2.1 Description of observations

2.1.1 Surface data

The topography of China is presented in Fig. 1a, and pink rectangles outline the four regions of interest (ROI) for the current study: northeast China (NEC), the Yangtze River Delta (YRD), the Pearl River Delta (PRD), and the North China Plain (NCP). The environmental monitoring station locations are indicated with red dots in Fig. 1b. They routinely measure PM with diameters ≤2.5 µm (PM_2.5), which are released to the public in real time with relatively high credibility (Liang et al., 2016). The locations of meteorological stations are indicated in Fig. 1c (data source: http://data.cma.cn/en, last access: 11 January 2018). The wind speed and wind direction at these stations are quality-controlled and archived by the China Meteorological Administration. We also utilized the MPL data and Sun-photometer data in Beijing, a megacity located within the NCP. The MPL located in Beijing was operated continuously by Peking University (39.99^∘ N, 116.31^∘ E) from March 2016 to December 2017, with a temporal resolution of 15 s and a vertical resolution of 15 m. The near-surface blind zone for lidar is around 150 m. Background subtraction, saturation, after-pulse, overlap, and range corrections are applied to raw MPL data (He et al., 2008; Yang et al., 2013). In this study, we use Level 1.5 aerosol optical depth (AOD) at 550 nm from the Beijing RADI (40^∘ N, 116.38^∘ E) Aerosol Robotic Network (AERONET) site, with hourly time resolution. As observations from multiple sources and platforms are used, we present descriptions of these observations in Table 1.

Figure 1(a) Topography of China. The black rectangles outline the five regions of interest: northeast China (NEC): 40.5–50.2^∘ N, 120.1–135^∘ E; North China Plain (NCP): 33.8–40.3^∘ N, 114.1–120.8^∘ E; Pearl River Delta (PRD): 22.2–24^∘ N, 111.9–115.4^∘ E; and Yangtze River Delta (YRD): 27.9–33.5^∘ N, 116.5–122.7^∘ E. Locations of (b) environmental stations and (c) meteorological stations. (d) Blue lines indicate CALIOP daytime orbits (in ascending node). Ground-based lidar and Sun photometer are deployed in Beijing (red triangle).

Download

2.2 Retrieving PBLHs

2.2.1 PBLH derived from MPL

MPL data from Beijing were used to retrieve the PBLH for this study. Multiple methods have been developed for retrieving the PBLH from MPL measurements, such as signal threshold (Melfi et al., 1985), maximum of the signal variance (Hooper and Eloranta, 1986), minimum of the signal profile derivative (Flamant et al., 1997), and wavelet transform (Cohn and Angevine, 2000; Davis et al., 2000). To derive the PBLH from MPL data, we implement a well-established method developed by Yang et al. (2013) and adopted in multiple studies (Lin et al., 2016; Su et al., 2017a, b). This method is tested to be suitable for processing long-term lidar data. Initially, the first derivative of a Gaussian filter with a wavelet dilation of 60 m is applied to smooth the vertical profile of MPL signals and to produce the gradient profile. The aerosol stratification structure is indicated by multiple valleys and peaks in the gradient profile. To exclude misidentified elevated aerosol layers above the PBL, the first significant peak in the gradient profile (if one exists) is considered the upper limit in searching for the PBL top. Then, the height of the deepest valley in the gradient profile is attributed to the PBLH; discontinuous or false results caused by clouds are subsequently eliminated manually. Moreover, we further estimated the shot noise (σ) induced by background light and dark current for each profile, and then added threshold values of ±3σ to the identified peaks and valleys of this profile to reduce the impact of noise. Figure S1 in the Supplement presents an example of the PBLH retrievals derived from MPL backscatter over Beijing. To validate MPL-derived PBLH, the values are compared with summertime radiosonde PBLH results retrieved using the Richardson number method (e.g., Vogelezang and Holtslag, 1996) from potential temperature profiles acquired at the Beijing station (39.80^∘ N, 116.47^∘ E) at 14:00 CST. Figure S2a shows good agreement ($R=\sim \mathrm{0.7}$) between MPL- and radiosonde-derived PBLHs over Beijing.

Figure 2The schematic diagram of retrieving the PBLH from CALIPSO.

