictor for the CBD trend. Then, we project the standardized SLP field to the correlation pattern, weighting each station by the cosine of the latitude, and obtain a score index S(n), which shows the relative score with which the predictor is expressed. Next, we use the least squares method to compute the residual T₁(n) and SLP trends by removing the linear component associated with the S(n). The above steps are repeated for the residual T(n) and SLP until we find the variance in residual T(n) explained by the third predictors is too small to be ignored (not shown). Thus, we only use two SLP trend predictors to dynamically adjust the CBD trend at each station, and the residual CBD T₂(n) is considered to be the adjusted CBD trend. The procedure is the same as that in Wallace et al (2012) and Hu et al (2018). We also apply 500 hPa geopotential height as a predictor and find similar results, which are not discussed here. Finally, the raw CBD is analyzed for comparison. The results are shown in figure 4.

Zoom In Zoom Out Reset image size

Both the spatial patterns of the raw CBD trend in summer (figure 4(a)) and winter (figure 4(b)) resemble the annual CBD trend pattern (figure 1(b)). In both winter and summer, most stations in West China show a positive trend, while considerable stations in East China show notable negative trends. After dynamic adjustment, the residual CBD trend is weaker than the raw one nationwide in summer (figure 4(c)). Some stations in northern Yunnan and the Pearl River Delta even reveal an opposing trend. The national average trend is 0.39 d/35 y and 0.28 d/35 y for raw and adjusted CBD, respectively. In winter (figure 4(d)), the residual CBD trend also becomes weak in most stations after dynamic adjustment. The results suggest that the observed CBD trends in summer and winter are partly caused by atmospheric circulation.

Figure 5 provides an overview of the historical change of the raw and adjusted CBD change in SWC and YHRV in summer and winter during 1980–2014. In SWC, the average contributions of residual parts for the total CBD trend in summer (figure 5(a)) and winter (figure 5(c)) are 14.25% and 60.38%, indicating the dominant role for atmospheric circulation in summer. As shown in figure 5(b), the raw summer CBD in YHRV features a notable decreasing trend during 1980–2014. After adjustment, the summer residual CBD in YHRV still decreases although its magnitude is weaker than the raw trend. Additionally, it is clear that the residual CBD in YHRV (figures 5(b) and (d)) still exhibits notable long-term changes, suggesting that other factors such as air pollution may contribute to the change of CBD. Therefore, in figure 6, we depict the contemporaneous variations in anthropogenic emissions of SO₂, NO_X, and VOC in YHRV, which are the main precursors of fine particles. We find that the downtrend in the residual CBD mainly appears after 1990, which is coincident with the rapid increase in emissions of SO₂, NO_X, and VOC after 1990 (figure 6(a)), suggesting that air pollution may also contribute to the decrease in summer CBD in YHRV. Similar results are obtained in winter (figures 5(d) and (b)). Overall, the average contributions of residual CBD to total CBD trend in summer and winter are 43.62% and 35.84% separately, indicating that atmospheric circulation contributes to a larger part of CBD variations in YHRV. Generally, PLS is effective in reducing atmospheric circulation-induced variability and revealing the role of anthropogenic emissions in YHRV.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size