Quantiﬁcation of Regional Ozone Pollution Characteristics and Its Temporal Evolution: Insights from Identiﬁcation of the Impacts of Meteorological Conditions and Emissions

: Ozone (O 3 ) pollution has become the major new challenge after the suppression of PM 2.5 to levels below the standard for the Pearl River Delta (PRD). O 3 can be transported between nearby stations due to its longevity, leading stations with a similar concentration in a state of aggregation, which is an alleged regional issue. Investigations in such regional characteristics were rarely involved ever. In this study, the aggregation (reﬂected by the global Moran’s I index, GM), its temporal evolution, and the impacts from meteorological conditions and both local (i.e., produced within the PRD) and non-local (i.e., transported from outside the PRD) contributions were explored by spatial analysis and statistical modeling based on observation data. The results from 2007 to 2018 showed that the GM was positive overall, implying that the monitoring stations were surrounded by stations with similar ozone levels, especially during ozone seasons. State of aggregation was reinforced from 2007 to 2012, and remained stable thereafter. Further investigations revealed that GM values were independent of meteorological conditions, while closely related to local and non-local contributions, and its temporal variations were driven only by local contributions. Then, the correlation (R 2 ) between O 3 and meteorology was identiﬁed. Result demonstrated that the westerly belonged to temperature (T) and surface solar radiation (SSR) sensitive regions and the correlation between ozone and the two became intense with time. Relative humidity (RH) showed a negative correlation with ozone in most areas and periods, whereas correlations with u and v were positive for northerly winds and negative for southerly winds. Two important key points of such investigation are that, ﬁrstly, we deﬁned the features of ozone pollution by characterizing the temporal variations in spatial discrepancies among all stations, secondly, we highlighted the signiﬁcance of subregional cooperation within the PRD and regional cooperation with external environmental organizations.


Introduction
The Pearl River Delta (PRD), the largest city cluster in South China, has long suffered from severe air pollution due to rapid urbanization and intensive anthropogenic activities. Following the introduction of a series of stringent air pollution control measures, levels of most atmospheric pollutants in the PRD, such as SO 2 , NO x , CO, PM 10 , and PM 2.5 , have gradually decreased in recent years; however, tropospheric ozone (O 3 ) is the exception and it became the primary pollutant since 2015 and the amount of days exceeding the

Identification of the Impacts from Local, Non-Local and Meteorological Factors on O3
To better understand the correlations between O3 and meteorological variables, time (t) data series X(t) were separated into short-term (ST), seasonal (SE), and long-term (LT) components as expressed in Equation (1) [37,38].
The sum of seasonal and long-term trend components is the baseline, and each component can be determined using a KZ filter, which repeats the iterations of a moving average to remove the high-pass signal defined by: where k is the number of values included on each side, the window length m = 2k + 1, i is the interval time, j is the window variables, and Y is the output time series. Different time scales can be obtained by changing the window length and the number of iterations [39,40].

Identification of the Impacts from Local, Non-Local and Meteorological Factors on O 3
To better understand the correlations between O 3 and meteorological variables, time (t) data series X(t) were separated into short-term (ST), seasonal (SE), and long-term (LT) components as expressed in Equation (1) [37,38].
Atmosphere 2021, 12, 279 4 of 15 The sum of seasonal and long-term trend components is the baseline, and each component can be determined using a KZ filter, which repeats the iterations of a moving average to remove the high-pass signal defined by: where k is the number of values included on each side, the window length m = 2k + 1, i is the interval time, j is the window variables, and Y is the output time series. Different time scales can be obtained by changing the window length and the number of iterations [39,40]. A KZ (15,5) filter with a window length of 15 with five iterations removes cycles of 33 days referring to the baseline variations (BL).
BL(t) = KZ (15,5) The long-term trend can be separated from the raw data by KZ (365, 3) with a period >632 days, and then the seasonal and short-term component ST(t) can be derived by SE(t) = KZ (15,5) ST(t) = X(t) − BL(t) = X(t) − KZ (15,5) To explore the factors driving the temporal evolution of O 3 aggregation and the impacts of meteorological conditions, local and non-local contributions were identified. The methods were as described in our related studies [28]. Briefly, a multiple linear regression (MLP) model was used to perform meteorological adjustments. Local and nonlocal sources were identified with an empirical orthogonal function (EOF) model and their contributions were estimated with absolute principal component scores (APCS). In our previous investigations, we treated the first principal as non-local, and local contributions were determined by subtracting non-local values from the original data. The statistic models were developed with R language (version 3.5).

