Overall, the seasonal patterns in both AOD and PW are pretty consistent between AERONET and the NAAPS-RA (Fig. 3). For example, peak AOD values in North America and Europe occur during the summer months in both datasets. Likewise, peak AOD values are found over the Sahel in winter and spring due to a combination of dry-season biomass burning and dust associated with the northeasterly Harmattan winds with shifts in peak AOD further north in summer due to increased dust activity over the Sahara. Like the Sahel, peak AOD values associated with fire activity during regional dry seasons are also found in both datasets for Central and South America, southern Africa, and Southeast Asia. Boreal regions, which also exhibit seasonality due to fires in summer months, are not as well sampled in the AERONET dataset, making it harder to see seasonal shifts in AOD. However, this seasonality is found in the NAAPS-RA. Likewise, northward shifts in PW are seen in AERONET and the NAAPS-RA in the summer and a southward shift in the winter months. A more in-depth discussion of the data by region, which is consistent with verification regions presented in Lynch et al. (2016) and Rubin et al. (2016), is presented below:

North America. The largest number of AERONET sites is present in this region, with ∼180 included in the analysis. AERONET data counts are the highest in the summer months (JJA), which also coincides with peak mean AOD and PW values in both the AERONET and NAAPS-RA datasets (Fig. 3). Summertime peak AOD values are associated with ABF and smoke aerosol types, concentrated to the north, and a combination of ABF and transported dust to the south (Fig. 4). Despite JJA being associated with the highest AOD values, the IQR is only around 0.1–0.2 (Fig. 5). The 90th percentile AOD values in JJA for North America are mainly associated with large smoke events, particularly originating from the Pacific Northwest and boreal regions (Fig. 5). High AOD values are also observed in MAM months, concentrated in the Southeastern United States (Figs. 2 and 5), associated with smoke (originating from Central American fires) and ABF/pollution aerosol types (Fig. 5).

The NAAPS-RA daily correlations (Fig. 7) within seasonal aggregates indicate similar but not identical spatial patterns relative to the AERONET dataset. The dominant positive-correlation regions include the Eastern/Southeastern United States, as is found in AERONET. Likewise, stronger European AOD and PW correlations are found in the summer months (JJA), in eastern Asia in the winter season (DJF), and the Middle East in the fall (SON). The NAAPS-RA results are helpful in that they provide a more complete perspective on the AOD and PW relationships. In addition to strong positive correlations in Southeastern United States and East Asia during DJF, the NAAPS-RA also indicates strong positive correlations in parts of Southwest Asia (Iran, Afghanistan, and Pakistan), India, South America, and southern Africa, which are minimally if not at all sampled by AERONET. The spatial extent of the observationally sampled relationships can also be seen. For example in MAM, the AERONET correlation at the Tamanrasset site in Algeria is 0.55, with a consistent NAAPS-RA correlation of 0.53. In the reanalysis, the correlations, greater than 0.5, extend to the east of Tamanrasset. Likewise, the spatial extent of correlations for maritime regions can be seen in the reanalysis, where AERONET sites are rare. In JJA months, correlations at the Dakhla site in Morocco are 0.45 in the AERONET dataset. Although the NAAPS-RA correlation at Dakhla is weaker (R=0.31), the positive relationship observed in both datasets on the west coast of Africa can be seen extending out into the Atlantic ocean in the reanalysis, consistent with dust transport pathways. Correlations associated with aerosol transport are also seen in southern Africa in the reanalysis, extending out into the ocean.

