Using Lidar Technology To Assess Urban Air Pollution and Improve

Jun 20, 2019 - Using Lidar Technology To Assess Urban Air Pollution and Improve .... The Supporting Information is available free of charge on the ACS...
0 downloads 0 Views 4MB Size
Article Cite This: Environ. Sci. Technol. XXXX, XXX, XXX−XXX

pubs.acs.org/est

Using Lidar Technology To Assess Urban Air Pollution and Improve Estimates of Greenhouse Gas Emissions in Boston Yanina D. Barrera,*,† Thomas Nehrkorn,‡ Jennifer Hegarty,‡ Maryann Sargent,† Joshua Benmergui,† Elaine Gottlieb,† Steven C. Wofsy,† Phil DeCola,§,∥ Lucy Hutyra,⊥ and Taylor Jones†,§

Downloaded via GUILFORD COLG on July 23, 2019 at 19:33:15 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.



School of Engineering and Applied Sciences and Department of Earth and Planetary Sciences, Harvard University, Cambridge, Massachusetts 02138, United States ‡ Atmospheric and Environmental Research, Inc., Lexington, Massachusetts 02421, United States § Sigma Space Corporation, Lanham, Maryland 20706, United States ∥ Department of Atmospheric and Oceanic Sciences, University of Maryland, College Park, Maryland 20742, United States ⊥ Department of Earth and Environment, Boston University, Boston, Massachusetts 02215, United States S Supporting Information *

ABSTRACT: Simulation of the planetary boundary layer (PBL) is key for forecasting air quality and estimating greenhouse gas (GHG) emissions in cities. Here we conducted the first long-term and continuous study of PBL heights (PBLHs) in Boston, MA, using a compact lidar instrument. We developed an image recognition algorithm to estimate PBLHs from the lidar measurements and evaluated simulations of the PBL from seven numerical weather prediction (NWP) model versions, which showed different systematic errors and variability in simulating the PBLHs (discrepancies from −2.5 to 4.0 km). The NWP model with the best overall agreement for the fully developed PBL had R2 = 0.72 and a bias of only 0.128 km. However, this model predicted a notable number of anomalously high carbon dioxide concentrations at ground stations, because it occasionally significantly underestimated the PBLH. We also developed a novel method that combines lidar data with footprints from a Lagrangian particle dispersion model to identify long-range transport of air pollution in the nocturnal residual layer. Our framework was powerful in evaluating the performance of models used to estimate air pollution and GHG emissions in cities, which is critical to track progress on emission reduction targets and guide effective policies.

1. INTRODUCTION With the pending withdrawal of the U.S. from the Paris Climate Accord, cities and states have pledged to meet the commitments of the accord by showing efforts to mitigate greenhouse gas (GHG) emissions. Establishing a baseline of GHG contributions from each city is key to tracking progress and enabling effective policies and actions to control emissions. Estimating GHG emissions, however, poses several challenges, since emissions are not only sensitive to human activities, but also atmosphere−land surface interactions, such as the heat island effect and biogenic activity due to plants and soil. This complexity provides an impetus to develop high spatial resolution models that can adequately represent the spatial variability of land surface activities that influence emissions. These comprehensive models in turn must be evaluated against both meteorological and GHG measurements. The planetary boundary layer (PBL) defines the mixing height of air pollutants. It is the turbulent domain that connects the surface environment to the large-scale atmosphere. The PBL also serves as a locus for vertical and horizontal transport. The air © XXXX American Chemical Society

motions in the PBL are not resolved in numerical weather prediction (NWP) models; hence its height, wind speed, and other properties are simulated using a variety of parametrizations. It is very challenging to accurately simulate the PBL and to validate the simulations of PBL heights (PBLHs). In studies that compare atmospheric measurements of PBLH against NWP simulations, the strongest agreement is typically found during summer months,1,2 with larger systematic errors of PBLHs during spring and winter seasons.1−5 These errors in PBL simulations can have significant impacts on estimating greenhouse gas fluxes.3,6 There is often a residual layer (RL), lying just above the PBL, defined as residual air from an earlier PBL, that separates the present PBL from the free atmosphere.7−9 Typically, the RL forms at night, as the surface cools and stratification strands air Received: Revised: Accepted: Published: A

January 30, 2019 June 18, 2019 June 20, 2019 June 20, 2019 DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 1. (a) The NRB signal for September 25, 2012, at 14 UTC and (b) the result of applying the WCT filter at 14 UTC, improving signal detection of changes in NRB and identification of the growing PBL and the RL. (c) NRB image with PBL and RL. (d) STILT footprint sums plotted vs receptor height, followed 24 h backward in time, starting at 0, 3, 6, 9, 12, 15, and 21 UTC. By this approach, the RL was identified between 1.5 and 2.5 km, where the footprint sum zeros out (mostly, around 2 km), indicating that the particles traveling backward in time did not touch the surface during the previous day.

pollutants from the previous PBL in the RL, where they remain above the nocturnal PBL and can be transported long distances. However, the entrapment and entrainment processes associated with the RL may make it difficult to distinguish local from regional air pollution. Relatively few studies have focused on the RL8,10−12 because it does not interact with the surface. In this study, we use observations from a MiniMPL sensor from the Micro Pulse LiDAR division of Hexagon Geosystems (originally developed by Sigma Space Corp.) to observe aerosol loading of the PBL and to estimate PBL and RL heights (herein referred to as “RLH”) in Boston, MA. This sensor is part of the Northeast Urban Network under NASA’s Carbon Monitoring System (CMS). The greater Boston area does not rely on energy-intensive industries,13 and up to half of the air pollution is estimated to be regional.14 Because strong emissions are anticipated from traffic and heating in Boston during the day, with smaller emissions at night, we can exploit our MiniMPL sensor to identify diurnal cycles of the PBL and to detect outside influences. We used image recognition techniques and fuzzy logic1 to identify and estimate the PBLH and RLH from the MiniMPL data. PBLHs from the MiniMPL were then utilized to evaluate NWP models, including four configurations of a forecast model

