Satellite Mapping of PM2.5 Episodes in the Wintertime A “Static” Model Using Column Water Vapor

: The use of satellite Aerosol Optical Thickness (AOT) from imaging spectrometers has been successful in quantifying and mapping high PM2.5 (particulate matter mass < 2.5 µm diameter) episodes for pollution 10 abatement and health studies. However, some regions have high PM2.5 but poor estimation success. The challenges in using Aerosol Optical Thickness (AOT) from imaging spectrometers to characterize PM2.5 worldwide was especially evident in the wintertime San Joaquin Valley (SJV). The SJV’s attendant difficulties of high-albedo surfaces and very shallow, variable vertical mixing also occur in other significantly polluted regions around the world. We report on more accurate PM2.5 maps (where cloudiness permits) for the whole-winter period in the SJV, Nov 19, 15 2012–Feb 18, 2013. Intensive measurements by including NASA aircraft were made for several weeks in that winter, the DISCOVER-AQ California mission. We found success with a relatively simple method based on calibration and checking with surface monitors and a characterization of vertical mixing, and incorporating specific understandings of the region’s climatology.We estimate PM2.5 to within ~7 µg m –3 root-mean-square (rms) error and with R values of ~ 0.9, based on remotely 20 sensed MAIAC (Multi-Angle Implementation of Atmospheric Correction) observations, and that certain further work will improve that accuracy. Mapping is at 1 km resolution. This allows a time sequence of mapped aerosols at 1 km for cloud-free days. We describe our technique as a “static estimation.” Estimation procedures like this one, not dependent on well-mapped source strengths or on transport error, should help full source-driven to PM2.5, both slope and intercept effects, in a daily calibration allows regressioin to exploit portions of the PM2.5 vs. f (AOT/CWV) scatterplots that reveal proportionality. High-spatial-resolution estimates of the 11 AM–3:30 PM PBL heights for momentum may as helpful as CWV when available and when the PBL estimation has been examined for accuracy; this could be explored. Such PBL data is not available for the whole MODIS-Aqua period.(2004–present), while CWV is. The DISCOVER-AQ (Deriving Information on Surface Conditions from Column and Vertically Resolved Observations Relevant to Air Quality) mission conducted its second field deployment in the California San Joaquin Valley region during January and February 2013. The mission’s overarching goal is to better understand how remotely-sensed column measurements can be used to diagnose near-surface air quality. To achieve this objective, the DISCOVER-AQ sampling strategy requires extensive probing of the vertical structure of the lower troposphere as it relates to both trace gases and aerosols. This strategy was implemented by using the NASA P-3B aircraft to perform three circuits of spirals from 0.3 to ~3 km over 6 air quality monitoring ground sites at three different times of the day (mid-morning, midday, and mid-afternoon local time). In addition, missed approach maneuvers were performed at 7 airports along the flight path (5 of which were located near profile sites), which provided profile data from as low as 25 m up through the 0.3 km bottom limit of the spirals. A total of 170 spirals and 157 missed approaches were flown, which generated detailed vertical distributions for a large variety of trace gases, aerosol properties, and meteorological variables. I n t r o du c t i o n a nd O v e r v i e w o f D - A Q C A n The DISCOVER-AQ (Deriving Information on Surface Conditions from Column and Vertically Resolved Observations Relevant to Air Quality) mission conducted its second field deployment in the California San Joaquin Valley region during January and February 2013. The mission’s overarching goal is to better understand how remotely-sensed column measurements can be used to diagnose near-surface air quality. To achieve this objective, the DISCOVER-AQ sampling strategy requires extensive probing of the vertical structure of the lower troposphere as it relates to both trace gases and aerosols. This strategy was implemented by using the NASA P-3B aircraft to perform three circuits of spirals from 0.3 to ~3 km over 6 air quality monitoring ground sites at three different times of the day (mid-morning, midday, and mid-afternoon local time). In addition, missed approach maneuvers were performed at 7 airports along the flight path (5 of which were located near profile sites), which provided profile data from as low as 25 m up through the 0.3 km bottom limit of the spirals. A total of 170 spirals and 157 missed approaches were flown, which generated detailed vertical distributions for a large variety of trace gases, aerosol properties, and meteorological variables.


