Satellite soil moisture data assimilation impacts on modeling weather variables and ozone in the southeastern US - Part 2: Sensitivity to dry deposition parameterizations

. Ozone (O 3 ) dry deposition is a major O 3 sink. Realistically representing this process in models is important for accurately simulating O 3 concentrations and exceedances, as well as assessing the O 3 impacts on human and ecosystem 15 health. As a follow-up study of Huang et al. (2021), soil moisture (SM) data from NASA’s Soil Moisture Active Passive mission are assimilated into the Noah-Multiparameterization land surface model within the NASA Land Information System framework, semicoupled with Weather Research and Forecasting model with online Chemistry regional-scale simulations covering the southeastern US. Major changes in the used modeling system include enabling the dynamic vegetation option, adding the irrigation process, and updating the C H (i.e., surface exchange coefficient for heat) scheme. Two different dry 20 deposition schemes are implemented, i.e., the Wesely scheme and a “dynamic” scheme, in the latter of which dry deposition parameterization is coupled with photosynthesis and vegetation dynamics. It is demonstrated that, when the “dynamic” scheme is applied, the modeled O 3 dry deposition velocities as well as the total, stomatal and cuticular O 3 fluxes are overall larger and 2–3 times more sensitive to the SM changes due to the data assimilation (DA). We also highlight that, the configuration of the SM factor controlling stomatal resistance (i.e., the β factor which presents dependencies on soil type and 25 hydrological regime) can strongly affect the quantitative results. Referring to multiple observation and observation-derived evaluation datasets, which may be associated with variable extents of uncertainty, the model performance of vegetation, surface fluxes, weather, and surface O 3 concentrations, shows mixed responses to the DA, some of which display land cover dependency. Finally, using model-derived concentration- and flux-based policy relevant O 3 metrics as well as their matching exposure-response functions, the relative biomass/crop yield losses for several types of vegetation/crops are estimated to be 30 within a wide range below 20%. Their sensitivities to the model’s dry deposition scheme and the implementation of SM DA are discussed. and ecosystem health. Specifically: 1) MDA8 O 3 fields over unban and nonurban regions were investigated linked to their respective population ranges; and 2) LULC-specific O 3 stomatal uptake 𝐹 ! in the format of Phytotoxic Ozone Dose above the critical level of 𝑦 275 nmol m -2 s -1 (POD y ), as well as crop-specific AOT40, were evaluated based on equations (15) and (16). and DA results based on different LSMs dry deposition 3 health. O 3 –SM transport multiplicative dry deposition schemes. Our Noah_W and P1_W related results indicate the influences of SM on air quality via its feedbacks to weather; and results from the Noah_D and CLM_D regarding both the indirect via adjusting phenology and and direct SM effects on O 3 . The complex SM impacts on O 3 dry deposition as well as surface O 3 concentrations based on the coupled photosynthesis-R s calculations rely heavily on the application of water stress function (β scheme), soil properties and hydrological regime. The WRF-Chem results from this case indicate that, to more accurately simulate MDA8, land DA be combined aggressive to identify other of uncertainty in O 3 modeling chemistry, and extra-regional pollution contributions) and reduce their negative impacts on model performance.