(Weather and Research Forecasting (WRF)), two operational products (the North American Mesoscale (NAM) and Global Data Assimilation System (GDAS), and one reanalysis product (the North American Regional Reanalysis (NARR)]. We also compared the performance of two Lagrangian particle dispersion models (LPDMs), the Stochastic Time-Inverted Lagrangian Transport (STILT) and the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT), which were coupled to WRF and NAM meteorology to estimate carbon dioxide (CO2) fluxes in Boston.6 We detected long-range transport of air pollution in the nocturnal RL by developing a novel method that assesses MiniMPL-retrieved RLHs against information from STILT footprints,15 along a vertical column. Each footprint quantifies the sensitivity of simulated concentrations at each altitude at the receptor location to upwind emission sources.16

2. METHODS Sampling Sites and Measurements. The MiniMPL sampling site in Boston, MA (42.350 N, 71.104 W), is located approximately 32 m above ground level on a rooftop of a building at Boston University, between the Charles River to the B

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

at 14 UTC, and Figure1b shows the result of applying a firstderivative Gaussian WCT on the NRB profile, which transforms the NRB image by detecting small signal shifts in the NRB, highlighting atmospheric structures of interest. In Figure 1c, the black and gray colors indicate the lowest backscattering of light and the red and white colors, the highest backscattering of light in the NRB profile. The PBL was identified at roughly 0.6 km and the RL at 1.6 km at 14 UTC, respectively, in this example. The WCT-filtered NRB profile for the entire day was then processed using image thresholding, singular value decomposition, and fuzzy logic (Figure S2, SI) to identify our atmospheric structures of interest: the PBL and RL. The RL experiences fluctuations throughout the night and day, especially during the morning hours if the RL is being entrained with the growing PBL. We used singular value decomposition (SVD) to identify the largest NRB signals above the PBL, which we define as the MiniMPL-retrieved RL. Often, the RL signal can blend with the PBL signal once the PBL is fully developed. Our fuzzy logic incorporated the anticipated timing of a growing PBL, with the peak PBLH limited to 100 m below the daytime RL. Figure 1c shows the lidar NRB with the retrieval of the PBLHs and RLHs at our Boston site, after processing. NWP Model Setup. In the first phase of our NWP analysis, three configurations of the forecast model WRF-ARW,20,21 using different options for PBL parametrizations and urban canopy treatment, were used to compute the PBL heights and compare against MiniMPL-retrieved PBLHs in Boston, MA. Each of the WRF configurations included a single-layer urban canopy model22 to account for the effect of the urban landscape (buildings, roads, etc.) on the fine-scale meteorological circulation. All WRF runs used the NARR reanalysis for initial and lateral boundary conditions. For the Boston site, we compared three different WRF configurations: (1) MYJ_v341, WRF v.3.4.1 using a second-order MYJ PBL scheme and urban roughness lookup tables, which was used for the Boston methane study by McKain et al.;23 (2) MYJ_v361, WRF v.3.6.1 using a MYJ PBL scheme and urban roughness database, with enlarged inner domains, but otherwise identical to MYJ_v341, which was used for the Boston carbon dioxide study by Sargent et al.;6 and (3) YSU_v361, a variant of configuration 2, which differs only in the use of the first-order YSU PBL scheme with the topographic wind correction24 enabled. The topographic wind correction, which is only available in connection with the YSU PBL scheme, resulted in an improved fit with observed lowlevel winds, but this version also had larger temperature errors. All three configurations were compared over a 2-month period (from June 20 to August 31, 2013), while the MYJ_v361 and YSU_v361 configurations were compared for an 8-month period (from June 20, 2013 to February 28, 2014). MYJ_v341 includes four nested grids (27, 9, 3, and 1 km grid spacing), with physics options corresponding to those of the Turb-U configuration used in the Salt Lake City study by Nehrkorn et al.25 The two v.3.6.1 runs (MYJ_v361 and YSU_v361) used a different set of four nested grids covering the entire northeast corridor, with grid spacing of 36, 12, 4, and 1.33 km (Figure S3, SI). A total of 40 vertical levels were used with average vertical spacing of 250 m, from 250 m to 2.0 km, and 500 m, from 2.0 to 5.0 km. In addition, an updated WRF version (v3.6.1 vs v3.4.1) and city-specific urban parameter database from NUDAPT-44 (MYJ_v341 used lookup tables based on land-use category) were used in these runs. The innermost nested grids (1 km and 1.33 km, respectively) from WRF contain the highest resolution