Download

2.3 Statistical analysis methods

As a widely used parameter, the Pearson correlation coefficient derived from linear regression analysis measures the degree to which the data fit a linear relationship. This approach is less meaningful for characterizing nonlinear relationships. We find that the PBLH and PM_2.5 are correlated but not linearly correlated under most conditions. We found by trial and error that an inverse function ($f\left(x\right)=A/x+B$) fits our data well. Following Winship and Radbill (1994), we derived the fitting parameters (A and B) and the coefficient of determination (R²) of the PBLH–PM relationship using this inverse fitting function. Similar to the concept in the linear fitting, we define the slope in the inverse fit as A. Thus, the slope of linear fit represents the linear slope between PBLH and PM_2.5, whereas the slope of an inverse fit represents the linear slope between $-\frac{\mathrm{1}}{\mathrm{PBLH}}$ and PM_2.5. The sign of correlation coefficient for the inverse fit is the same as that of the slope. Obviously, the correlation coefficient and slope of the inverse fit for a positive relationship will be positive. Moreover, the normalized sample density at each location in a scatter plot represents the probability distribution in two dimensions (Scott, 2015). Then setting the weighting function of the inverse fit equal to the normalized density produces the best-fitting results representing the majority of cases. In general, we attempt both regular linear regression and inverse fitting to characterize the PBLH–PM relationships, and we provide the correlation coefficients and slopes for both fitting methods. In each case, the magnitude of correlation coefficient represents how well the observations are replicated by the fitting model, and the magnitude of slope represents the sensitivity of PM_2.5 to PBLH changes.

Figure 3Spatial distributions of climatological mean PBLH derived from CALIPSO for (a) March–April–May (MAM), (b) June–July–August (JJA), (c) September–October–November (SON), and (d) December–January–February (DJF) during the period 2006–2017. Spatial distributions of climatological mean of early-afternoon PBLH obtained from MERRA for (e) MAM, (f) JJA, (g) SON, and (h) DJF during the period 2006–2016.

Download

In addition, the statistical significance of the PBLH–PM relationships is tested by two independent statistical methods, namely the least squares regression and the Mann–Kendall (MK) test (Mann, 1945; Kendall, 1975). Least squares regression typically assumes a Gaussian data distribution in the trend analysis, whereas the MK test is a nonparametric test without any assumed functional form, and is more suitable for data that do not follow a certain distribution. To improve the robustness of the analysis, a correlation is considered to be significant when the confidence level is above 99 % for both least squares regression and the MK test. Hereafter, “significant” indicates that the correlation is statistically significant at the 99 % confidence level.

3.1 Climatological patterns of PBLH and surface pollutants

The climatology of the PBLH, especially its seasonal variability, is very important for air-pollution-related studies. We utilized the CALIPSO measurements from 2006 through 2017 to represent the spatial distribution of seasonal mean PBLH with interpolation, as shown in Fig. 3a–d. A smoothing window of 20 km was applied to the original PBLH data at 1∕3 km horizontal resolution. The seasonal climatological patterns of MERRA-derived PBLH are presented in Fig. 3e–h for a similar period. In general, the climatological pattern of MERRA PBLH is similar to that of CALIPSO, though the MERRA values are higher in spring and summer, and the peak values are lower in autumn and winter. Both CALIPSO and MERRA PBLHs are generally shallower in winter, when the development of the PBL is typically suppressed by the weaker solar radiation reaching the surface, and are generally higher in summer, especially for inland regions.

Note that there are still considerable differences between the CALIPSO- and MERRA-derived PBLH climatological patterns, which can be attributed to sampling biases, different definitions, and model uncertainties. First, since the spatial coverage and time resolution are quite different between the CALIPSO and MERRA datasets, the sampling used to calculate the climatologies are quite different. Moreover, MERRA PBLHs are derived from turbulent fluxes computed by the model, whereas CALIPSO usually identifies the top height of an aerosol-rich layer. Although turbulent fluxes would significantly affect aerosol structures, the different definitions still can cause differences between CALIPSO and MERRA PBLHs. The detailed relationship between of CALIPSO- and MERRA PBLHs is presented in Fig. S2d. Quantitatively, CALIPSO PBLH values exhibit considerable differences from MERRA results; the correlation coefficient of ∼0.4 indicates that the observations presented here will likely be useful for future model refinement. The reanalysis data do take into account large-scale dynamical forcing and have the ability produce the general PBLH climatology pattern (Guo et al., 2016). However, the reanalysis data do not consider the impact of aerosols, except with limited upper atmospheric measurement data assimilated, so the effects of aerosol–PBL interactions are poorly represented (Ding et al., 2013; Simmons, 2006; Huang et al., 2018). Thus, the current reanalysis data have limited ability to support a detailed investigation of PBLH–PM relationships.