Although positive correlations are dominant throughout the world, negative correlations were also identified in both the AERONET and NAAPS-RA datasets from daily data. In the AERONET dataset, negative correlations are limited to the tropic/subtropics, with negatively correlated regions mostly associated with biomass burning. The strongest negative correlations in the AERONET dataset are shown in Tables 1–4, with NAAPS-RA values shown for comparison. During all seasons, negative correlations are found in the Sahel region in both AERONET and the NAAPS-RA, with the negative relationships extending further northwards in the boreal spring and summer months. This results in an exceptionally strong dipole between Saharan and Sahelian outflow and is likely related to shifts in the ITCZ. This points to aerosol sources (biomass burning and dust) and scavenging as a cause of the negative AOD and PW relationship. The NAAPS-RA shows these negative correlations extending into the Atlantic Ocean with seasonally dependent differences. Negative correlations extend into the Caribbean in JJA and to the northern parts of South America in MAM, consistent with seasonal transport pathways. Other negative-correlation regions include Southeast Asia, South America, and southern Africa. For these regions, the strongest negative correlations are associated with the respective dry, burning seasons. For example, negative correlations are strongest in peninsular Southeast Asia in MAM and in insular Southeast Asia and South America in SON. In these cases, negative AOD and PW relationships are likely a result of higher aerosol emission occurring under dry conditions, which lead to more fire activity. Southern Africa is an exception during JJA, in which smoke aerosol is dominant (Fig. 4). However, this is consistent with previous studies which have found elevated free tropospheric water vapor levels associated with southern African smoke events (Adebiyi et al., 2015; Pistone et al., 2021). Correlations in both AERONET and the NAAPS-RA are positive in JJA and negative in MAM and SON when smoke aerosol is also present but not at its peak. One of the largest AERONET negative correlations occurs at the Jomson site, Nepal, in JJA, with a value of -0.65 (Table 3), although nearby sites show small or statistically insignificant correlations. The Jomson site is located at 2825 m, with maximum PW values around 2, while the nearby Pokhara site is 2000 m lower in altitude, with maximum PW values around 5; therefore, Jomson is likely a regional outlier due to altitude effects. For Jomsom and the surrounding regions, the NAAPS-RA indicates no statistically significant correlation. While NAAPS and AERONET are in general agreement in the locations of negative correlations, this discrepancy is likely related to mesoscale or small-scale features that are not captured in a global, 1∘ model.

While the global plots of AERONET and NAAPS-RA AOD and PW relationships give a sense of spatial agreement, a scatterplot comparison of the quantitative values generated from the two datasets is used to take a closer look at the consistency between the observed and predicted relationships. A seasonal comparison of AERONET and NAAPS-RA regressions is shown as a scatterplot in Fig. 9, including site-by-site (a) correlations, (b) Theil–Sen slopes, and (c) the PW mean difference for high-AOD events. In addition to the three scatterplot comparisons, all locations for which for the sign of the AOD and PW relationships differed between the AERONET and the NAAPS-RA datasets were identified. For these identified sites, the distribution of AERONET correlations is plotted by season in Fig. 9d. This is included as a means to examine the strength of the observed AOD–PW relationship under conditions when the datasets disagree. Overall, the observations and model are in general agreement in the sign of the correlations (Fig. 9a), with similar results found for the Theil–Sen slope and the PW mean differences for high-AOD events (Fig. 9b, c). Differences in the sign of the correlation are found for 15.5 %, 9.5 %, 10.2 %, and 10.1 % of analyzed AERONET sites for the DJF, MAM, JJA, and SON months, respectively. For all seasons except JJA, these differences are mostly associated with a negative correlation in the AERONET data and a positive value in the NAAPS-RA. Differences in correlation sign occur for sites in which the AERONET-generated correlations are weak, mostly falling below 0.20 (Fig. 9d), with the exception of the Jomson AERONET site in JJA in which AERONET indicated a strong negative correlation and the reanalysis had a slight positive but statistically insignificant relationship, as previously discussed. For the strongest correlation sites, AERONET and NAAPS are in good agreement in DJF and MAM (Tables 1 and 2). For JJA and SON, NAAPS-RA has a tendency to produce weaker correlations relative to AERONET (Tables 3 and 4). Some differences are expected given that the event sampling is different between the AERONET observations and the 16-year NAAPS-RA. However, the overall agreement in the correlations between the two datasets provides some confidence in the NAAPS-RA for generating regionally and seasonally varying AOD and PW relationships on a global scale.

With the consistency between AERONET and NAAPS-RA established, a more thorough evaluation of the strength of the slope of AOD–PW relationship has been conducted. As previously discussed, the AOD and PW relationship in both the AERONET and NAAPS-RA datasets was quantified using a Theil–Sen regression in order to fit a slope to the change in AOD per centimeter PW (Figs. 7 and 8). Examples of NAAPS-RA and AERONET Theil–Sen fittings for eight AERONET sites scattered over the globe, each with their own unique aerosol environment, are shown in Fig. 10. Included are positive- and negative-correlation examples shown for each season (DJF – Beijing, China, and Lamto, Côte D'Ivoire; MAM – Houston, Texas, and Ilorin, Nigeria; JJA – Helsinki, Finland, and Dakar, Senegal; and SON – Dhadnah, UAE, and Palangkaraya, Indonesia), with the selected sites having some of the strongest correlations for the respective seasons in the AERONET and NAAPS-RA datasets (Tables 1–4). Good agreement is shown between the NAAPS-RA and AERONET-generated Theil–Sen slopes at the selected sites with the largest differences occurring at the Lamto and Palangkaraya sites in which relatively fewer AERONET observations are available. These fittings are calculated for each grid and AERONET site and are used to generate the results in Figs. 7 and 8. The examples in Fig. 10 show the insensitivity of the Theil–Sen regression to outliers, while the correlation coefficient is quite sensitive to such values. In DJF, Beijing exhibits a large change in AOD with PW, as high-AOD events are more frequent at this location (Figs. 5 and 10). However, places like Houston, Helsinki, and Dhadnah have relatively smaller Theil–Sen slopes as a result of high-AOD events, with values around 1 occurring less frequently and not influencing the slope. For these locations, the range of frequently observed AOD events is much smaller (Figs. 5 and 10), resulting in small changes in AOD with PW. Although there is certainly scatter in the data points in Fig. 10, statistically significant trends exist. The scatter in the data points occurs more so at negative-correlation locations (Fig. 10), resulting in smaller correlation coefficients. While the relationships for both positive and negatively correlated locations are statistically significant and the Theil–Sen regression gives an overall trend, the scatter indicates that differences in AOD and PW relationships will occur from day to day. This is expected as the AOD–PW relationship is based on a combination of transport covariance and local meteorology–source relationships.