Introduction
The San Joaquin Valley (SJV) is an important agricultural area, characterized by poor air quality (Figure 1). The SJV gives an example of a region with frequent air pollution episodes, challenged by difficulties as varied particle characteristics with hard-to-quantify sources from domestic burning and spatially distinct ammonia and nitrate precursors. The 60,840 km 2 area (with approx. 4 million residents) is located southeast of San Francisco, between the Coastal Mountain Range to the west and the Sierra Nevada Range to the east (Sorek-Hamer et al., 2013). Figure 1 describes the particularly high particulate pollution characterizingthe San Joaquin Valley. Previous studies in this region reported a range of correlations between satellite-borne AOT and daily/ hourly collocated ground PM2.5 measurements in this region. Using linear tools resulted in little or no correlation (Engel-Cox et al., 2004;Ballard et al., 2008;Justice et al., 2009), while applying non-linear methods improved the correlation to inland Mediterranean valleys, with the aim of allowing air pollution professionals to understand particulate episodes and to improve sources and simulation details (e.g., transport error) for source-driven models. Goals of the Dalhousie group are to improve annual average exposure: they see that as the principal driver for health effect (Van Dankelaar et al., 2010. A main goal of NASA's MAIA (Multi-Angle Imager for Aerosols) mission is similarly deliver new data for a each-day mapping of PM2.5 exposure sufficient for full studies of health effects (Diner et al., 2018, https://maia.jpl.nasa.gov/). In pursuit of that goal for the MODIS Aqua dataset, we will indicate some preliminary, meteorology-based ideas for estimating high aerosol concentration when clouds prevent the use of remote sensing data.

100
Due to the complex meteorology of the San Joaquin flows and uncertainties surrounding the sources of ammonia, nitrogen oxides, and residential-burning smoke, we attempt to separate out some certain aspects of complex 3-d source-driven modeling (Bey et al., 2001, Nolte, 2015, Appel et al., 2017, Friberg et al. 2018) with a "static model" which does not attempt to simulate transport, but rather uses observational records related to vertical mixing and AOT. The spatial maps produced can give a more detailed check on the 3-d process modeling. They also 105 allow the whole MODIS record of winters to be analyzed efficiently so as to reveal patterns and trends. We emphasize this and further extension the interpretation of satellite radiances, attempting to remain close to physical interpretations by using both MAIAC AOT and CWV retrievals. MAIAC Column Water Vapor (CWV)  retrievals have been quite acceptability validated with the AERONET CWV measurements in higher CWV environments (Martins et al., 2017(Martins et al., , 2018. It has not been previously recognized as a tool for improving 110 ground PM estimation and in particular, in the SJV.

115
The Multi-Angle Implementation of Atmospheric Correction (MAIAC) is an operational algorithm developed for MODIS Collection 6 (C6) data (Lyapustin et al., 2011a,b). This algorithm applies a dynamical time series technique to derive the MODIS surface bidirectional reflectance factor and atmospheric retrievals at a 1 km resolution, such as AOT, and CWV (Lyapustin et al., 2008;2011b). MAIAC AOT retrievals present an expected error within 15% and relatively good correlation coefficient (R) with AERONET measurements in the study area 120 .
MAIAC data has been used from both Terra and Aqua satellite with a daily overpass at ~10:30 and ~13:30 local sun time (+ ca.1.5 hours), respectively. Data has been obtained for the period of Winter 2012-2013(November 2012-April 2013. We surveyed this entire period and included, for estimation, all wintertime high-PM2.5 episodes for this specific winter, selecting November 19-February 18, as described in several later figures (discussed in context: Figures 3, 4, and 7).

AERONET AOT and CWV
The factor in the denominator, M (for "magnification") describes the relationship of the optical extinction to "dry particle mass smaller than 2.5 µm aerodynamic diameter" which is the motivated definition of PM2.5. (PM2.5 165 also has a definition by a United States "Federal Reference Method" which is formulated to approximate the physical definition as closely as possible.) The factor M then is a function of particle composition and the extinction coefficients bExt associated with the components, one of which may be largely absorbed water. Particle composition and ambient relative humidity, RH, then interact with each other to determine the water content. It is significant that RH is a function of temperature and therefore altitude, with highest RH at the top of a well-mixed layer.

170
This work emphasizes and attempts to exploit features of regional aerosol haze palls that parallel features of aerosol mass and a different measure of water vapor.  Figure 2 illustrates a conceptual idea of the fair-weather simulation we focus on. Both regional particulate 175 pollution and water vapor originate from the Earth's surface. Each tends to create relatively well mixed layers over several days, transported most significantly by a repeated daily cycle of mixing. The mixing of momentum is most active from just before noon through the mid-afternoon, creating an afternoon mixed layer, and water vapor and aerosol most typically mixes well up to this layer. Turbulent mixing depths vary from day to day, and these can create lofted layers of pollution cut off from the surface on the day of AOT and CWV observation. Flows in the San

180
Joaquin can be greatly influenced by the nearby mountains, with flows day and night promoting some upslope transport of material which can recirculate, detached from mixing on following days. Consideration of subsidence of air into the San Joaquin mixed layer suggests a flow-through time for aerosol and water of 2-3 days for some situations (Caputi et al., 2018). Mixing of entrained and mixed layer air allows for continued accumulation of pollutant aerosol in the valley as Figure 2 shows.