northwest and a major interchange of the I-90 highway to the southeast, surrounded by apartment buildings of similar height. The MiniMPL is an aerosol backscattering lidar and a miniaturized version of the standard MPL used in the NASA MPLNET global lidar network. An extensive description of this lidar instrument can be found in Ware et al.4 The MiniMPL operates similarly to other lidars. The units are compact and continuously take measurements with little need for on-site maintenance. The MiniMPL emits light at 532 nm wavelength and records the backscattering from aerosols and atmospheric molecules. The MiniMPL-retrieved normalized relative backscatter (NRB) profiles represent the backscattering of light (in photon counts km2/μs μJ), after correcting and normalizing the measurements. All MiniMPL data were averaged over 5-min bins at a vertical resolution of 30 m. The MiniMPL sensor used in this study had a blind zone below 200 m, roughly. At times, in the late afternoon hours, the image-processing algorithm had difficulty distinguishing the PBL from other NRB signals, when aerosol signals can be stronger than the fully developed PBL or the collapsing PBL. Therefore, our image-processing algorithm focused on PBLHs above 200 m and 100 m below the RLH, or aerosol signals, from 9 to 21 UTC (coordinated universal time). Additionally, periods during which WRF simulated rain were screened out in this analysis (precipitation greater than 0 mm). Sea breezes in the afternoon hours identified by the occurrence of easterly winds after 16 UTC were also screened out to remove strong aerosol layer signals in the marine layer that can lift the land-based PBL from the surface, giving very complex atmospheric structure. Results for a typical day are shown in Figure 1. Observations of carbon dioxide were measured with a Picarro cavity ringdown spectrometer at our Boston MiniMPL sampling location, as well as at four other stations in Massachusetts and New Hampshire.6 The Boston University and Copley Square sites sample at 29 m above ground level (agl) and 215 m agl, respectively, providing observations of the vertical gradient in the atmosphere. The Copley site is typically within the afternoon PBL but above the nocturnal PBL. We developed a CO2 concentration curtain at the boundary of the study region [Figure S1, Supporting Information (SI)] based on a combination of tower measurements, eddy flux measurements, and modeled vertical concentration gradients to calculate CO2 concentrations entering the urban domain.6 The background site was chosen on the basis of the wind direction, and the associated background concentration was subtracted from the observed CO2 at the urban site to define CO2 enhancements in the urban core (Figure S1, SI). PBL and RL Retrieval Using Lidar Data. The MiniMPL was used to identify the growing PBL from the early morning to late afternoon, roughly between the hours of 9−21 UTC (4−17 EST or 5−18 EST), when NRB signals are elevated due to emissions from the transportation, commercial, residential, and industrial sectors in our urban domain. Moisture and aerosols trapped within the PBL17 produce a sharp gradient in the NRB at the top of the PBL. Figure 1 illustrates the signal processing of the lidar NRB profile for September 25, 2012. We used a wavelet covariance transform (WCT) as described in several previous studies, except we used the first derivative Gaussian wavelet instead of the Haar wavelet1,18,19 to give clearer edges and structures in the NRB signals from our MiniMPL units. Additionally, a sigmoidal curve model was fitted to the data to interpolate through time intervals when the wavelet did not identify the growing PBL. Figure 1a shows the lidar NRB profile C

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

footprint analysis using the product from STILT is shown in Figure 1d for September 25, 2012. This day is fairly complicated, with enhanced NRB signals detected all night and day, illustrating the effectiveness of our method. We found that the sum of footprints at each level, generated by STILT, first reached near zero at receptor heights between 1.5 and 2.0 km (Figure 1d), also identified as the top of an aerosol layer (Figure 1b). Our MiniMPL algorithm identified the RL between 1.6 and 2.0 km (Figure 1c). At our sampling location large NRB signals are sometimes seen in the RL, as pollution from the previous day or from longrange transport can remain aloft in the RL, crossing political or jurisdictional boundaries (herein referred to as “transboundary”). Days which showed enhanced nighttime NRB signals (NRB > 0.3) in MiniMPL profiles in Boston were flagged as potential transboundary air pollution episodes. After screening for clouds using NASA’s Geostationary Operational Environmental Satellite (GOES) infrared imagery, which can be downloaded from the Plymouth State Weather Center (https://vortex.plymouth.edu/main/disclaimer.html), the STILT vertical footprints were evaluated to ascertain if the enhanced nighttime NRB signals were in fact transboundary, i.e., showing sensitivity to emission sources far from Boston.