The global and seasonal patterns in the positive and negative Theil–Sen slopes are consistent with the correlation analysis results (Figs. 7 and 8). The biggest Theil–Sen slopes tend to occur where larger IQR ranges are present (Fig. 5), as was shown for the Beijing Theil–Sen slope example in Fig. 10. The largest slopes in both datasets are centered on Beijing in the DJF months, with values exceeding 1 cm-1. Beijing consistently has some of the largest positive changes in AOD with PW in the AERONET dataset for all seasons with values, including 95 % confidence intervals, of 1.1(1.0–1.2), 0.35(0.32–0.38), 0.46(0.43–0.51), and 0.26(0.22–0.3) cm-1 for DJF, MAM, JJA, and SON, respectively. The NAAPS-RA is largely consistent with AERONET for the DJF and MAM months, with corresponding values of 1.13(1.08–1.18), and 0.31(0.29–0.33) cm-1. Less sensitivity to PW is found in the reanalysis for JJA and SON, with corresponding values at Beijing of 0.13(0.11–0.14), and 0.18 (0.17–0.20) cm-1. This is likely due to an underestimation of haze formation within NAAPS, as with other global models (e.g., Sessions et al., 2015; Xian et al., 2019), and also possibly due to the underestimation of AOD in summer from NAAPS due to a low AOD bias in the assimilated satellite AOD datasets in the East Asian region (Eck et al., 2018). Large positive changes in AOD with PW extend through Asia, the Middle East, and northern Africa, all regions impacted by high-AOD events. As is the case in the correlation results, the strong dipole in slopes is clear over northern Africa with positive slopes to the north and negative slopes in the southern Sahel region. Likewise, negative slopes are mainly associated with burning regions, with the exception of southern Africa in JJA. Statistically significant correlations and slopes at high latitudes, particularly Antarctica, indicate aerosol–water vapor transport in the model since local sources are limited, although AOD and PW values are low (Figs. 5 and 6).

Recall that the scatterplot comparison of AERONET- and NAAPS-generated changes in AOD with PW is shown in Fig. 9b. Again, there is generally good agreement between the datasets, consistent with the correlation comparison. The signs of the slopes are the same with the exception of 14.6 %, 9.2 %, 12.3 %, and 8.1 % of sites for DJF, MAM, JJA, and SON, respectively. Sites where differences in sign are observed have weak correlations (Fig. 8d). The NAAPS-RA has a tendency to underpredict negative changes in AOD with PW relative to AERONET for SON months where peak negative slopes are generated from AERONET. This is also shown in Table 4 as well as the global maps in Fig. 7 where differences can be seen, particularly in the Sahel and Southeast Asia. At the Kuching site in Borneo in SON, the AERONET-generated slope is -0.37(-0.48 to -0.28) cm-1, with a reanalysis value of -0.11(-0.13 to -0.09) cm-1, likely due to strong mesoscale variability and poor constraints on biomass burning in Borneo (Reid et al., 2013; Wang et al., 2013). Additionally, this could again be due to satellite retrieval screening of smoke as cloud and NAAPS failing to simulate the highest AOD smoke events in Borneo, especially in the dry El Niño years such as 2015 (Eck et al., 2019; Shi et al., 2019). The reanalysis also tends to underpredict positive slopes for JJA at the AERONET sites where the largest slopes are observed. This difference is not restricted to a particular region but can be seen in East Asia, Africa, and Mexico City (Fig. 7). As an example, the Tamanrasset site in Algeria exhibits slopes in the AERONET data of 0.26(0.23–0.30) cm-1 and in the reanalysis of 0.07(0.06–0.08). Likewise, at the Lubango site in Angola, the slope is 0.27(0.21–0.34) cm-1 in the AERONET data and 0.13(0.12–0.14) cm-1 in the reanalysis. The Tamanrasset site is at 1377 m altitude in the Ahaggar Mountains, which is significantly higher than the surrounding terrain in the Sahara. The Lubango site in Angola is at 2047 m, also higher than a portion of the surrounding terrain. This terrain/altitude influence is likely a factor in the discrepancies. The differences in slopes for JJA are also shown in Table 3.