185
Particles and water vapor are emitted and accumulate in the same region, and they are mixed similarly each midday and afternoon by convective stirring. The height of mixing can be determined by variations in the buoyancy flux from the surface and varying vertical subsidence velocities, responding to larger scale weather patterns, during successive days. Figure 2 does not show the effect of particle transport or water vapor transport for a specific 190 location, but the PBL top, which is strongly controlled by local heat and water vapor fluxes at the surface.
If the mixing height is lower on succeeding days, then any water vapor and any particles at the top of the mixed layer are trapped in an "elevated layer" which does not mix to the surface. Other common ways in which elevated layers can be formed are mixing along the side of the valley (small -scale anabatic and katabatic winds) and by differential transport, i.e., wind shear. Fires, power plant plumes, and long-distance synoptic transport can form 195 layers that are quite separated at higher altitudes in the troposphere. Eventually there is removal of both species. Wet removal of particles is particularly effective, and the specific humidity of the air is very effectively removed by the condensation accompanying cooling and rising, according to the Clausius Clapeyron equation. Similar processes then limit the vertical spread of particles and specific humidity.

200
Water vapor molecules also accumulate in the atmosphere over a period of several days (typically a somewhat longer period), and both aerosols and water vapor are cleared from a particular place by cloud removal processes (venting, rainfall)  where CWV is in g / cm2 , @AB ( ) and ̅ @AB ( , ) correspond to the vertically distributed water vapor and appropriately average water density of the mixed lay. CWV is available from the MAIAC analyses yielding AOT.
Making the assumption that the heights are the nearly equivalent for water vapor and aerosol, we may write PM2.5 is calculated at EPA reference temperature (25 C) and pressure (1 atm), water vapor quantities in g cm-3 and ∆ ? @AB is in cm.
Work reported by Shook et al., 2018, described  We found in ensuing work that approximating ̅ @AB ( , ) by @AB ( = 0, ) added only a small amount to the variance explained by the regression given other limitations of the approximation.
(Possibly relative humidity effects or the correlation of water density with temperature could be complicating 220 correlated factors.) We calibrate the relationship (AOT) using data at official PM2.5 stations, and make the calibration daily. It is our observation that varies only over a small range when there are several MODIS observations on the same day, and that it varies in a limited way between neighbouring stations in a local region. The definition of "region" is based on that similarity, and it suggests similarity of∆ #$ and M, i.e., similar aerosol characteristics and boundary 225 layer behavior. This similarity does not apply when the wind shifts greatly between times or between stations, e.g.
when a front passes. Fortunately for our understanding of pollution episodes, frontal passage days tend not to have high PM2.5 .
We distil these understandings when we formulate a regression equation where the subscripts i describe "instance" or calendar date, and the subscripts s describe "station," so that AOT and 230 PM2.5 form a two-dimensional table.
Given the independent nature of i and s, the regression must be solved by "mixed effects" methods described below. The subscript s need only be independent of i, so later we will use it to denote "situation" or the hour of the day when there are many observations made at one station on one day i. It is not assumed that the consecutive order of the day observations necessarily describe any continuity in i. Observations show that there is often continuity, but 235 that the continuity is quickly broken when frontal passages or rain affect the region. surface monitors. These we will call "elevated layers." In-situ measurements on aircraft and also lidar measurements from ground lidars looking downward from aircraft (Sawamura et al., 2017), and satellite lidar (CALIPSO) reveal aerosol layers with significant optical thickness above the mixing layer. Similarly, airborne measurements in the DISCOVER-AQ intensive measurements of 2013 suggest a fraction of water vapor lies above the mixed layer for water. Allow these portions of total AOT and CWV layers to be quantified as AOTe and CWVe ('e' stands for 260 elevated). There can be several individual layers. AOTe and CWVe refer to the total amounts of extinction and water vapor mass. Thus, there is an approximate equation upon which to base regression estimation: While elevated layers of water and aerosol are common, we will see that it appears that this regression equation   Figure 4(a).