Correspondingly, Fig. 4 presents the spatial distributions of seasonal mean PM_2.5 as measured at the surface stations. Both the PBLH and PM_2.5 over China exhibit large spatial and seasonal variations. The PM_2.5 seasonal pattern is generally coupled to that of PBLH; the lowest values occur in summer and the highest in winter. As a high PBLH facilitates the vertical dilution and dissipation of air pollution, the contrasting patterns of PBLH and PM_2.5 are consistent with expectation. The NCP is a major polluted region, with mean PM_2.5 concentrations overwhelmingly above 100 µg m⁻³ during winter. Both the PBLH and PM_2.5 also show strong seasonality over the NCP. The PRD is a relatively clean region, and PM_2.5 maintains low values (<50 µg m⁻³) through all seasons. As a reference, the seasonal means and standard deviations of PBLH and PM_2.5 over four ROIs are listed in Table S1 in the Supplement.

Figure 4Spatial distributions of climatological mean of early-afternoon PM_2.5 concentration (in µg m⁻³) for (a) MAM, (b) JJA, (c) SON, and (d) DJF during the period 2012–2017.

Download

From the seasonal climatologies, we find a coupling pattern between PBLH and PM_2.5, although one cannot assume a causal relationship from these plots alone. In subsequent sections, we use the lidar PBLH retrievals to investigate the PBLH–PM relationships in more detail.

3.2 Regional relationships between PM and PBLH

If the common factor driving large-scale variations in both PM and PBLH is meteorology, a regional analysis of their relationship could elucidate the meteorological impacts. We investigate the CALIPSO PBLH and surface PM_2.5 data case by case. By matching the available CALIPSO retrievals within 35 km of the surface PM_2.5 observations, we show the scatter plots for PBLH versus surface PM_2.5 for the four ROIs in Fig. 5. Despite the overall negative correlations, the correlations between PBLH and PM_2.5 have large spreads and differences. Both regular linear regression and inverse fitting are applied to characterize the PBLH–PM relationships. Significant negative correlations between PM_2.5 and PBLH are found over the NCP, with a Pearson correlation coefficient of −0.36. In addition, the nonlinear inverse function shows high consistency with the average values for each bin, and characterizes the PBLH–PM relationship with a somewhat higher correlation coefficient (−0.49). PBLH also shows significant negative correlation with PM_2.5 over the YRD and NEC, whereas the weak PBLH correlation with PM_2.5 over the PRD is not statistically significant. The correlation coefficients for the inverse fit are generally larger than the Pearson correlation coefficients, indicating that the nonlinear fit may be more suitable for characterizing the PBLH–PM relationships. Such improvements are obvious for the NCP and the YRD, but are not significant over the PRD and NEC.

Figure 5The relationship between CALIPSO-derived PBLH and early-afternoon PM_2.5 over (a) the NCP, (b) the PRD, (c) the YRD, and (d) NEC. The black dots and whiskers represent the average values and standard deviations for each bin. The red dashed lines indicate the regular linear regressions, and the black lines represent the inverse fit ($f\left(x\right)=A/x+B$). The detailed fitting functions are given at the top of each panels, along with the Pearson correlation coefficient (red) and the correlation coefficient for the inverse fit (black). Here and in the following analysis, R with asterisks indicates that the correlation is statistically significant at the 99 % confidence level. The color-shaded dots indicate the normalized sample density.

Download

We note that the ranges of PM_2.5 for these ROIs are significantly different; therefore, the background pollution level is likely to be an important factor for the PBLH–PM relationship. We thus normalize the PM_2.5 by MODIS AOD, a widely used parameter to represent the total-column aerosol amount, to qualitatively account for background or transported aerosol that is not concentrated in the PBL. The relationships between PBLH and PM_2.5/AOD over four ROIs are presented in Fig. 6. Clearly, after normalizing PM_2.5 by AOD, the spread of these scatter plots and the regional differences are significantly reduced, and the correlations become more significant for all ROIs, especially for the PRD. This is because transported aerosol aloft can contribute to variability in total-column AOD that is unrelated to the PBLH.