While the correlation and slope evaluation is used to define a seasonal AOD and PW relationship across the datasets, it is expected that variations in the aerosol and water vapor relationship will exist across air masses. As a result, a probability distribution evaluation is another useful way to examine the data. The seasonal evaluation of the probability distributions is included in Fig. 7, next to the correlation and slope results. The plots show the statistically significant difference in the mean for the PW distribution associated with high-AOD events (AOD values more than 1 standard deviation above the mean) and the full PW distribution. Red regions/sites indicate that the PW mean for high-AOD events is statistically higher than the full distribution mean (i.e., higher moisture levels). Blue regions/sites indicate a lower PW mean for high-AOD events (i.e., drier conditions). Regions or sites in white have no statistically significant difference. The spatial pattern in the probability distribution evaluation is similar to the correlation and slope analysis; however, the probability distribution evaluation highlights different regions than the previous analyses. For example, across all seasons, larger changes in PW for high-AOD events are observed in Argentina, South America, including at the CEILAP-BA (Buenos Aires, Argentina), with values of 0.88, 0.94, 1.01, and 1.00 cm for DJF, MAM, JJA, and SON in the AERONET dataset and values of 0.61, 0.35, 0.95, and 0.73 cm in the NAAPS-RA dataset. This is a region that is impacted by both local pollution and transported biomass burning (Resquin et al., 2018). Larger changes in PW for high-AOD events are also observed over northern Australia during MAM, which is consistent with peak bushfire season in the region. Larger changes in PW are also found over the United States and Canada, consistent with patterns in the correlation evaluation but with more pronounced values relative to other locations. ABF is generally the dominant aerosol type, with biomass burning from Central America and western US/boreal regions during the MAM and JJA seasons, respectively. Likewise, Eurasian boreal regions associated with biomass burning activity during JJA are more pronounced in the PW distribution evaluation. The peak in values in the Southeastern United States is found during the DJF season. During MAM and SON, the peak areas include most of the eastern United States, extending into Canada and Central America. However, regions that were more pronounced in the correlation and slope evaluation have smaller differences in mean PW for high-AOD events, Beijing being a good example of this. Based on AERONET, the difference in mean PW at Beijing is 0.21 cm, while the difference is 0.35 cm in the NAAPS-RA for DJF when the strongest correlations and largest slopes were found. However, sites like Stennis, Mississippi (Table 1) which had a much smaller slope than Beijing (0.04 cm-1 compared to 1.1 cm-1) have a much larger difference in mean PW, with an AERONET value of 1.08 cm and a NAAPS value of 1.16 cm. This is because the probability distribution evaluation is taking into account those infrequent, outlier events that do not affect the Theil–Sen slopes. Locations where the IQR is relatively small, such as the United States, Europe, Australia, and parts of South America and southern Africa, have greater differences in mean PW, despite having small Theil–Sen slopes, due to the impact of outlier events. For many of these regions, the outliers are associated with biomass burning, indicating that PW is a useful tracer for such events.

Like the correlation and slope evaluation, a comparison of AERONET- and NAAPS-generated differences in mean PW was conducted by season. Similar to the previous two comparisons, AERONET and NAAPS are in agreement in the sign of PW difference for most locations, demonstrated by the global plots in Fig. 7 and the scatterplots in Fig. 8c. The percentage of sites that have differences in sign between the two datasets is 8.9 %, 6.87 %, 5.86 %, and 4.15 % for DJF, MAM, JJA, and SON, respectively. These percentages are smaller than those of sites with differences in the correlation and slope analysis. However, like the previous evaluations, most sites that exhibit sign differences between AERONET and NAAPS had weak AOD and PW relationships (R<0.20, Fig. 8d), with the exception of some outliers in which small-scale features that cannot be resolved in the global model may be at play. The comparisons between AERONET and NAAPS-RA across the different evaluations indicate that NAAPS is generating AOD and PW relationships that are pretty consistent with the observational data. Although differences in magnitude are present, the direction of the relationships is very consistent, providing confidence in the use of the NAAPS-RA for further exploring the AOD and PW relationship, particularly in the vertical and accounting for hygroscopic affects.