Determination of Relationships between O 3 and Meteorological Factors
MLP was conducted using stepwise regression between baseline O 3 values and meteorological factors in determining the coefficients of determination [41,42]. We ignored the short-term component as it was weakly correlated compared with the relations between baseline O 3 and meteorological variables [35].
A BL (t) = a BL + ∑ b BLi ·M BLi + ∈ BL (t) (6) where A BL (t) and M BL are the baselines of the O 3 and meteorological factors, respectively. The parameters a, b, and ∈ are fitted parameters and the residual term. The coefficient of determination (R 2 ) of MLP reflects the relation between O 3 and meteorological conditions, and R 2 values between O 3 and single meteorological factors were obtained if M BLi in Equation (6) contains one meteorological factor. The negative sign will be added to R 2 if there exists a negative correlation between O 3 and the single meteorological factor.

Calculation of Degree of Aggregation Dispersion of Stations with Similar O 3 Levels
The global Moran's I index (GM) can be used as a spatial autocorrelation analysis technology to explore the dispersion or unification of attribute values in a given region. In this case, it reflects the correlations of O 3 concentrations at different stations, taking the spatial weights of all stations into account. GM values range from −1 to 1, with positive/negative values indicating positive/negative correlations among all O 3 stations. GM values approaching 1 or −1 represent strong positive or negative relations, respectively, and a GM value approaching 0 indicates no obvious association. GM was calculated as follows.
where I is GM, x j is the observed value of a region, w ij is the spatial weight matrix, and S is variance. We utilized local Moran'I (LM) to explore the correlation of a station with its adjacent stations in a small area. LM constitutes the normalized O 3 concentration of a station and the adjacent station and their scatter plots in quadrant can discern their correlations. It exposes the heterogeneity shadowed by GM and is often involved in recognizing pollution hotspots in geography [43][44][45]. Three stations closest to a site were used to calculate the normalized lagged O 3 concentration, which represents the average O 3 level adjacent to a site. The normalized O 3 value of a site and its lagged value were assigned to a two-dimensional plot and correlations between the two were visualized according to their locations in quadrants ( Figure 2). "high-high" (H-H) in the first quadrant [(1) in Figure 2] indicates a site with a high attribute that is encircled by high-attribute sites.

Aggregation of Stations with Similar O3 Concentrations over Different Time Scales
O3 is a regional pollution issue with stations showing similar concentrations distributed close together. The concentration is always affected by meteorological conditions, local precursor emissions and transport of non-local O3 or precursors from outside the local area. The heterogeneity or consistency of average O3 concentrations at multiple stations may vary with changes in these factors, causing fluctuations in the correlations between each station and its adjacent stations. In this section, the aggregation on different time scales, long-term evolution, and driving factors are analyzed to explore the pollution characteristics. To clarify the factors that drive annual GM values, the impacts of meteorological conditions and local and non-local contributions were identified and their annual GM values are shown in Figure 3. We found that GM values were almost constant regardless of fluctuations in meteorological conditions. This implies that the annual average spatial distribution of O3 is independent of fluctuations in meteorological conditions. Our previous investigation demonstrated that O3 concentrations were sensitive to meteorological conditions in the western region of the PRD [28], while O3 concentrations were high in the northeastern region during the period 2013 to 2017 [46]. Meteorological conditions still have an important influence on O3 concentrations, and we discuss the spatiotemporal evolution of the correlation between O3 and meteorological conditions in Section 3.2.
The annual GM (red line, local GM) decreased when non-local contributions were removed and the annul non-local GM retained the same negative values. Therefore, we inferred that the overall temporal GM was likely driven by local contributions as local and annual GM showed similar temporal variations. Negative non-local GM values near 0 imply that O3 transported from outside of the area was distributed discretely in the PRD. GM values would be increased by the impact of non-local contributions as the discrepancies between O3 concentrations at different stations were evened out by non-local contri-