Figure 6Similar to Fig. 5, but for the relationship between CALIPSO PBLH and early-afternoon PM_2.5/AOD (unit: µg m⁻³ per AOD) over the four ROIs. Here, the AOD data are obtained from MODIS.

Download

Compared to CALIPSO data, the MPL has a much higher signal-to-noise ratio and can continuously observe at one location. Therefore, Fig. 7 shows the relationship between MPL-derived PBLH and PM_2.5 over Beijing (a major city in the NCP), as well as the relationship between PBLH and normalized PM_2.5. We find the PBLH–PM relationships derived from MPL over Beijing are similar to those derived from CALIPSO over the NCP. Probably because of higher data quality, the correlation coefficients for both fitting methods are slightly higher for the relationships derived from surface observations than those from CALIPSO. Consistent with the results over the NCP, the PBLH shows a significantly nonlinear relationship with PM_2.5 over Beijing. As the inverse fitting method better characterizes the PBLH–PM relationships than the regular linear fitting, we only use the inverse fitting method for the PBLH–PM relationships in the main text.

Figure 7(a) Relationship between MPL-derived PBLH and PM_2.5 over Beijing. (b) Relationship between MPL-derived PBLH and PM_2.5/AOD (unit: µg m⁻³ per AOD) over Beijing. The AOD data are obtained from AERONET. Here, linear (red) and inverse fits (black) are both shown. We only include data acquired during 10:00–15:00 local time, when the PBL is well developed.

Download

Figure 8The relationship between CALIPSO PBLH and PM_2.5 over the NCP for (a) MAM, (b) JJA, (c) SON, and (d) DJF. (e) General relationship between PM_2.5and PBLH aggregated over all seasons, with individual observations for each day plotted as gray dots. The box-and-whisker plots showing 10th, 25th, 50th, 75th, and 90th percentile values of PM_2.5 for each bin. The green, blue, pink, and red dots present the mean values for MAM, JJA, SON, and DJF, respectively.

Download

The most negative correlations between PBLH and PM_2.5 appear over the NCP, likely a testament to intense PBL–aerosol interactions, which may be caused by concentrated local sources. Comparing with southeast China, absorbing aerosol loading is much greater over the NCP, and may have strong interaction with the PBL through a positive feedback (Dong et al., 2017), which may contribute to the significant nonlinear relationships over the NCP. Note that the PBLH–PM_2.5 correlations are apparently stronger for heavily polluted regions than for clean regions. However, after normalizing PM_2.5 by AOD, the correlations are improved preferentially for clean regions (where aerosol aloft makes a larger fractional contribution to the total AOD), and thus, the differences between clean and polluted regions are reduced (Fig. S3). It further indicates that the background pollution level plays a critical role in interpreting the PBLH–PM observations.

As the NCP experiences the most pronounced seasonality in both PBLH and PM_2.5, the relationship over this region also shows the most prominent seasonal differences (Fig. S4). Figure 8 focuses on the seasonal dependence of the PBLH and PM_2.5 relationship over the NCP. The magnitude of the slope between 1∕PBLH and PM_2.5 for this region is ∼90 (unit: km µg m⁻³), with a correlation coefficient of −0.55 during winter, and only ∼40 in summer. For comparison, the seasonally aggregated relationship between PBLH and PM_2.5 is presented in Fig. 8e. PM_2.5 concentrations do not increase linearly with decreasing PBLH. Specifically, PM_2.5 increases rapidly with decreasing PBLH when PBLH is lower than 1 km, but changes much more slowly for PBLH >1.5 km. The seasonal mean values for PM_2.5 and PBLH are presented as colored dots in Fig. 8e, and the whiskers represent the standard deviations. For winter, the PBLH is generally shallow and PM_2.5 concentrations are high, and thus PBLH shows the most significant negative correlation with PM_2.5. Conversely, in summer, the PBLH is generally higher, PM_2.5 concentrations are lower, and the PBLH–PM_2.5 relationship is virtually flat. Such seasonally distinct PBLH–PM_2.5 relationships have not previously been studied quantitatively, and have the potential for improving PM_2.5 monitoring and predictions.