In addition to calculating the full column-integrated AOD and PW correlations in the NAAPS-RA, the correlations were also evaluated by vertically integrating the extinction and specific humidity through previously defined pressure levels in the atmosphere that correspond to a boundary layer and the lower, middle, and upper free troposphere. This evaluation was conducted seasonally, like the fully integrated analysis, with results shown in Fig. 11. In addition to the global plots, histograms of the AOD and PW correlations for the full column and the vertical components of the atmosphere are shown in Fig. 12. It is notable that stronger positive correlations exist when looking at limited parts of the atmosphere compared to the fully integrated column. This is most evident in the global plots for ocean regions, particularly in the Southern Hemisphere, where correlations exceeding 0.5 occur compared to the fully integrated correlations that are on the order of 0.2. This result is not unexpected given that the vertical components of the atmosphere look different depending on things like vertical mixing, a local aerosol and water vapor source compared to a long-range transport event, and the relative humidity profile. Additionally, some regions exhibit stronger correlations in certain portions of the atmosphere. For example, dust-dominated regions such as the Sahara, Arabian Peninsula, and the Gobi and Taklimakan deserts have the strongest correlations in the MT. This is higher up in the atmosphere than expected, given for example, studies have shown East Asian dust heights to range from 1.9 to 3.1 km (Liu et al., 2019), and the typical description of the Saharan Air Layer (SAL) includes dust-laden air between approximately 850 and 500 hPa (Karyampudi et al., 1999), with several other studies identifying Saharan dust up to ∼5 km for summertime dust transport (Mortier et al., 2016; Veselovskii et al., 2016; Tesche et al., 2011). This indicates that the model may be transporting too much dust aerosol and water vapor higher into the atmosphere and that this transport is well correlated. Correlations over North America and eastern Europe are strongest in the BL to LT. Wintertime correlations over East Asia/Beijing are pretty consistent throughout the column. Negative-correlation regions associated with smoke aerosol, including the Sahel, southern Africa, and Southeast Asia, have the strongest correlations in the LT and largely disappear beyond this point. The shift in correlations with vertical location is also evident in the histograms (Fig. 12) when compared to the full column distribution. This is particularly the case for the lower and middle free troposphere, where the number of grids with correlations greater than 0.5 increases. Additionally, the shift of the correlations to mostly positive can be seen in the middle and upper free troposphere histograms.

The final consideration in this work is hygroscopicity. Although the effects of clouds on the AOD and PW relationship are also important to understand, this effect cannot be investigated using NAAPS since the model does not account for the processing of aerosol in cloud droplet, rapid gas-to-particle conversion in cloud droplets, or the high RH halo in the immediate vicinity of clouds. However, this should be considered in follow-on work. While relationships between AOD and PW have been demonstrated, this signal can be either from co-transport or a confounding relationship between enhanced PW and RH. The correlation between RH and PW is shown in Fig. 13 by season and for the boundary layer and parts of the free troposphere. The largest spatial variations in the PW and RH correlation occur in the boundary layer, as anticipated, with strong correlations found over Africa, extending into the Indian Ocean/India, located further north during JJA and further south during DJF, and similar patterns during MAM and SON. Other regions of high correlation in the boundary layer include the region off the coast of South America, parts of Australia, and limited locations in the tropical oceans. Beyond the boundary layer, the overall patterns are generally consistent throughout the vertical column, with strong correlations in the subtropics and tropics (>0.9) and some variations on the extent by season. In JJA, for example, this high correlation region extends further north, while in DJF the high correlation region extends further into the Southern Hemisphere. In this highly correlated region, hygroscopic growth is expected to be a significant driver in AOD and PW relationships when dust is not the dominant aerosol type. RH and PW correlations are higher over ocean regions than over land in the Northern Hemisphere, which should be impactful for sea salt aerosol and PW correlations. The impact of the RH and PW correlations on the AOD and PW relationship is shown in calculated seasonal relationships between “dry” AOD, which excludes the impact of hygroscopic growth, and PW in the NAAPS-RA in Fig. 14. This figure includes the correlations, the slope of the “dry” AOD and PW relationship, and the statistically significant different in mean PW for high “dry” AOD events. The removal of hygroscopic growth from the AOD calculation had the following outcomes on the resulting correlations: (1) the previously positive correlation was reduced in magnitude, (2) the previously negative correlation coefficient became more negative, (3) the sign of the correlation flipped from positive to negative, and (4) there was little to no change in the correlation. Regions such as the eastern United States and Europe fall into the first category, where positive AOD and PW correlations are found for all seasons, but the correlation coefficient is greatly reduced. For the eastern United States, peak correlation coefficients were in the approximate 0.6–0.7 range with hygroscopic growth and fell below 0.5 without it. This is especially true in JJA when RH and PW correlations are the strongest. Likewise, positive correlations in Europe are still present but weakened. In these cases, hygroscopic growth amplifies an existing positive relationship that is somewhat weak when evaluating seasonal data by correlation. In regard to the second category, this corresponds to regions dominated by smoke aerosol that previously exhibited negative AOD and PW relationships, such as peninsular Southeast Asia during the MAM months and insular Southeast Asia during the SON months. Additionally, increases in negative correlations are found for aerosol transport from Asia across the Pacific Ocean. In these cases, hygroscopic growth reduces an existing negative relationship between aerosol and water vapor. Ocean regions mostly account for the third category, where the correlation flipped from a weak positive to negative value. Regions that are dominated by dust, including the Sahara and Arabian Peninsula, fell into the fourth category as there is no hygroscopic growth for dust in NAAPS.