Aggregation of Stations with Similar O 3 Concentrations over Different Time Scales
O 3 is a regional pollution issue with stations showing similar concentrations distributed close together. The concentration is always affected by meteorological conditions, local precursor emissions and transport of non-local O 3 or precursors from outside the local area. The heterogeneity or consistency of average O 3 concentrations at multiple stations may vary with changes in these factors, causing fluctuations in the correlations between each station and its adjacent stations. In this section, the aggregation on different time scales, long-term evolution, and driving factors are analyzed to explore the pollution characteristics. butions. These observations emphasize that O3 pollution is a regional issue and was intensified by local contributions from 2007 to 2012. Furthermore, O3 concentrations at most stations in the PRD increased during this period [28], and the increasing GM values imply that O3 levels increased faster at stations with previously low levels, thereby reducing the difference compared to high O3 stations. Non-local contributions had no effect on temporal GM fluctuations, whereas they enhanced O3 concentrations at low-level stations. Therefore, local and regional cooperation is necessary to restrict O3 pollution. It should be noted that GM dropped to 0.1 (p > 0.05) in 2018, implying that O3 concentrations were discretely distributed, which may have been related to abnormal weather in that year. Fluctuations in T and SSR intensified, accompanied by significant differences in temperature and precipitation compared to previous years, and there were several typhoons in 2018 [47]. Different stations were affected to varying degrees by meteorological conditions, resulting stations in high/low O3 levels encompassed by low/high levels stations, which should be explored further in future studies. Analysis of the monthly GM values for each year compared with the averages of all years ( Figure 4) revealed that months with high O3 levels (marked with digits) were usually coupled with high GM values, indicating that O3 concentrations at most stations throughout the region became more similar in O3 seasons. GM values were low or even negative in months with low O3 concentrations. The polarization of GM values in different months demonstrates that control of O3 during periods of high pollution requires the co- To clarify the factors that drive annual GM values, the impacts of meteorological conditions and local and non-local contributions were identified and their annual GM values are shown in Figure 3. We found that GM values were almost constant regardless of fluctuations in meteorological conditions. This implies that the annual average spatial distribution of O 3 is independent of fluctuations in meteorological conditions. Our previous investigation demonstrated that O 3 concentrations were sensitive to meteorological conditions in the western region of the PRD [28], while O 3 concentrations were high in the northeastern region during the period 2013 to 2017 [46]. Meteorological conditions still have an important influence on O 3 concentrations, and we discuss the spatiotemporal evolution of the correlation between O 3 and meteorological conditions in Section 3.2.
The annual GM (red line, local GM) decreased when non-local contributions were removed and the annul non-local GM retained the same negative values. Therefore, we inferred that the overall temporal GM was likely driven by local contributions as local and annual GM showed similar temporal variations. Negative non-local GM values near 0 imply that O 3 transported from outside of the area was distributed discretely in the PRD. GM values would be increased by the impact of non-local contributions as the discrepancies between O 3 concentrations at different stations were evened out by non-local contributions. These observations emphasize that O 3 pollution is a regional issue and was intensified by local contributions from 2007 to 2012. Furthermore, O 3 concentrations at most stations in the PRD increased during this period [28], and the increasing GM values imply that O 3 levels increased faster at stations with previously low levels, thereby reducing the difference compared to high O 3 stations. Non-local contributions had no effect on temporal GM fluctuations, whereas they enhanced O 3 concentrations at low-level stations. Therefore, local and regional cooperation is necessary to restrict O 3 pollution.
It should be noted that GM dropped to 0.1 (p > 0.05) in 2018, implying that O 3 concentrations were discretely distributed, which may have been related to abnormal weather in that year. Fluctuations in T and SSR intensified, accompanied by significant differences in temperature and precipitation compared to previous years, and there were several typhoons in 2018 [47]. Different stations were affected to varying degrees by meteorological conditions, resulting stations in high/low O 3 levels encompassed by low/high levels stations, which should be explored further in future studies.
Analysis However, heterogeneity between a single station and its adjacent regions within a small district will be shadowed by GM values. Hence, local autocorrelation analysis was performed to examine these features. As shown in Figure 5a, stations were distributed mainly in the first and third quadrants, indicating that stations were surrounded by other stations with similar O 3 levels, consistent with high positive GM values. CW, TC, TL, and LY with low concentration levels located in or near Hong Kong (HK) are associated with the L-L pattern because of their relatively low precursor emissions [19]. Furthermore, dilution by the sea breeze and increased precipitation in coastal regions would also lead to low O 3 levels in these areas. The remaining stations mostly fell within the first quadrant, indicating that these stations simultaneously experienced high O 3 levels compared with those of sites in or near HK. O 3 values were highest at TH, and the three nearest stations,