and were utilized to compare PBL simulations at the nearest grid to our Boston sampling site. In the second phase of our NWP study, MiniMPL-retrieved PBLHs were also compared from September 2013 to November 2014 with the following models: WRF-ARW at 27 km grid, NARR at 32 km grid (Mesinger et al.26), NAM at 12 km grid (Janjic et al.27), and GDAS at 0.25° grid (Kleist et al.28). PBL simulations were compared directly with PBLHs retrieved at our Boston MiniMPL sampling location. For both phases of our NWP analysis, if precipitation greater than 0 mm or sea breeze was detected in WRF meteorological simulations, those time periods were removed from the study. Detection of Air Pollution Episodes in the Nocturnal RL. Like the PBLHs, the residual layer heights were retrieved from lidar data using image recognition and fuzzy logic. After we performed WCT on our lidar NRB profiles, we processed the image using SVD (Figure S2, SI). In linear algebra, SVD is a numerical technique that uses matrix decomposition. SVD looks at subsets of data (splitting the matrix into linearly independent components) and can be used as a powerful tool to transform images, differentiating noise from signal. Here, our signal is the RL and our matrix or image is the miniMPL NRB profile for a single day. We used the first singular values in our SVD matrix, which contains the largest amount of information and highlights the intensity of light from our image matrix. We then used SVD to detect the RL in the miniMPL NRB profile, where stronger backscattering of light is expected due to pollution remaining aloft from the previous day. In the morning and afternoon hours, stronger NRB signals are anticipated due to an increase in anthropogenic activity, or emissions, that are associated with an increase in traffic, heating/cooling, etc. These emissions include greenhouse gases such as carbon dioxide and methane, particulate matter, and secondary formation of aerosols.29−32 Therefore, we postulated that SVD would be an effective method to separate the PBL signal from the RL signal. At times, in the afternoon hours, the fully developed daytime PBL can blend with the residual layer top. We addressed this problem by using fuzzy logic to identify the RL when it is at least 100 m above the PBL height (Figure 1c). WRF identifies PBL heights but resolving for RLH requires much finer vertical resolution aloft, which would be very computationally expensive. Therefore, as a secondary evaluation tool, we created a novel method using the coupled WRF− STILT model to identify the RL at our MiniMPL sampling location. The STILT model released an ensemble of 500 particles at each of 10 levels (“receptors”) from 0.25 to 3 km along a vertical column at the lidar site. Particles were transported on the basis of wind fields and parametrized turbulence from the WRF MYJ_v361 configuration followed backward in time for 24 h, every 3 h. The STILT turbulence parameters used the standard settings KBLS = 1 (boundary layer stability derived from heat and momentum fluxes) and KBLT = 1 (PBL turbulence scheme based on Beljaars/Holtslag and Betchov/Yaglom). The “footprint”, which quantifies the impact of upwind surface fluxes on changes in atmospheric concentration at the receptor, is then computed on the basis of the amount of time particles residing in the layer close enough to the surface to “feel” its influence (we used 50% of the mixing height, following standard STILT practice). We estimated the top of the RL as the height of the lowest receptor for which the sum of its footprints (over all map locations and back trajectory times in a 24-h period) is near zero; i.e., the Lagrangian particles do not touch the ground in the first 24 h. For example, a vertical

3. RESULTS AND DISCUSSION Lidar-Retrieved PBL Heights vs NWP PBL Simulations. We assessed the quality of PBLH simulations from a number of NWP models by comparing them to MiniMPL-retrieved PBLH from 9 to 21 UTC. This analysis was completed in two phases. In the first phase, three WRF configurations (MYJ_v341, MYJ_v361, and YSU_v361) were compared to MiniMPL PBL heights for a summer period (from June 20 to August 31, 2013), and two configurations (MYJ_v361 and YSU_v361) were compared for an autumn period (from September 1 to November 30, 2013) and winter period (from December 1 to February 28, 2014). The two v.3.6.1 configurations (MYJ_v361 and YSU_v361) performed better than the older v.3.4.1 configuration during the summer period (Figures S4a, SI). MYJ_v361 showed the best performance during all three periods, which is our highest resolution NWP configuration (1.33 km × 1.33 km). Agreement between NWP model simulations and MiniMPL retrievals of PBLH improved when the days with sea breezes were omitted from the analysis (Figure S4a−c, SI). In the late afternoon, simulated WRF PBLH decayed quicker than MiniMPL-retrieved PBLHs. This may be due to MiniMPL data biasing the PBLH high, as enhanced NRB signals persist from aerosols and atmospheric molecules that remain aloft once the PBL is fully developed and the RL begins to form. However, we attempted to minimize this bias by limiting the peak PBLH to 100 m below the RL in our image processing algorithm. Additionally, WRF simulations may overestimate the speed of the PBL collapse in the late afternoon. Results reported by Hegarty et al.33 highlight that the PBLHs derived from 8 pm local time radiosonde data often agreed with PBLH derived from standard MPL technology, while the WRF PBLHs were lower than both the MPL and radiosonde data (viz., their Figure 5). In the second phase, PBL simulations from two forecast models (WRF MYJ_v361 and WRF-ARW), two operational products (GDAS and NAM), and one reanalysis product (NARR) were compared with MiniMPL-retrieved PBL heights from September 2013 to November 2014. We examined the PBL growth as mean percent errors (MPE) of PBLH for each D

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 2. PBL growth (9−21 UTC) from NWP models as mean percent error (a−e) and bias (f−j), in comparison to the MiniMPL measurements in Boston, MA, from September 2013 to November 2014, for each metereological season: fall (Sept−Nov), winter (Dec−Feb), spring (Mar−May), and summer (June−Aug).

Table 1. The Daily Maximum PBLH and Time When the Maximum PBLH Occurred (NWP simulations vs MiniMPL retrievals), Compared to MiniMPL Data at the Boston Sampling Sitea NWP model

grid

WRF MYJ_v361 WRF-ARW NAM NARR GDAS

1.33 km 27 km 12 km 27 km 0.25°

NWP vs MiniMPL daily maximum PBLH

time of daily maximum PBLH

R2 = 0.72, RMSE = 0.424 km, bias = 0.128 km R2 = 0.61, RMSE = 0.423 km, bias = −0.062 km R2 = 0.47, RMSE = 0.915 km, bias = 0.470 km R2 = 0.60, RMSE = 0.517 km, bias = −0.250 km R2 = 0.43, RMSE = 0.521 km, bias = 0.061 km

R2 = 0.80, RMSE = 2.82 h, bias = −0.469 h R2 = 0.77, RMSE = 2.92 h, bias = −0.485 h R2 = 0.57, RMSE = 4.76 h, bias = −1.735 h R2 = 0.67, RMSE = 3.97 h, bias = −1.556 h R2 = 0.45, RMSE = 5.38 h, bias = −1.429 h

Data was filtered by WRF meteorological data and dates kept in MiniMPL NRB profile analysis.