In regard to the slope of “dry” AOD and PW, the same categories apply with similar spatial patterns relative to the correlation analysis. Additionally, the same is true when examining the difference in mean PW for high “dry” AOD cases. In this case, it is found that (1) an increase in PW is still statistically significant, but the difference in mean PW is much less; (2) a decrease in PW for high “dry” AOD cases is still statistically significant, with a larger decrease when not considering hygroscopic growth; (3) the sign of the difference flipped from an increase in PW to a decrease, or the difference became statistically insignificant; or (4) the PW difference did not change much due to dust-dominated conditions. While the modeled differences in PW are statistically significant when excluding hygroscopic growth, they are small, with peak differences on the order of a few millimeters. The results here indicate that hygroscopic growth of aerosol plays an important role in the AOD and PW relationship. While PW is still a good tracer for AOD as shown in this work, it should be kept in mind that there is a difference in water vapor as a tracer for AOD and for aerosol mass. It is expected that the relationship between “dry” AOD and PW would be a closer representation of the dry aerosol mass to PW relationship.

In order to further understand regional differences in observed AOD and PW relationships, individual sites in which strong AOD and PW relationships were identified and that had several years of observational data available were selected for further analysis. These sites included (1) Tallahassee, Florida for Southeastern US pollution (Fig. 15); (2) Beijing, China, for Asian haze and dust (Fig. 16); (3) Izaña, Canary Islands, for Saharan dust (Fig. 17); and Alta Floresta, Brazil, for South American biomass burning (Fig. 18). For the four identified AERONET sites, the daily-averaged AOD and PW time series are examined for seasons in which correlations were found to be strong. This includes DJF for the Tallahassee and Beijing sites and JJA for Izaña. At these sites, the identified relationships between AOD and PW were positive. The Alta Floresta site, which exhibited negative AOD and PW relationships in the presented results, is further examined for the SON biomass burning season. The daily-averaged data are included since this is what was analyzed in the previous analyses. Additionally, the AERONET data, without any averaging, are further examined for individual cases from the site-specific time series for which peaks in AOD and/or PW were found. NAAPS-RA AOD and PW fields are also shown for the selected cases (Figs. 15–18).

At the Tallahassee site, the predominant aerosol type is ABF/pollution, and although AOD values are generally low during DJF (mean values in the 0.1–0.2 range, Fig. 3), strong AOD and PW relationships were found, with correlations of 0.74 in the AERONET dataset and 0.66 in the NAAPS-RA and PW mean differences around 1 cm for high-AOD events (Table 1). The daily-averaged AOD and PW time series for the 2018–2019 DJF season are shown in Fig. 15a. The time series indicate, consistent with the correlation analysis, that the daily-average AOD and PW generally move together. There are several joint peaks in AOD and PW that occur during the time period, and three selected cases are examined further, including 1 January, 7 February, and 17 February 2019, with these events identified in the Fig. 15a time series using red arrows. The AERONET AOD and PW time series for these three cases are shown in Fig. 15b–d, respectively. The 17 February case has the least data points available, making it harder to evaluate diurnal changes in AOD and PW; however, the 1 January and 7 February cases have a good number of data points throughout the afternoon and later into the evening. For these two cases in particular, the changes in AOD and PW throughout the day are generally consistent with each other, indicating that the AOD and PW relationships can extend to sub-daily timescales. AOD and PW plots for the three identified cases are shown from the NAAPS-RA in Fig. 15e as a means to assess the types of aerosol events that are impacting Tallahassee when coordinated peaks in AOD and PW are observed. For all three cases, coincident transport of AOD and PW is observed in the reanalysis fields, associated with a frontal system. This type of frontal transport was commonly found for events in which coincident PW and AOD peaks are observed at Tallahassee. As the DJF season in the Southeastern United States has significant frontal activity, this is likely an important factor in enhanced AOD–PW relationships during this season.