GM on Different Time Scales
GM reflects the autocorrelation of the O3 concentrations of all stations using a single index. This index indicates only the degree of aggregation or dispersion of O3 concentrations in the region. However, heterogeneity between a single station and its adjacent regions within a small district will be shadowed by GM values. Hence, local autocorrelation analysis was performed to examine these features. As shown in Figure 5a, stations were distributed mainly in the first and third quadrants, indicating that stations were surrounded by other stations with similar O3 levels, consistent with high positive GM values. CW, TC, TL, and LY with low concentration levels located in or near Hong Kong (HK) are associated with the L-L pattern because of their relatively low precursor emissions [19]. Furthermore, dilution by the sea breeze and increased precipitation in coastal regions would also lead to low O3 levels in these areas. The remaining stations mostly fell within the first quadrant, indicating that these stations simultaneously experienced high O3 levels compared with those of sites in or near HK. O3 values were highest at TH, and the three nearest stations, HG, HJC, and LH, had similarly high values. These sites are located in the north of the PRD and in northerly winds are the most susceptible to non-local O3 from inland. With a southerly wind, O3 from the south will settle in areas to the north. Both situations could facilitate the accumulation of O3 in areas north of the PRD. JGW and TJ were distributed in the fourth quadrant because of the influence of nearby HK, which had the lowest O3 concentrations. Therefore, programs to restrain O3 in the PRD should take into account geographical location and the effects on upwind areas.
Variations in the scatter diagram of LM for the different years are shown in Figure  5b. Hong Kong remained L-L and most other regions stabilized in the first quadrant throughout the investigation. However, although some stations, such as LH and DH, switched among different quadrants, they finally settled in the first or second quadrant, indicating that these stations coexisted with surrounding high O3 regions. The pattern of To explore the monthly aggregation in local areas as part of the whole region, we calculated the lagged O3 levels of all stations with the monthly averages from all periods and the results are shown in Figure 5c. The stations were allocated to the quadrants from January to April and during December. This was consistent with the monthly GM values, which were low during the same periods (Figure 4). High O3 concentrations occurred from May to November in the PRD with high monthly GM values ( Figure 4) and with monthly local autocorrelations having H-H and L-L patterns (Figure 5c). These observations show that the discrepancies of O3 concentration from the whole region were shrank in high O3 level months, which implied high/low O3 stations were enclosed with high/low stations around during high pollution periods. Inversely, high/low O3 stations were encircled by low/high stations around during non-high pollution periods relatively. These observations indicate the need to formulate different O3 control measures according to specific local pollution conditions.   To explore the monthly aggregation in local areas as part of the whole region, we calculated the lagged O 3 levels of all stations with the monthly averages from all periods and the results are shown in Figure 5c. The stations were allocated to the quadrants from January to April and during December. This was consistent with the monthly GM values, which were low during the same periods (Figure 4). High O 3 concentrations occurred from May to November in the PRD with high monthly GM values ( Figure 4) and with monthly local autocorrelations having H-H and L-L patterns (Figure 5c). These observations show that the discrepancies of O 3 concentration from the whole region were shrank in high O 3 level months, which implied high/low O 3 stations were enclosed with high/low stations around during high pollution periods. Inversely, high/low O 3 stations were encircled by low/high stations around during non-high pollution periods relatively. These observations indicate the need to formulate different O 3 control measures according to specific local pollution conditions.