3.3 Association with horizontal transport

The PBLH mainly affects the vertical mixing and dispersion of air pollution, but horizontal transport also plays a critical role in surface air quality. Figure 9a–b present the PBLH–PM_2.5 relationships over China under strong wind (WS > 4 m s⁻¹) and weak wind (WS < 4 m s⁻¹) conditions. Under strong wind conditions, PM_2.5 is found to be much less sensitive to PBLH than for weak wind. In addition, Fig. 9c–d show the aerosol extinction profiles as a function of PBLH under strong and weak wind conditions, as retrieved by the MPL in Beijing, with the Klett method applied (Klett, 1985). In both strong and weak wind conditions, we find clear aerosol extinction gradients at the top of the PBL. Nonetheless, under strong wind, the aerosol extinction is typically low in the PBL, and the surface extinction does not change significantly with different PBLHs. In this situation, the strong wind likely plays a dominant role in affecting PM_2.5 concentration by ventilating the PBL. Under weak wind, the response of near-surface pollutants to PBLH is more nonlinear, and both aerosol extinction and PM_2.5 fall rapidly as the PBLH increases from 600 to 1200 m.

Figure 9The relationship between CALIPSO PBLH and PM_2.5 over China for (a) strong wind (WS > 4 m s⁻¹) and (b) weak wind (WS < 4 m s⁻¹). The aerosol extinction profiles at ∼550 nm derived from the MPL in Beijing change with different MPL-derived PBLHs under (c) strong wind and (d) weak wind conditions. In (c) and (d), the black dots indicate the location of the PBL top.

Download

We further consider the relationship between PBLH and PM_2.5 under different wind direction regimes for Beijing. Two different regimes are easy to identify: a northerly wind and a southerly wind; these are divided by the red line in Fig. 10a. The northerly air comes from arid and semiarid regions in northwest China and Mongolia, and is usually strong and clean. The southerly wind comes from the southern part of the NCP, with high humidity and aerosol content. To relate the connections between WS, PBLH, and surface air quality, at least qualitatively, the ventilation rate (VR) can be represented as VR = WS × PBLH (Tie et al., 2015). Figure 10b–c and d–e present the PBLH–PM_2.5 and VR–PM_2.5 relationships under southerly wind and northerly wind conditions, respectively. For all wind conditions, the VR shows a reciprocal relationship with surface PM_2.5. Under northerly wind conditions, both PBLH–PM_2.5 and VR–PM_2.5 relationships are flatter and have lower correlation coefficients. The northerly wind is apparently effective in removing pollutants and may play a dominant role in affecting air quality. For the southerly wind, the PM_2.5 concentration is highly sensitive to PBLH and VR values.

Figure 10(a) Relationship between wind direction/wind speed and PM_2.5 over Beijing. The red line divides the northerly wind and southerly wind regimes. (b–c) The relationship between PM_2.5 and MPL-PBLH/ventilation rate (VR = WS×PBLH, unit: km m s⁻¹), for southerly winds over Beijing. (d–e) The relationship between PM_2.5 and MPL-PBLH/VR, for northerly winds over Beijing.

Download

To further illustrate the coupling effects of PBLH and WS on surface pollutants, Fig. 11a presents the relationship between early-afternoon WS and PM_2.5 concentration across China. Overall, WS is negatively correlated with PM_2.5, although a few stations over southwest China show positive correlations. A negative correlation might be expected in general, as strong winds can be effective at removing air pollutants; however, other factors such as wind direction must also be considered, as, for example, upwind sources could increase pollution under higher wind conditions. There are positive correlations between PBLH and near-surface WS in most cases (Fig. S5a), and thus, low PBLH and weak WS tend to occur together over much of China. These unfavorable meteorological conditions for air quality would exacerbate severe pollution episodes.

Figure 11(a) Spatial distribution of linear correlation coefficients (R) for the WS–PM_2.5 relationship. (b) Spatial distribution of fitting parameter (A) for the VR–PM_2.5 relationship. The function ${\mathrm{PM}}_{\mathrm{2.5}}=A/\mathrm{VR}$ is used to characterize the relationship between VR and PM_2.5, with A (unit: km µg m⁻³) as the fitting parameter. Both WS and PM_2.5 are obtained from surface data, and PBLH is derived from CALIPSO. Here and in the following analysis, dots marked with black circles indicate where the relationship is statistically significant at the 99 % confidence level.