Observations and an Overview of Pollution Episode Trends
In this section, we will show how components of a mixed effects model that utilize CWV contribute to its 275 explanatory power. We examine the relationships and predictive ability for PM2.5 observed at SJV measurement stations for the winter season encompassing high-pollution periods, Nov 19, 2012 to February 17, 2013. Stations from Bakersfield in the South to Stockton in the North were included. Figure 3(b) shows several episodes affecting most of the Valley; one period with more stations reporting includes the DISCOVER-AQ period. This period has additional P3-B aircraft data which motivated this work, but are too lengthy to describe in this publication.     inclusion of the and variables suggests an over-fitting. Mixed effects convention emphasizes these "main effects" separately and therefore specifies there must be a single linear constraint on the terms such as and ; also, and . Importantly, in Section 6 and certain figures below, we describe the (more intuitive) combination of main and random effects, e.g. we graph ← + and ← + .

Which Information Contributes to PM2.5 Maps
Using the MAIAC 1x1 km estimates for each day for the location of each aerosol monitoring station 295 and the PM2.5 measured at overpass time for that day, we may solve the estimation equation, Eq. 6.
The complete simulation of PM2.5 measurement at all stations where MAIAC data allowed is shown in Figure 4(b). The technique can be used for all years and the whole area of the SJV where MODIS data is available. We used the complete model as described in Equation 6, "slopes and "intercepts", but without any time-independent spatial variation allowed ( \ ). Three features deserve immediate comment. First, 300 there are patterns of gradual increase of PM2.5 up to 45-80 µg m -3 followed by relatively sudden decrease to levels near 5 µg m -3 . Second, the regression technique using AOT / CWV, as estimated individually for each day, captures the variation rather well, for all days where estimates can be made. MODIS observations were not available, almost always due to cloud cover. We expect that these are readily explained in terms of weather phenomena especially typical of the American West during wintertime. Pollution episodes are ended with the approach of warm fronts with high clouds, followed in a few days by the cleansing effects of rain, air mass replacement, and higher wind. We will return to this topic later. To understand what information is used by the technique and how important is that information, based on the series of regression estimates we present; we argue that there is a cumulative aspect to explanation. For example, when we include one statistical variable, e.g. α c , describing variation by day but constant for all stations of the day, then the regression with AOT/CWV becomes much more informative. A general relation describing the slope of PM2.5 with AOT/CWV becomes more useful when an appropriate intercept (offset) is provided.

315
Consider both Table 1 and Figure 5 as they describe cumulative effects of adding information. First, we note that AOT alone is not very informative about PM2.5. This would seem to follow naturally from Equation 1, since variations in mixing depth and composition are not considered. Figure 5a show many station observations with high PM2.5 but low AOT, and vice versa. Slight but significant improvement is made when column water vapor, CWV, is introduced to provide some information on mixing depth and dilution. R improves to 0.48 but the remaining error 320 is nearly as high. Still, some linear relationship begins to show for perhaps 60% of the data.
A side comment regarding significance: R and remaining RMS error (in µg m-3) are shown in Table 1. We also performed two other tests not tabulated. An analysis of the Kuhlbach-Liebler divergence (Hastie et al., 2009 other examples of regression, they show similar trends as R and RMS error, i.e. accuracy becomes increasingly hard to improve as R increases. Another test was leave-out-one cross validation (Hastie et al., 2009). Each individual station was omitted, and the regression based on the remaining stations was tested against observations at that station. The cross-validated mean squared error was about 7.8 µg m -3 at most for the most informative regressions shown. Now consider a popular alternative to the use of satellite data. The third regression shown estimates of satellite data to particulate estimation, this has been shown to surpass, or at least approximate the only α c , i.e, assign a single PM2.5 estimate for each station based only on the individual day. Color-coded maps of PM2.5 drawn for a region have a single color which varies from day to day. In many applications of satellite data to particulate estimation, this has been shown to surpass, or at least approximate the results of use of AOT (Sorek-Hamer et al., 2017). R ~ 0.78, RMS error ~ 10 µg m-3. Its success emphasizes the regional similarity of conditions defining PM2.5 concentrations, and their extensive spatial correlation. An explanation is that respirable PM2.5 is defined by daily weather and orientation to major sources.ß a b c d e f Once the regional similarity of pollutant conditions is recognized, it becomes appealing to combine 340 information. The fourth estimate, Figure 5(d), does just this and shows a notable increase in R, 0.88, and decrease in     An appealing alternative is to estimate only slope variations, β c . This is nearly as useful as estimating just α c , R ~ 0.85 RMS error ~ 10 µg m -3 . Each is useful. Do the two parameter estimations give distinct information?