Spatial Distribution of Meteorological Conditions-O 3 Correlations and Its Temporal Evolutionary Characteristics
The results outlined in Section 2 indicate that annual GM values were independent of meteorological conditions. However, O 3 concentrations have been shown to be markedly influenced by meteorological fluctuations [29,[48][49][50][51][52]. This section discusses the correlation between meteorological fluctuations and their spatiotemporal evolution. As shown in Figure 6a [32][33][34]. R 2 values in the northeastern region were relatively low, with a minimum value of 0.24 at TH (Figure 6a), indicating that O 3 in these areas was likely regulated mainly by changes in its precursors or by non-local transportation.
R 2 values between O 3 and single meteorological variables are shown in Figure 6b-f. The R 2 of SSR and T (b and c) had similar spatial distributions and governed the overall picture of MET R 2 , implying that they were the major factors influencing O 3 concentrations. R 2 values were low in coastal regions, but high in western and central-western areas. This was because precursor emissions of O 3 were concentrated in the western and centralwestern areas [36], and temperature and solar radiation can influence O 3 production directly or indirectly. Areas with high precursor emissions are more sensitive to T and SSR and will probably experience higher O 3 levels as T and SSR will increase with climate change and with the alleviation of particulate matter pollution. RH showed slight negative correlations in most areas, especially in coastal cities (Figure 6d), which was likely associated with wet deposition of O 3 precursors. The R 2 values of u and v (Figure 6e-f) had similar spatial distributions and magnitudes, and the negative correlations near the ocean were probably associated with dilution by sea breeze. We speculate that the positive R 2 values in the north reflect transport of O 3 and its precursors from inland. and will probably experience higher O3 levels as T and SSR will increase with climate change and with the alleviation of particulate matter pollution. RH showed slight negative correlations in most areas, especially in coastal cities (Figure 6d), which was likely associated with wet deposition of O3 precursors. The R 2 values of u and v (Figure 6e-f) had similar spatial distributions and magnitudes, and the negative correlations near the ocean were probably associated with dilution by sea breeze. We speculate that the positive R 2 values in the north reflect transport of O3 and its precursors from inland. The annual geographical distributions and average annual variations in R 2 between O 3 and all meteorological factors are shown in Figures 7 and 8. High correlations were seen in the southwest in most years, but there were large discrepancies between different years (Figure 7). In the long term, the variations in R 2 showed no obvious tendencies for the whole region, except that R 2 was relatively low from 2016 to 2018. Examination of the spatial distribution of R 2 for each meteorological variable with O 3 (Figures S1-S5) implied that the sensitivity of the western area to meteorological conditions was due mainly to SSR and T and that u and v were responsible for the totally high R 2 values in the northeast from 2015. R 2 values in SSR and T had similar spatial distributions throughout all periods and were higher during the last 8 years than the first few years. Negative correlations were seen between O 3 and RH in most periods, especially in coastal areas. The R 2 values of u and v with O 3 were positive in the north and became more intense with the years, signifying that ozone was more sensitive to wind in the north, while values were negative in the south and the last to become positive, signifying that O 3 in the south was likely induced by wind.
Annual MET R 2 (black dashed line in Figure 8) values showed a slight decrease from 2007 to 2018, accompanied by increases in R 2 values for SSR and T and reductions in RH, u, and v. Hence, MET R 2 was suppressed by RH, u, and v overall. R 2 values for SSR and T remained highly consistent in tendency and magnitude due to the high correlation between SSR and T. The decline in NO x and increase in VOCs were relatively steady over the last decade [1], so the peaks in 2012 and 2016 were probably related to the marked fluctuations in SSR and T ( Figure S6). The increases in R 2 values for T and SSR imply that the PRD, especially the areas with O 3 concentration sensitive to meteorological conditions, will likely suffer more severe O 3 pollution in the future at present emission levels. On the whole, u, v, and RH acted as diluters initially based on the negative R 2 , and this occurred mainly in the southern parts of the PRD; O 3 showed positive correlations with u and v, and was independent of RH in the last few years. Xue et al. reported that Hong Kong was experiencing increasing O 3 transport from the PRD [53]. Therefore, we assume that the negative correlations of u-R 2 and v-R 2 values occurred when dilution by wind was dominant, and O 3 concentrations in the PRD were relatively low. More O 3 was transported to the south from the PRD when O 3 levels in the PRD were high, leading to positive u and v R 2 values in the southern regions of the PRD. These findings imply that efforts to reduce emissions may be offset by adverse meteorological conditions and indicate that it is necessary to clarify O 3 transport by the wind to restrict levels in the PRD. spatial distribution of R 2 for each meteorological variable with O3 ( Figures S1-S5) implied that the sensitivity of the western area to meteorological conditions was due mainly to SSR and T and that u and v were responsible for the totally high R 2 values in the northeast from 2015. R 2 values in SSR and T had similar spatial distributions throughout all periods and were higher during the last 8 years than the first few years. Negative correlations were seen between O3 and RH in most periods, especially in coastal areas. The R 2 values of u and v with O3 were positive in the north and became more intense with the years, signifying that ozone was more sensitive to wind in the north, while values were negative in the south and the last to become positive, signifying that O3 in the south was likely induced by wind. Annual MET R 2 (black dashed line in Figure 8) values showed a slight decrease from 2007 to 2018, accompanied by increases in R 2 values for SSR and T and reductions in RH, u, and v. Hence, MET R 2 was suppressed by RH, u, and v overall. R 2 values for SSR and T remained highly consistent in tendency and magnitude due to the high correlation between SSR and T. The decline in NOx and increase in VOCs were relatively steady over the last decade [1], so the peaks in 2012 and 2016 were probably related to the marked fluctuations in SSR and T ( Figure S6). The increases in R 2 values for T and SSR imply that the PRD, especially the areas with O3 concentration sensitive to meteorological conditions, will likely suffer more severe O3 pollution in the future at present emission levels. On the whole, u, v, and RH acted as diluters initially based on the negative R 2 , and this occurred mainly in the southern parts of the PRD; O3 showed positive correlations with u and v, and was independent of RH in the last few years. Xue et al. reported that Hong Kong was experiencing increasing O3 transport from the PRD [53]. Therefore, we assume that the negative correlations of u-R 2 and v-R 2 values occurred when dilution by wind was dominant, and O3 concentrations in the PRD were relatively low. More O3 was transported to the south from the PRD when O3 levels in the PRD were high, leading to positive u and v R 2 values in the southern regions of the PRD. These findings imply that efforts to reduce emissions may be offset by adverse meteorological conditions and indicate that it is necessary to clarify O3 transport by the wind to restrict levels in the PRD.