a

Many inversion modeling studies only use afternoon observations, when the PBL is well-developed and PBLH errors are expected to be smaller. To get a model evaluation metric applicable to these studies, we compared the daily maximum PBLH of the NWP models with the MiniMPL. Figure 2 shows the mean bias and RMSE once the PBL is fully developed (daily maximum PBLH) for each season. PBLH errors are strongly autocorrelated in time within each day (Figure S5, SI). The forecast models performed the best, where WRF MYJ_v361 PBL simulations showed the best agreement in estimating the daily maximum PBLH throughout our study period (Table 1), with good correlation (R2 = 0.72), RMSE of 0.424 km, and relatively low bias of 0.128 km. WRF MYJ_v361 showed RMSE and bias similar in magnitude to those reported by other studies that evaluated WRF PBLH.34,35 During the winter months, models either systematically underestimated (NARR, WRFARW) or overestimated (GDAS, NAM, WRF MYJ_v361) the daily maximum PBLH (Figure 3a). All models systematically underestimated the time at which the fully developed PBL depth

model, during each season (Figure 2). Because the PBL has a strong diurnal cycle during well-mixed conditions, systematic errors in the NWP models caused errors in the PBLH diurnal cycle. Overall, the forecast models had small mean percent errors at all times. The operational and reanalysis products had large errors before the daytime PBLH develops (morning hours) and smaller errors after the daytime PBLH develops (afternoon hours). It is important to note, however, that the MiniMPL unit may bias PBLH high in the morning hours if the nocturnal RL or PBL is being entrained. This potential bias should dissipate by midday, as all data were analyzed above 200 m agl (conservative). The heights of the PBL were biased low during all seasons in NARR and WRF-ARW (Figures 2h,i). PBLHs were biased low in the afternoon hours for all seasons in NARR, during spring in NAM, and during fall in WRF MYJ_v361 (Figures 2f,g,j). The WRF MYJ_v361 model produced the best overall agreement with the MiniMPL diurnal cycle of PBLHs. Lastly, although the operational product from NAM shows the largest systematic errors overall, it less often shows a low bias in PBLH compared to the other models. E

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 3. (a) Mean percent errors and (b) root-mean-square errors are shown for each NWP model in order to evaluate systematic errors of the fully developed PBL at our MiniMPL sampling site in Boston, MA, for four meteorological seasons: fall (Sept−Nov), winter (Dec−Mar), spring (Apr− May), and summer (June−Aug).

Figure 4. Carbon dioxide enhancements [CO2 (urban) − CO2 (background)] in Boston, MA, are shown for afternoon hours (from 11 a.m. to 4 p.m. EST) for November 2−16, 2014. The gray bars indicate the days where lidar PBLH data were available. The blue and red lines show WRF−STILT and NAM−HYSPLIT model predictions and the black line observed CO2 enhancements. The red and blue dots indicate when the model PBL height was too low by 40% or more compared to the lidar-retrieved PBL height.

occurs in MiniMPL data, with discrepancies from −0.469 to −1.735 hours.

Estimating Greenhouse Gas Emissions in Boston, MA, using Lidar Data. We also used MiniMPL data to better F

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 5. Encroachment of aerosol pollution detected in lidar data within the nighttime RL on November 20 (a) and November 23 (b), 2014; enhanced NRB signals detected in the daytime PBL on November 23, following the 5 day transboundary air pollution episode in Boston (b); the RL was detected roughly between 1.5 and 2.5 km on both days using lidar data (a, b) and vertical footprints (c, d); and STILT footprints within the innermost grid at the Boston receptor location, 24 h backward in time, at 529 m agl at 0 UTC on November 20 (e) and 1029 m agl at 0 UTC on November 23 (f), show observations sensitive to sources in Pennsylvania, New York, and Connecticut.

location. The observed CO2 enhancements were calculated by subtracting the CO2 concentrations at the Boston miniMPL site from the background air entering one of our background sites (Figure S1, SI). This was determined on the basis of the wind direction that day. These comparisons focused on afternoon hours when the PBL is well-developed (Figure 2 shows significantly better NWP model PBLH performance from 16 to 21 UTC than from 9 to 16 UTC). There were significant differences in the CO2 enhancements calculated using the WRF−STILT and NAM−HYSPLIT frameworks, which were highly correlated with differences in PBLH between the two models. The HYSPLIT and STILT

understand the impact of simulated PBLH on estimated CO2 concentrations calculated by two inverse model frameworks. The forecast model WRF was coupled with STILT and the forecast model NAM coupled to HYSPLIT, and both were run backward in time to give the adjoint (inverse) model. Both inverse models were combined with a high-resolution bottomup CO2 inventory36 to estimate CO2 fluxes in the Boston6 area from concentrations observed at the surface and at 215 m. The sum of modeled CO2 enhancements in the urban core and concentrations at the boundary of the study region (90-km radius circle around Boston) were compared to observed CO2 enhancements measured at our Boston MiniMPL sampling G