345
Estimation of varying offsets β c and sensitivities α c does indeed help, reducing the variance by another 10%.
Combining the use of AOT, CWV, and individual daily intercepts and slopes yields R ~ 0.90 and RMS error ~ 6.72 µg m -3 . Nevertheless, Figure 5(e) shows that certain stations have persistent deviations from the general swarm of points, Tranquility (pale green) is predicted high and Porterville and neighbors (red), are predicted low. This analysis of residuals suggests that there may be spatial variations that can be specified for our stations, γ d , but 350 are general enough that they can be extended to maps. For this publication, we attempted a very simple variation, an east-west variation (longitude). This did improve the scatterplot for most stations, especially when considering values above ~10 µg m -3 . RMS error decreased slightly to 6.48, and the R estimate also rose slightly, to 0.911. These changes are close to the range of sample variability. The maps shown in Figure 6 also show more convincing (subjective) agreement in magnitude and pattern. Nevertheless, many of the highest observations are underestimated by about 20%.
We used CWV rather than the RAP planetary boundary layer height for momentum, 11 to 15:30 local time.
This was available in the 2012-2013 winter at times within a half hour of overpass time, However, this PBL height is not always recorded in the high-resolution RAP archive. We compared a regression very similar to the most detailed regression of With this, the R value was 0.917 and the RMS error was 6.25 µg m -3 ; these are only insignificantly better than the CWV-based estimate R of 0.912; the RMS error was 6.43 µg m -3 . Mid-afternoon PBL depth is consequently useful. However, the CWV-based estimate may be used with all years of the MODIS data, while the best-available meteorology for PBL depth varies considerably, as high-resolution NOAA models advanced through the years. 365

6.
Results: Maps of Estimated PM2.5 The major purpose of this work, viz. to combine AOT, CWV, and daily calibration in order to allow maps of estimated PM2.5 for all regions where MODIS can provide optical thickness data. Results using the full model with α c , β c , and γ d are shown (Figure 5f). Out of the 42 days in the calibration set, we consider 6 days of single major air pollution episode during middle of January, 2013, a period that was largely sampled by the DISCOVER-AQ ground

Intensification of PM2.5 Episodes: Pollutant Accumulation vs Confinement
The well performing mixed effects models (equations 5 and 6) led us to examine the repeated development of air pollution episodes to a maximum, striking patterns seen in Figures 3, 4, and 6. How did the independent values 390 for various models in Table 1 vary within episodes and between episodes? Our description of the development leads to some answers in Figure 7.  tops; Afternoon PBL depths and mixed layer depths decrease day by day until a depth of 300-400 m is reached.
( Figure 7d). Escape from the Valley may slow, allowing accumulation of pollution from within the region or from upwind. This further increases the surface PM2.5. Relatively local sources add to both AOT and PM2.5, and can transport them 50-100 km downwind, occasionally from east-valley sources to west-valley pollution hotspots (the 425 map of Figure 6d). Both subsidence and surface buoyancy flux are broad-scale weather phenomena (~300 km) , and so AOT-to-PM2.5 relationships are similar on a given day with a given history of weather. Finally, warm-frontal rain approaches the region.
An examination of HSRL2 data for the DISCOVER-AQ period (Sawamura et al., 2017) suggests that there can be considerable vertical variability of aerosol extinction; the fact that AOT tends to average the whole afternoon 430 mixed layer allows our generalized description to hold nevertheless.
Finally, we venture some ideas for filling in afternoon PM2.5 on days when MAIAC did not allow mapping to 2 m s -1 or higher. This behavior is most clearly observed in their graphs for the period January 23-January 27.
Similar behavior is observed in the period February 6-February 11, although the episode has more complex increase 445 than in the earlier, most intense episodes. The short spike up to 80 µg m -3 on the night of February 10 is not explained, and not reflected in the afternoon-only data of our Figure 7. Nevertheless, the averages shown by Young in Figure 2e do repeat the general observation that daily average PM2.5 and afternoon PM2.5 do tend to correlate well. For best-estimate maps of PM2.5, we suggest that the end-of-retrieval values of PM2.5 reduce gradually over a day or two. Maps of precipitation (e.g. from radar or other analyses) allow more detail. Estimates for a region 450 should then fall to ~7 µg m -3 whenever sustained winds rise to > 2 m s -1 from the NNW or > 3 m s -1 from any direction. Such wind speeds are held to mark air mass replacement (e.g, frontal passage). These ideas remain suggestions since our analysis for a single winter may not provide enough instances. The whole Aqua MAIAC period is available, but currently beyond NASA's resources.