Conclusions and Discussion
In this study, the aggregation (as reflected by GM) of O3 concentrations at all stations in the PRD and their temporal evolution were analyzed to elucidate regional issues related to O3 pollution. The impacts of meteorology, local and non-local contribution were identified to determine the driving factors of GM variation. The results show that stations with similar O3 levels aggregated distributed more in PRD. The increases in annual GM from 2007 to 2012 indicate that the differences in O3 concentrations among stations decreased and O3 approached to a similar level. Further investigation showed that GM values were independent of meteorological conditions and were markedly enhanced by non-local con-

Conclusions and Discussion
In this study, the aggregation (as reflected by GM) of O 3 concentrations at all stations in the PRD and their temporal evolution were analyzed to elucidate regional issues related to O 3 pollution. The impacts of meteorology, local and non-local contribution were identified to determine the driving factors of GM variation. The results show that stations with similar O 3 levels aggregated distributed more in PRD. The increases in annual GM from 2007 to 2012 indicate that the differences in O 3 concentrations among stations decreased and O 3 approached to a similar level. Further investigation showed that GM values were independent of meteorological conditions and were markedly enhanced by non-local contributions and that the temporal variations in GM were driven by local contributions. GM values were higher in O 3 seasons and became small in low O 3 months. Furthermore, stations near HK had similarly low levels and the remainder had high O 3 levels, as characterized by LM. Thus, regional O 3 issues became more prominent, which was mainly due to local and non-local contributions.
To reduce O 3 pollution in the PRD, further substantial reductions in emissions are required. Cooperation between regions within the PRD and with environmental agencies outside the PRD will be crucial to reduce transport from upwind areas. Ozone concentration in the westerly of PRD was more sensitive to T and SSR, and the R 2 between ozone and meteorological factors increased over the years, so O 3 concentrations will probably increase even if emissions are kept constant as the warming climate, and additional efforts are required to reduce pollution in these areas. Particularly, RH-R 2 values were negative in most areas and periods, which is reasonable due to wet deposition in O 3 and its precursors. The R 2 values of u and v were positive in northern regions and increased over the years, while being negative in southern regions and eventually becoming positive, implying that O 3 was more likely to be transported into the area by wind, especially in the northern regions of the PRD. Therefore, it is necessary to characterize the impacts of meteorological conditions for effective emission reduction, and additional attention and efforts are needed in the meteorology-sensitive regions.
Supplementary Materials: The following are available online at https://www.mdpi.com/2073-443 3/12/2/279/s1, Figure S1: Annual coefficients of determination (R 2 ) between the baseline of daily MDA8 and SSR. Figure S2: Annual coefficients of determination (R 2 ) between the baseline of daily MDA8 and T. Figure S3: Annual coefficients of determination (R 2 ) between the baseline of daily MDA8 and RH. Figure S4: Annual coefficients of determination (R 2 ) between the baseline of daily MDA8 and u. Figure S5: Annual coefficients of determination (R 2 ) between the baseline of daily MDA8 and v. Figure