Download

To consider horizontal and vertical dispersion jointly, we investigate the nationwide relationships between the VR and PM_2.5. In general, the VR is overwhelmingly negative correlated with surface PM_2.5 (Fig. S5b). Based on Fig. 10, the VR is typically reciprocal to PM_2.5 for different wind conditions, and thus, we use the function $\mathrm{VR}=A/{\mathrm{PM}}_{\mathrm{2.5}}$ to characterize the relationship between the VR and PM_2.5, with A as the fitting parameter. The spatial distribution of A, presented in Fig. 11b, shows the largest values over the NCP, indicating that the PM_2.5 concentration is highly sensitive to the VR there. Moreover, VRs are relatively large over the coastal areas, where sea–land breezes could play a role in dispersing air pollution. The detailed relationships and fitting functions for four ROIs are presented in Fig. S6. We note that although there are large regional differences in the PBLH–PM_2.5 relationship (Fig. 5), the VR–PM_2.5 relationships are similar for the different study regions. Therefore, by combining vertical and horizontal dispersion conditions, the overall VR apparently has a similar effect on PM_2.5 for all four ROIs.

3.4 Correlations with topography

The PBL structure and PM_2.5 concentration can both be affected by topography. We divided the sites into two categories based on elevation: plains (elevation < 0.5 km) and highlands (elevation > 1 km). Figure 12a–d present the correlation coefficients and slopes in the inverse fit between PM_2.5 and PBLH for the plains and highland areas. For calculating the correlation coefficient and slope, we require that the number of matched CALIPSO PBLH and PM_2.5 samples is larger than 15 for each site. Much higher correlation coefficients are found in the plains than the highlands, and the slope (i.e., linear slopes between $-\frac{\mathrm{1}}{\mathrm{PBLH}}$ and PM_2.5) in the plains is ∼3 times that in the highlands. A reciprocal relationship is shown between station elevation and the slope between $-\frac{\mathrm{1}}{\mathrm{PBLH}}$ and PM_2.5 (Fig. 12e). The magnitudes of slopes decrease dramatically with elevation increase, for elevations between 0 and 500 m. Local emissions also affect aerosol loading, and differences between plains and highland areas regarding local source activity could be important here as well. Figure 12e shows that the low-elevation regions are typically more polluted than highland areas, and the magnitudes of the slopes tend to be higher. Here, we utilized the inverse fitting method to reveal the different PBLH–PM relationships for the plains and highland areas, and we reach a similar conclusion by using the linear fitting method (Fig. S7).

Figure 12Stratification by terrain elevation. The correlation coefficients (R) and slopes (unit: km µg m⁻³) between CALIPSO PBLH and PM_2.5 for the inverse fit ($f\left(x\right)=A/x+B$) are shown for the (a–b) plains and (c–d) highland areas. Note the slope in the inverse fit is defined as A. (e) The slopes of the inverse fit (i.e., linear slopes between $-\frac{\mathrm{1}}{\mathrm{PBLH}}$ and PM_2.5) under different station elevations, with color shading indicating station mean PM_2.5 concentration. (f) Box-and-whisker plots showing the 10th, 25th, 50th, 75th, and 90th percentile values of the early-afternoon WS for plain and highland regions. The dots indicate the mean values.

Download

Returning to Fig. S3, stronger correlations for PBLH–PM_2.5 relationships are found over polluted regions, which also correspond to the plains areas, due to strong local emissions. Therefore, high aerosol loading is likely to be another factor contributing to the strong correlation between PBLH and PM_2.5 over the plains, whereas the low PM_2.5 concentration may contribute to the weak PBLH–PM_2.5 correlation over the highlands.

In addition, horizontal transport is associated with topography. Thus, we illustrate the distribution of WS for plains and highland areas in Fig. 12f. WS is generally larger for highland areas, especially for the strongest wind cases. In fact, the 10 % and 25 % quantiles of WS are nearly the same between plains and highland areas, whereas there are clear differences in the 75 % and 90 % quantiles. Strong wind cases account for 37% of the total over highland areas, but only 27 % of the total over the plains. As discussed in Sect. 3.3, strong wind can effectively remove surface pollutants, and can play a dominant role in determining local pollution levels. In this situation, PBLH might not play as critical a role in PM_2.5 concentration. Thus, mountain winds, along with less local emission, are likely to be leading factors accounting for the differences in PBLH–PM_2.5 correlations between plains and highland areas.

Other factors could come into play as well, such as the vertical distribution of aerosol, the insolation, and the actual single-scattering albedo of the particles; further examination of these phenomena is beyond the scope of the current paper.