Beijing is an urban site that commonly experiences high-AOD levels related to pollution, as well as transported dust and smoke events. Wintertime events are notorious for exhibiting some of the worst air quality in the world for a major population center (e.g., Wang et al., 2014; Gao et al., 2016; Zhang et al., 2018). Additionally, Beijing has the benefit of a long AERONET data record, with measurements dating back to 2001. Like Tallahassee, the AOD and PW time series at Beijing are evaluated for the 2018–2019 DJF season (Fig. 16), with strong positive relationships identified in both the AERONET and NAAPS datasets with correlations of 0.71 and 0.76, respectively, and the change in AOD with PW exceeding 1 cm-1 (Table 1). As indicated in Fig. 4, ABF/pollution is the dominant aerosol type, with some dust present and much higher AOD values observed at this location (Figs. 3 and 5). In East Asia, pollution buildup often occurs under stagnant weather conditions, where a stable atmosphere leads to limited vertical mixing (Wang et al., 2014; Li et al., 2019); the monsoon is also an important factor in determining synoptic conditions. In particular, the East Asian winter monsoon (EAWM) has been shown to be a controlling factor in aerosol concentrations during the winter season (Li et al., 2016; Jeong et al., 2017). With a strong EAWM, reduced aerosol concentration occurs over northern East Asia, including Beijing, due to stronger northerly winds. In weaker EAWM years, increased aerosol concentrations occur in the north due to weakened winds and more stagnant conditions. The daily-averaged AERONET AOD and PW time series for 2018–2019 DJF are shown in Fig. 16a. Consistent with the previously presented evaluations, the daily-average time series for this particular DJF time period are well correlated with AOD and PW moving up and down together. As was done for the Tallahassee site, several peaks in AOD and PW were selected for further evaluation and are highlighted with red arrows in Fig. 16a, including the peak on 3 January and its subsequent decrease on 4 January 2019 and the peak on 20 December 2018 and its subsequent decrease on 21 December 2018. The non-averaged AERONET AOD and PW time series for these two cases are shown in Fig. 16b and c, respectively, with a zoomed-in view of 21 December on Fig. 16d. For both of these events, high AOD from ABF/pollution and high PW values are observed, with a subsequent drop-off the following day, which is well coordinated in the full dataset. A closer look at the data on 21 December 2018 (Fig. 16d), like the previous Tallahassee examples, shows consistent movement between the measured AOD and PW, indicating the presence of correlations on short timescales. For both of these events, the NAAPS-RA AOD and PW fields are shown for both the peak and subsequent drop-off in Fig. 16e. The movement of the large-scale air mass can be seen in both the AOD and PW fields. For the 3–4 January 2019 event, NAVGEM meteorological fields indicate weakened northerly winds due to a region of high pressure over the eastern portion of the continent, leading to stagnant conditions at the surface in Beijing and local pollution and water vapor build-up. On the following day, the high-pressure system moved eastward. As a result, the Siberian High northerlies were no longer suppressed, and a more typical wintertime circulation resumes, with the winds jointly pushing the aerosol and water vapor southward and away from Beijing. For the 20–21 December 2018 case, extensive multi-level cloud cover can be seen in both MODIS Terra and Aqua images on 20 December, indicating that this may be a case where clouds played a role in gas-to-particle conversion in the polluted air and/or enhanced particle humidification in the high RH fields associated with the clouds, consistent with the findings of Eck et al. (2018). As the air mass moves on 21 December, a joint reduction in both AOD and PW is observed at Beijing, demonstrating the impact of large-scale transport. Thus in the Beijing case, the overall regional weather patterns are an important factor in the AOD and PW relationship.