Variation of Random-effects Model Parameters
The preceding section gives some background so that we may understand the parameters for the random effect model. We will discuss the full Equation 4; results with mild spatial dependence (Equation 5) are very similar. The  the mixing depth changes little. The following weather episode is notable for high and quite variable AOT (Figure   7e), and the fitting procedure does well.

490
At this point, concerns about the quality of the CWV estimate should be addressed. In our analysis of the difficult San Joaquin Valley, MAIAC CWV can be frequently low compared to AERONET CWV, some error can arise from the presence of clouds in neighboring footprints. In the figures and results shown CWV was based on the MAIAC data interpolated and extrapolated where cloud-contamination made the retrieval of lower accuracy . Figure 6 shows some small-scale variability .RAP analyses of CWV could also be used at 495 their 13-km model-imposed width with similar results, since CWV does not vary as rapidly spatially as AOT. A better direct use of the MAIAC CWV could uses spatial averaging with a width of 3 to 6 km. Random errors in the MAIAC CWV due to the low radiances used would be reduced; considerations of source patterns suggests that CWV might not truly vary at such small scales. Improved PM2.5 values could result. We are implementing this averaging.

500
As understanding of MAIAC CWV improves, its role in determining daily AOT-to-PM2.5 relationships should improve; calibration of MAIAC using sun-photometer measurements can be useful in the meantime (Just et al., 2019).
Note also that assimilated CWV from the National Weather Service models is constrained empirically, and so not as reliant transport descriptions as is aerosol. Here are some constraints surface-station humidity measurements constraing CWV below 0.4-1 km, thermal-radiation sounders on the GOES (Geostationary Operational

505
Environmental Satellite) satellites describe water vapor partial above that; radiosonde and GPS humidity sensors give further constraint. This allows GOES AOT estimates to be used with assimilated CWV, even though GOES lacks a reflective water vapor channel (S. Kondragupta, personal communication, 2018).

Conclusions 510
Goals: We sought broadly applicables methods to estimate PM2.5 maps from satellite AOT for very polluted regions poorly described by satellite data. Ths study focusrd on the whole polluted winter season of the San Joaquin Valley (SJQ), November 19, 2012 to February 18, 2013. We sought to fulfill the overarching goal of the whole DISCOVER-AQ mission-to find general relationships between extended satellite data observations and surface air 515 pollutant concentrations and to evaluate their success. We found success with a simple methodology that follows the meteorology of regions like the SJQ. This success recommends an approach to the remote-sensing to PM2.5 analysis, investingating important pollution regions in terms of their meteorology and sources, but carrying over methods from similar regions. For example, the Po Valley of Itally and the Northern Gangetic Plane of India may respond similary to analyses based on detailed mixing height data and related distribution indicators.

520
Direct results: We found that a combination of information utilizing (1) optical depth, (2) measures of vertical dispersion, e.g. CWV, and (3) daily calibration of PM2.5 to predictors produced significantly better quantification of PM2.5 than a competitive no-satellite-use method which we named "regional correlation" since it produces unfeatured maps of PM2.5 which vary only from day to day. Our maps of estimated PM2.5 extend for all cloud-free periods November 19 2012 to February 18, 2013, essentially the whole pollution season for this winter. For that 525 whole period, this first published attempt found good predictive value of R ~ 0.9 and rms error of 6.5 µg m -3 . Crossvalidation suggested rms error of ≲ 7 µg m -3 . Analysis of residuals suggested that better rms errors could be achieved if further workallowed for sub-regioality (use of smaller regions or a geographic characterization incorporating some spatial variation). Local variations in PM2.5 on the order of 1-3 km were noted using our method, but only when particulate accumulation could occur along-wind. Still, in order to estimate PM2.5 at ≲ 1-km 530 scales, we expect that it will be necessary to use refined geographic information system methods (Kloog et al. 2014).

DISCOVER-AQ comparisons advisable:
Our analyzed winter 2012-2103 period did include the more limited DISCOVER-AQ / California-2013 airborne-intensive study period, primarily focused on the area around Fresno. Analysis of that intensive suggested ideas (Shook et al., 2013) that motivated this work.The shorter 535 DISCOVER-AQ period does deserve more detailed comparison to our results. Aircraft in-situ profiles of gas and particle composition, lidar profiles, very detailed surface measurements of particulate composition, and source-andtransport modeling all deserve comparison. The distribution of atmospheric particles and precursor gases is more complex than this work might suggest. Somehow averaging appears to allow our general methods. The development of concepts and the length of this workdo not allow for such comparison. We hope that research will 540 be encouraged.