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

4. IMPLICATIONS Accurate simulation of the PBL is a key requirement for forecasting air quality and estimating GHG emissions from atmospheric observations. We have developed the first longterm and continuous record of PBL heights in Boston, MA, using a MiniMPL sensor. We compared five NWP models and found that the WRF forecast models agreed better with MiniMPL observations than the operational (GDAS and NAM) and reanalysis (NARR) products. Sea breeze conditions were generally not simulated well. In the wintertime, the operational and reanalysis products systematically overestimated PBL height in the afternoon. These results emphasize that PBL simulations need to be improved, especially for the low wintertime PBL heights that may cause haze events and public health nuisances. The lidar was also used to evaluate the performance of two LPDMs estimating CO2 emissions in Boston. The disagreement between the MiniMPL-derived and modeled PBL height during periods of low PBL was an excellent predictor for CO2 errors in models, enabling assessment of relative model performance. We noted that accurate estimates of PBL heights in meteorological models, coupled to inversion models, may improve GHG emissions estimates, but careful assessments of the impact of biases and anomalies are required. We also showed that the lidar, when applied within our PBLH/RLH and inverse analysis framework, provides an excellent tool to detect and help quantify transboundary contributions to urban pollution. The Intergovernmental Panel on Climate Change Task Force on National Greenhouse Gas Inventories (IPCC TFI) has focused on developing and refining GHG inventories at national scales. But recently, the Paris Agreement and Lima−Paris Action Agenda recognize the critical role of subnational entities, especially cities. Our work shows how a simple, robust lidar can help estimate GHG emissions and establish accurate baselines at urban and subnational scales. Future work will further explore detecting and identifying transboundary air pollution at night in Boston, MA, and Mineola, NY, using lidar, a high-resolution carbon dioxide emissions inventory, and the STILT model. This work is currently underway, in order to identify the source regions throughout the northeastern U.S. and investigate vertical profiles of CO2 enhancements within the nocturnal RL and its relationship to lidar NRB profiles.

model footprints are much more sensitive to errors in PBLH when the PBL is low (800 m) because the footprint and associated concentration enhancement are inversely proportional to PBLH. In Figures 2 and 3, the bias and errors of both WRF models were significantly lower than those of NAM. However, WRF-ARW and MYJ_v361 simulations more often estimated a PBL less than 500 m (51.7% and 33.7% of PBL simulations) than NAM (26.4% of PBL simulations). At times, these erroneously low model PBLHs produced very high CO2 enhancements in the WRF−STILT framework that were not present in the NAM−HYSPLIT framework (Figure 4). Although NAM did not have the best overall agreement with MiniMPL PBLH, it less often showed a low bias in PBLH, to which STILT and HYSPLIT are particularly sensitive. The lidar therefore enabled us to understand the source of unrealistic model CO2 enhancements in WRF−STILT and demonstrate the relative advantage of the NAM−HYSPLIT model. Detecting Air Pollution Episodes within the Nocturnal RL in Boston, MA. The lidar detected numerous transboundary air pollution events within the nocturnal RL in Boston during our study period. Most were undetectable at the ground but nevertheless contributed to pollution in Boston when entrained in the growing PBL the next day (e.g., Figure 1). The Boston area does not contain strong emission sources of particulate matter at night; hence, when we observed dense aerosol layers with relatively high NRB signals (NRB > 0.3) above the PBL at night, we flagged them as potentially due to long-range transport of aerosol pollution. For these events, we used STILT to compute footprints from 10 selected altitude levels above our MiniMPL site, enabling us to identify nighttime high aerosol backscattering events that originated in source regions outside of the Northeastern Corridor 1 or 2 days previously. All selected release times for conducting STILT footprint analyses were screened for clouds using GOES satellite infrared imagery to eliminate errors in NRB signals due to cloud presence. Figure 5 shows an example 5-day transboundary air pollution episode detected on November 19−23, 2014. Panels a and b show encroaching aerosols in the nighttime RL in Boston on the second and last day of this episode. Results show strong agreement between RLHs retrieved by our MiniMPL and using our STILT vertical footprint analysis (Figures 5c,d). All footprint maps were created using the “maps” package version 3.3.0, “mapdata” package version 2.3.0, and “fields” package version 9.6 for R programming language. Figure 5e shows the footprints retrieved at a receptor height of 529 m agl beginning at 0 UTC on November 20 and Figure 5f a receptor height of 1029 m agl beginning at 0 UTC on November 23. During the last day of the event, on November 23, 2014, the lidar detected enhanced NRB signals (>0.3) within the nighttime RL and later, in the daytime PBL. STILT footprints indicate that the air above Boston was highly influenced by emission sources from Pennsylvania, New York, and Connecticut. These regions contain coal mining, traffic, wintertime biomass burning, and heavy industrial sources that can contribute significantly to aerosol pollution and airborne toxic air contaminants in Boston and which can be uniquely detected by lidar using the PBL analysis framework. We detected a total of 35 transboundary air pollution events over a 15-month period, with source regions ranging from Pennsylvania to Canada to New York.



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.9b00650.



Detailed evaluation of PBLH and RLH retrievals from MiniMPL data, as well as NWP and LPDM simulations (Table S1 and Figures S1− S7) (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]; telephone: +16174966361. ORCID

Yanina D. Barrera: 0000-0002-9666-7197 Steven C. Wofsy: 0000-0002-3133-2089 Notes