The third site that is examined is Izaña in the Canary Islands (Fig. 17). The Izaña site, which is located approximately 300 km west of the African coast, is particularly useful for evaluating aerosol and water vapor relationships for free tropospheric dust. It has also been noted in the community that against a Saharan Air Layer free subsidence regime, the infrared signals of Saharan Air Layer dust are quite small relative to co-transported water vapor (Gutleben et al., 2019; Ryder, 2021; Barreto et al., 2022). Nevertheless, forecasters find the water vapor signal useful in tracking dust events (Kuciauskas et al., 2018). Izaña is a mountain site located at approximately 2400 m, above a strong subtropical temperature inversion layer, which makes it ideal for monitoring free tropospheric plumes. In the summer months, frequent and intense Saharan air mass outbreaks in the subtropical free troposphere impact the site. Particularly large AOD dust storms were observed transporting Saharan dust across the Atlantic in the summer of 2020, including the so-called “Godzilla” dust event in June that has been examined in detail in previous studies (Francis et al., 2020). As a result, this time period was evaluated in further detail at Izaña, with a focus on several dust events. In the analysis conducted in this work, positive relationships between AOD and PW were found in both the AERONET and NAAPS datasets at Izaña during the JJA season, with the AERONET dataset having a correlation of 0.66 and the NAAPS-RA indicating a weaker correlation of 0.48 (Table 3). As noted for other sites, this difference may be due to altitude effects at the Izaña site which may not be captured in NAAPS. The daily-average AOD and PW time series for the 2020 JJA season at Izaña are shown in Fig. 17a. The AOD and PW are pretty well correlated, although there are PW peaks present without the presence of aerosol. As sources of aerosol and water vapor are different, this is not unexpected. Three of the joint AOD and PW peaks from the time series, as indicated by the red arrows in Fig. 17a, were selected for further evaluation. This includes 14 June, 18 July, and 31 July 2020 with time series shown in Fig. 17b–d, respectively. As was shown for the previous pollution cases, the AOD and PW are well correlated in the non-averaged dataset, showing the presence of correlations for dust events on timescales less than a day. The NAAPS-RA AOD and PW fields are shown for the 14 June 2020 case in addition to the associated drop-off of both AOD and PW on 18 June 2020, as well as for the 18 and 31 July 2020 events in Fig. 17e. For the 14 June 2020 case, CALIPSO indicates dust tops near Izaña at around 5 km. A cutoff low was present to the northwest of Africa and a subtropical high over the western coast of Africa, which resulted in increased dust generation and recirculation. The southwesterly winds transport the dust to Izaña on 14 June, and at the same time, moist ocean air shown by the PW fields is transported to Izaña as well, resulting in a spike in both AOD and PW at the AERONET site. On 18 June, the moisture and dust begin pushing south and west across the Atlantic, resulting in decreases in AOD and PW at the same time at Izaña. Similar examples of dust and water vapor co-transport are shown for 18 and 31 July 2020 cases. These results are consistent with previous studies which have indicated enhanced water vapor mixing ratios associated with the dust in Saharan Air Layer events (Marsham et al., 2008; Jung et al., 2013; Kanitz et al., 2014). Thus, Izaña is a good example of subtropical free tropospheric co-transport of water vapor and dust.

The final site evaluated was Alta Floresta, Brazil, for the 2019 SON season, when biomass burning is the dominant aerosol type. Here, there is a strong seasonal dependency: drier seasonal conditions are associated with a July–October biomass burning season. At this site and other regions of biomass burning, negative correlations between AOD and PW were identified in the presented evaluations. For Alta Floresta, SON correlations of -0.38 and -0.47 were generated from the AERONET and NAAPS-RA datasets, respectively (Table 4). The daily-average AOD and PW time series are shown in Fig. 18a. An important thing to note about this time series is that a clear downward shift in the PW fields occurs during October, consistent with monthly site climatologies from AERONET (3.61 cm in September, 4.15 cm in October, 4.5 cm in November). As fires are associated with dry conditions, the AOD fields decrease as the water vapor increases. This shift towards wetter and decreased aerosol conditions is driving the seasonal negative correlations for biomass burning regions. However, despite this overall shift, peaks in AOD and PW are generally positively correlated. Several such cases are highlighted in the time series, including 15 and 23 September 2019. The 9 September 2019 event is also highlighted in which the daily-averaged AOD peaks and the PW is at a low. For all three cases, the AOD and PW time series shown in Fig. 18b–d show good positive correlations, with AOD and PW changing in the 9 September 2019 case; this appears to be a more locally driven event with the extent of the smoke being more limited and the air being drier than the surrounding areas, suggesting a different air mass (Fig. 18e). Additionally, there is much more small-scale variability in the AOD and PW fields (Fig. 18b). For this type of event, the daily-average PW fields might not be as good of an indicator of what is going on with AOD. For the other two cases, the NAAPS-RA plots indicate larger spatial extent of the smoke with more moisture associated with the air mass. In this case, daily-average PW is a better indicator for large-scale smoke events. However, for all three smoke cases, AOD and PW are correlated on an event level. Thus, in this case the AOD–PW relationship identified in the previously presented AERONET and NAAPS-RA evaluations represents the end of the burning season, with wet-season onset in the middle of a “climatological season”. However, PW is a good positive indicator of AOD associated with smoke on an event level, consistent with what has previously been shown in the literature for case study evaluations.