Usefulness of Column Water Vapor:
A major finding was that the usefulness of CWV does not become apparent unless there is daily calibration of the AOT/CWV relationship to PM2.5. We attribute this primarily (a) details of CWV: e.g., CWV's dependence on mixed layer temperature on the timescale of days, (b) to CWV above the mixed layer for aerosols, presumably responding to other H2O sources upwind, and (c) variations in 545 composition: the relation of PM2.5 to light extinction. We believe that allowing for a full linear relationship each day for AOT/CWV to PM2.5, both slope and intercept effects, in a daily calibration allows regressioin to exploit portions of the PM2.5 vs. f(AOT/CWV) scatterplots that reveal proportionality. High-spatial-resolution estimates of the 11 AM-3:30 PM PBL heights for momentum may as helpful as CWV when available and when the PBL estimation has been examined for accuracy; this could be explored. Such PBL data is not available for the whole 550 MODIS-Aqua period. (2004-present), while CWV is.
Accompanying insights on pollution episodes. We found that this approach allowed a broad description of the buildup of six air pollution episodes and the balance of the roles of accumulation of pollutants versus limited vertical mixing. Episodes were as in earlier descriptions (Watson and Chow, 2002). Each appears important in different phases of reptitive PM2.5-increase cycles. PM2.5 to AOT relationships suggest a few days residence time 555 for particles (actually prticulate extinction) in the Valley. The first 1-3 days after MODIS described full cloud cover could still show high, slowly decreasing PM2.5. Unpublished analysis (see Young et al. 2018) suggests that this high PM2.5 dropped preciptously when surface winds rose to > 4 m s -1 from a quadrant centered on the NNW.
Best-estimate extensions to cloudy periods of the remote-sensing-based record can be made using the typical meteorology of the San Joaquin or presumably other areas, and verified by extensive checks. Widely available data Role of "Static" models: Our estimation approach aimed to avoid the use of modeling driven by source estimation and transport simulatoin. Principally we wished to provide dataseta that allowed independetn comparison to such three-d atmospheric chemistry models (e.g., Friberg, et al., 2018). When we used RAP-model CWV rather 565 than spatially averaged or calibratee (Just et al., 2019, manuscript in progress) MAIAC CWV, that goal was not fully reached, although RAP CWV is strongly constrained by surface, satellite, snd other observations. An aspirational goal is to provide an economical, accurate, and calibrated estimation of PM2.5 for the whole MODIS Aqua period to date, and then beyond. The opportunities to use MISR,VIIRS, MAIA, and even geostationary imaging are appealing! 570 11.

585
Included below are three portions of a poster presented by Michael Shook at the American Geophysical Union (Shook et al., 2013) soon after the DISCOVER-AQ measurements in California. That work gave a syncretic overview of many trace species measurement besides water vapor which motivated our investigation, as For some profiles, the BLHT could not be identified.
These cases usually had one of three problems: •  • Afternoon profile BLHTs seem to be bimodal First half of the campaign: BLHTs from 0.3 to 0.6 km Second half of the campaign: BLHTs from 0.5 to 0.7 km • 52 of 56 afternoon profiles had clear BLHTs Acknowledgements: This research was funded by NASA's Earth Venture-1 Program through the Earth System Science Pathfinder (ESSP) Program Office. We wish to thank the ESSP Program Office for the support. We also would like to thank the pilots and flight crew of the NASA P-3B through the NASA Wallops Flight Facility for their support and important contributions.

A43A-0228: Daily Evolution of Boundary Layer Properties based on NASA DISCOVER-AQ Airborne Profiles over the California San Joaquin Valley
Composite profiles of different constituents scaled by the profile BLHT and the average concentration of the constituent within the boundary layer. The dotted, colored lines represent individual profiles, and the solid black lines represent the median profile for that constituent and time period. Z = pressure altitude, Zi = profile BLHT, C = constituent concentration, and Ci = average constituent concentration in the boundary layer Example daily flight track showing the path of the P-3B and the repeated spirals at each of the six profile sites. The aircraft flew from Bakersfield to Porterville and continued around the circuit clockwise back to Bakersfield. Each flight day usually included three such circuits.
To visualize BL variability and vertical gradients, composite profiles for eight different constituents were created. Constituents were chosen to represent a variety of lifetimes and production/removal processes.

Conclusions:
• In addition to boundary layer heights increasing throughout the day, afternoon boundary layer heights were also higher in the second half of the campaign than they were in the first half.
• For many constituents, profile fluctuations (i.e. BL standard deviation-to-average ratio) decreased throughout the day, probably due to increased mixing and decreased stability in the BL.
• Only H 2 O and aerosol scattering appeared to have a consistent non-zero vertical gradient. CO 2 consistently had zero vertical gradient, and the other constituents had too much variability to define a consistent gradient.
However, for all other parameters besides O 3 , median trends were usually negative, and by the afternoon almost 75% of trends were negative. These results suggest that the BL was not always well-mixed.