The authors declare no competing financial interest. H

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology



(16) Lin, J. C.; Gerbig, C.; Wofsy, S. C.; Andrews, A. E.; Daube, B. C.; Davis, K. J.; Grainger, C. A. A near-field tool for simulating the upstream influence of atmospheric observations: The stochastic time-inverted Lagrangian transport model (STILT). J. Geophys. Res. 2003, 108 (D16), 4493. (17) Seibert, P.; Beyrich, F.; Gryning, S.-E.; Joffre, S.; Rasmussen, A.; Tercier, P. Review and intercomparison of operational methods for the determination of the mixing height. Atmos. Environ. 2000, 34, 1001− 1027. (18) Davis, K. J.; Gamage, N.; Hagelberg, C. R.; Kiemle, C.; Lenschow, D. H.; Sullivan, P. P. An objective method for deriving atmospheric structure from airborne lidar observations. J. Atmos. Oceanic Technol. 2000, 17, 1455−1468. (19) Brooks, I. M. Finding boundary layer top: Application of a wavelet covariance transform to lidar backscatter profiles. J. Atmos. Oceanic Technol. 2003, 20, 1092−1105. (20) Skamarock, W. C.; Klemp, J. B. A time-split nonhydrostatic atmospheric model for weather research and forecasting applications. J. Comput. Phys. 2008, 227, 3465−3485. (21) Powers, J. G.; Klemp, J. B.; Skamarock, W. C.; Davis, C. A.; Dudhia, J.; Gill, D. O.; Coen, J. L.; Gochis, D. J.; Ahmadov, R.; Peckham, S. E.; et al. The Weather Research and Forecasting (WRF) Model: Overview, system efforts, and future directions. Bull. Am. Meteor. Soc. 2017, 98, 1717−1737. (22) Kusaka, H.; Kondo, H.; Kikegawa, Y.; Kimura, F. A simple singlelayer urban canopy model for atmospheric models: Comparison with multi-layer and slab models. Bound. Layer Meteor. 2001, 101 (3), 329− 358. (23) McKain, K.; Down, A.; Raciti, S. M.; Budney, J.; Hutyra, L. R.; Floerchinger, C.; Herndon, S.; Nehrkorn, T.; Zahniser, M. S.; Jackson, R. B.; et al. Methane emissions from natural gas infrastructure and use in the urban region of Boston, Massachusetts. Proc. Natl. Acad. Sci. U. S. A. 2015, 112, 1941−1946. (24) Jiménez, P. A.; Dudhia, J. On the Ability of the WRF Model to Reproduce the Surface Wind Direction Over Complex Terrain. J. Appl. Meteor. Climatol. 2013, 52, 1610−1617. (25) Nehrkorn, T.; Henderson, J.; Leidner, M.; Mountain, M.; Eluszkiewicz, J.; McKain, K.; Wofsy, S. C. WRF simulations of the urban circulation in the Salt Lake City area for CO2 modeling. J. Appl. Meteor. Clim. 2013, 52, 323−340. (26) Mesinger, F.; DiMego, G.; Kalnay, E.; Mitchell, K.; Shafran, P. C.; Ebisuzaki, W.; Jovic, D.; Woollen, J.; Rogers, E.; Berbery, E. H.; et al. North American Regional Reanalysis. Bull. Am. Meteor. Soc. 2006, 87, 343−360. (27) Janjic, Z. I. A nonhydrostatic model based on a new approach. Meteor. Atmos. Phys. 2003, 82, 271−285. (28) Kleist, D. T.; Parrish, D. F.; Derber, J. C.; Treadon, R.; Wu, W. S.; Lord, S. Introduction of the GSI into the NCEP Global Data Assimilation System. Wea. Forecasting. 2009, 24, 1691−1705. (29) Jimenez, J. L.; Canagaratna, M. R.; Donahue, N. M.; Prevot, A. S. H.; Zhang, Q.; Kroll, J. H.; DeCarlo, P. F.; Allan, J. D.; Coe, H.; Ng, N. L.; et al. Evolution of Organic Aerosols in the Atmosphere. Science 2009, 326, 1525−1529. (30) Kelly, J. T.; Parworth, C. L.; Zhang, Q.; Miller, D. J.; Sun, K.; Zondlo, M. A.; Baker, K. R.; Wisthaler, A.; Nowak, J. B.; Pusede, S. E.; et al. Modeling NH4NO3 Over the San Joaquin Valley During the 2013 DISCOVER-AQ Campaign. J. Geophys. Res. Atmos. 2018, 123, 4727− 4745. (31) Schroder, J. C.; Campuzano-Jost, P.; Day, D. A.; Shah, V.; Larson, K.; Sommers, J. M.; Sullivan, A. P.; Campos, T.; Reeves, J. M.; Hills, A.; et al. Sources and secondary production of organic aerosols in the northeastern United States during WINTER. J. Geophys. Res. Atmos. 2018, 123, 7771−7796. (32) Zhang, Q.; Jimenez, J. L.; Canagaratna, M. R.; Allan, J. D.; Coe, H.; Ulbrich, I.; Alfarra, M. R.; Takami, A.; Middlebrook, A. M.; Sun, Y. L.; et al. Ubiquity and dominance of oxygenated species in organic aerosols in anthropogenically-influenced Northern Hemisphere midlatitudes. Geophys. Res. Lett. 2007, 34 (13), L13801.

ACKNOWLEDGMENTS We would like to thank engineers John Budney and Bruce Daube for their help maintaining the MiniMPL network. This work was funded by the NASA Carbon Monitoring System (NASA NNX12AP10G and NASA NX12AM82G), the National Science Foundation (NSF) Collaborative Research Awards 1265614 and 1302902, the NSF Major Research Instrumentation Program (AGS-1337512), and the Environmental Defense Fund (1046-000000-10800).