Future Investigations:
• Incorporate balloon-borne measurements from Huron and Porterville to refine current BLHTs and potentially to help identify additional BLHTs at those sites • Find the cause of the higher afternoon boundary layer heights later in the campaign, possibly through backtrajectory analysis, and analyze its effects on BL concentrations • Evaluate any trends in boundary layer height or vertical gradients among the six profile sites For some profiles, the BLHT could not be identified.
These cases usually had one of three problems: • Suspected BLHT was near or below the bottom of the profile The mission's overarching goal is to better understand how remotely-sensed column measurements can be used to diagnose near-surface air quality. To achieve this objective, the DISCOVER-AQ sampling strategy requires extensive probing of the vertical structure of the lower troposphere as it relates to both trace gases and aerosols. This strategy was implemented by using the NASA P-3B aircraft to perform three circuits of spirals from 0.3 to ~3 km over 6 air quality monitoring ground sites at three different times of the day (mid-morning, midday, and mid-afternoon local time). In addition, missed approach maneuvers were performed at 7 airports along the flight path (5 of which were located near profile sites), which provided profile data from as low as 25 m up through the 0.3 km bottom limit of the spirals. A total of 170 spirals and 157 missed approaches were flown, which generated detailed vertical distributions for a large variety of trace gases, aerosol properties, and meteorological variables. To more clearly identify trends in BLHT from day to day across the campaign, profiles were separated by starting time into three groups: 0800-1030, 1030-1300, and 1300-1600 PST.
These times loosely correspond to the start and end times of the three circuits. The distributions of the BLHTs in these time intervals were then analyzed. The distributions and time series of the BLHTs for each interval is shown below, along with an example profile from that interval. Constituent vertical profiles are colored by bearing from center of spiral.

Morning Profiles (0800-1030 LT)
• Morning BLHTs were consistently low (about 0.35 km or less) • Aircraft was often unable to get low enough to see a clear transition into the boundary layer; only 20 of 61 morning profiles had clear BLHTs • Not enough sample points to know if high outliers were part of a second mode

Afternoon Profiles (1300-1600 LT)
• Afternoon profile BLHTs seem to be bimodal First half of the campaign: BLHTs from 0.3 to 0.6 km Second half of the campaign: BLHTs from 0.5 to 0.7 km • 52 of 56 afternoon profiles had clear BLHTs Acknowledgements: This research was funded by NASA's Earth Venture-1 System Science Pathfinder (ESSP) Program Office. We wish to thank the E support. We also would like to thank the pilots and flight crew of the NASA P-3 Flight Facility for their support and important contributions.
Composite profiles of different constituents scaled by the profile BLHT and the average concentration of the co dotted, colored lines represent individual profiles, and the solid black lines represent the median profile for that c Z = pressure altitude, Zi = profile BLHT, C = constituent concentration, and Ci = average constituent concentratio Example daily flight track showing the path of the P-3B and the repeated spirals at each of the six profile sites. The aircraft flew from Bakersfield to Porterville and continued around the circuit clockwise back to Bakersfield. Each flight day usually included three such circuits.
To visualize BL variability and vertical gradients, composite profiles for eight different const were chosen to represent a variety of lifetimes and production/removal processes.

Conclusions:
• In addition to boundary layer heights increasing throughout the day, aftern also higher in the second half of the campaign than they were in the first hal • For many constituents, profile fluctuations (i.e. BL standard deviation-to-av the day, probably due to increased mixing and decreased stability in the BL. • Only H 2 O and aerosol scattering appeared to have a consistent non-zero had zero vertical gradient, and the other constituents had too much variabi However, for all other parameters besides O 3 , median trends were usual almost 75% of trends were negative. These results suggest that the BL was  System Science Pathfinder (ESSP) Program Office. We wish to thank the ESSP Program Office for the support. We also would like to thank the pilots and flight crew of the NASA P-3B through the NASA Wallops Flight Facility for their support and important contributions.
Composite profiles of different constituents scaled by the profile BLHT and the average concentration of the constituent within the boundary layer. The dotted, colored lines represent individual profiles, and the solid black lines represent the median profile for that constituent and time period. Z = pressure altitude, Z i = profile BLHT, C = constituent concentration, and C i = average constituent concentration in the boundary layer To visualize BL variability and vertical gradients, composite profiles for eight different constituents were created. Constituents were chosen to represent a variety of lifetimes and production/removal processes.