REFERENCES

(1) Lewis, J. R.; Welton, E. J.; Molod, A. M.; Joseph, E. Improved boundary layer depth retrievals from MPLNET. J. Geophys. Res. Atmos. 2013, 118, 9870−9879. (2) Collaud Coen, C. M.; Praz, C.; Haefele, A.; Ruffieux, D.; Kaufmann, P.; Calpini, B. Determination and climatology of the planetary boundary layer height above the Swiss plateau by in situ and remote sensing measurements as well as by the COSMO-2 model. Atmos. Chem. Phys. 2014, 14, 13205−13221. (3) Kretschmer, R.; Gerbig, C.; Karstens, U.; Biavati, G.; Vermeulen, A.; Vogel, F.; Hammer, S.; Totsche, K. U. Impact of optimized mixing heights on simulated regional atmospheric transport of CO2. Atmos. Chem. Phys. 2014, 14, 7149−7172. (4) Ware, J.; Kort, E. A.; DeCola, P.; Duren, R. Aerosol lidar observations of atmospheric mixing in Los Angeles: Climatology and implications for greenhouse gas observations. J. Geophys. Res. Atmos. 2016, 121, 9862−9878. (5) Su, T.; Li, J.; Li, C.; Xiang, P.; Lau, A. K.-H.; Guo, J.; Yang, D.; Miao, Y. An intercomparison of long-term planetary boundary layer heights retrieved from CALIPSO, ground-based lidar, and radiosonde measurements over Hong Kong. J. Geophys. Res. Atmos. 2017, 122, 3929−3943. (6) Sargent, M.; Barrera, Y.; Nehrkorn, T.; Hutyra, L. R.; Gately, C. K.; Jones, T.; McKain, K.; Sweeney, C.; Hegarty, J.; Hardiman, B.; Wofsy, S. C.; Wang, J. A. Anthropogenic and biogenic CO2 fluxes in the Boston urban region. Proc. Natl. Acad. Sci. U. S. A. 2018, 115 (29), 7491−7496. (7) Stull, R. B. An Introduction to Boundary-Layer Meteorology; Kluwer Academy Publishers: Dordrecht, Netherlands, 1988; p 14. (8) Blay-Carreras, E.; Pino, D.; Vilà-Guerau de Arellano, J.; van de Boer, A.; De Coster, O.; Darbieu, C.; Hartogensis, O.; Lohou, F.; Lothon, M.; Pietersen, H. Role of the residual layer and large-scale subsidence on the development and evolution of the convective boundary layer. Atmos. Chem. Phys. 2014, 14, 4515−4530. (9) Emeis, S.; Schafer, K. Remote sensing methods investigate boundary-layer structures relevant to air pollution in cities. Bound. Layer Meteorol. 2006, 121, 377−385. (10) Barlow, J. F. Progress in observing and modelling the urban boundary layer. Urban Clim. 2014, 10, 216−240. (11) Balsley, B. B.; Svensson, G.; Tjernstrom, M. On the scale dependence of the gradient Richardson number in the residual layer. Bound. Layer Meteor. 2008, 127, 57−72. (12) Wehner, B.; Siebert, H.; Ansmann, A.; Ditas, F.; Seifert, P.; Stratmann, F.; Wiedensohler, A.; Apituley, A.; Shaw, R. A.; Manninen, H. E.; Kulmala, M. Observations of turbulence-induced new particle formation in the residual layer. Atmos. Chem. Phys. 2010, 10, 4319− 4330. (13) U.S. Energy Information Administration. Massachusetts State Energy Profile. https://www.eia.gov/state/?sid=MA (accessed Nov 7, 2017). (14) Masri, S.; Kang, C. M.; Koutrakis, P. Composition and sources of fine and coarse particles collected during 2002−2010 in Boston, MA. J. Air Waste Manage. Assoc. 2015, 65 (3), 287−297. (15) Nehrkorn, T.; Eluszkiewicz, J.; Wofsy, S. C.; Lin, J. C.; Gerbig, C.; Longo, M.; Freitas, S. Coupled weather research and forecastingstochastic time-inverted Lagrangian transport (WRF-STILT) model. Meteorol. Atmos. Phys. 2010, 107 (1), 51−64. I

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology (33) Hegarty, J. D.; Lewis, J.; McGrath-Spangler, E. L.; Henderson, J.; Scarino, A. J.; DeCola, P.; Ferrare, R.; Hicks, M.; Adams-Selin, R. D.; Welton, E. J. Analysis of the Planetary Boundary Layer Height during DISCOVER-AQ Baltimore−Washington, D.C., with Lidar and HighResolution WRF Model. J. Appl. Meteor. Climatol. 2018, 57, 2679− 2695. (34) Lauvaux, T.; Miles, N. L.; Deng, A.; Richardson, S. J.; Cambaliza, M. O.; Davis, K. J.; Gaudet, B.; Gurney, K. R.; Huang, J.; O’Keefe, D.; et al. High-resolution atmospheric inversion of urban CO2 emissions during the dormant season of the Indianapolis Flux Experiment (INFLUX). J. Geophys. Res. Atmos. 2016, 121, 5213−5236. (35) Deng, A.; Lauvaux, T.; Davis, K. J.; Gaudet, B. J.; Miles, N.; Richardson, S. J.; Wu, K.; Sarmiento, D. P.; Hardesty, R. M.; Bonin, T. A.; et al. Toward reduced transport errors in a high resolution urban CO2 inversion system. Elem. Sci. Anth. 2017, 5, 20. (36) Gately, C. K.; Hutyra, L. R. Large uncertainties in urban-scale carbon emissions. Geophys. Res. Lett. 2017, 122 (11), 11242−11260.

J

DOI: 10.1021/acs.est.9b00650 Environ. Sci. Technol. XXXX, XXX, XXX−XXX