Introduction
Ground-level ozone (O3) is a regulated secondary air pollutant harmful to human and ecosystem health (Fleming et al., 2018;35 Mills et al., 2018a,b). It is closely connected with O3 at higher altitudes where O3 plays a more important role in the Earth's climate system. To better protect human health and public welfare, in 2015, the US primary and secondary National Ambient Air Quality Standards were lowered from 75 ppbv to 70 ppbv, in the format of daily maximum 8-h average (MDA8). Several other O3-exposure based metrics have also been applied or/and proposed to assess O3 impacts on vegetation, such as the accumulated O3 exposure over given thresholds (e.g., SUM40, SUM60, and AOT40), the averaged O3 exposure during 40 daylight hours (e.g., M7 and M12), and the sigmoidal-weighted W126 cumulative exposure (e.g., Fredericksen et al., 1996;van Dingenen et al., 2009; Hemispheric Transport of Air Pollution, 2010, and references therein; Avnery et al., 2011;Hollaway et al., 2012;Huang et al., 2013;Lapina et al., 2014;Mills et al., 2007Mills et al., , 2018a. To help comply with the evertightening air quality standards, an improved understanding of the individual processes affecting (near-)surface O3 concentrations and exceedances is demanded. Many O3-related processes are highly sensitive to environmental and/or 45 biophysical conditions (Huang et al., 2021, and references therein). These O3-related processes include dry deposition of O3 and its precursors, which is a major sink for near-surface O3 and depends on dry deposition velocities (Vd) and the deposited chemicals' concentrations. As recognized in numerous studies, accurately estimating dry deposition fluxes is critical to understanding O3 budgets and exceedances in the past, present, and future (e.g., Stevenson et al., 2006;Griffiths et al., 2021); it could also contribute to a more reasonable assessment of the O3 impacts on vegetation (e.g., Mills et al., 2011;50 Lombardozzi et al., 2015;Mills et al., 2018b;Ducker et al., 2018;Ronan et al., 2020), which is relevant to the budgets of other greenhouse gases as well.
Ozone uptake by plants is generally higher in warm/growing seasons and during the daytime when O3 concentrations and Vd values peak. As introduced in Huang et al. (2021) as well as references therein, over the land, surface resistance Rc, which is 55 composed of stomatal-mesophyll (Rs-Rm), cuticular (Rlu), in-canopy, and ground resistance terms, often exerts the strongest effects on the magnitude and variability of Vd. Vd also includes the aerodynamic resistance (Ra) and quasi-laminar sublayer resistance (Rb) terms.
using available observations to constrain (some of) the model land variables; and 3) including a wide range of observations and/or observation-derived carbon, water, and energy fluxes as well as vegetation states in model evaluation for broad geographical regions. Furthermore, it is important to explicitly connect the progress in dry deposition modeling with the impact assessments of O3 and other air pollutants on ecosystem health, productivity, and diversity. 105 A regional-scale land modeling and SM data assimilation (DA) framework coupled with weather and atmospheric chemistry modeling by the Weather Research and Forecasting model with online Chemistry (WRF-Chem) is implemented in this work.
Using this tool, we quantify and discuss the responses of Vd and its key components as well as O3 concentrations and plant uptake to SM changes due to the DA, for different soil texture, LULC and crop types. The central parts of this work rely on the Noah-Multiparameterization (MP, Niu et al., 2011) LSM with dynamic vegetation that enables the implementation of a 110 modified "dynamic" dry deposition scheme. With this modified scheme, both the indirect (i.e., via changing weather and vegetation fields) and direct effects of SM on dry deposition are considered in this modeling system. Results based on this modified and the WRF-Chem default Wesely schemes are compared and evaluated with independent datasets. As an extended work of Huang et al. (2021), here we continue to focus on the southeastern US during summer 2016 for which period prior Noah-and Wesely-based model calculations were conducted and aircraft observations are available. This 115 manuscript introduces the applied two dry deposition schemes in Section 2. It then presents results from this Noah-MP based modeling system, in comparison with those from Huang et al. (2021) (Sections 3.1-3.2). Discussions on O3 concentrations and fluxes based on all related WRF-Chem simulations are also connected with the assessment of O3 impacts on societies, ecosystem health, and crop yield (Section 3.3). Summary and suggestions on future directions are provided in Section 4.

Modeling and DA experiments design
The modeling tools and DA experiment design of this study were largely consistent with the Huang et al. (2021) study: we conducted model simulations over the southeastern US in a semi-coupled Land Information System (LIS)/WRF-Chem system without and with the assimilation of the enhanced SM retrievals from NASA's Soil Moisture Active Passive (SMAP) mission. Two dry deposition schemes were applied in cases without and with the SM DA. The model domain, horizontal and 125 vertical resolutions, atmospheric/land initialization and SM DA methods were adapted from our previous study based on the Noah LSM. Major model input datasets, physics and chemistry schemes were kept similar as before except a few aspects relevant to the upgrade of LSM from Noah to Noah-MP (version 3.6) and the implementation of an irrigation scheme to be introduced in Section 2.2.

130
Same as in Huang et al. (2021), the LULC and soil texture type inputs of our coupled modeling system were based on the International Geosphere-Biosphere Programme-modified Moderate Resolution Imaging Spectroradiometer and the State Soil Geographic datasets, respectively. Crop type data from Monfreda et al. (2008) were used in the irrigation scheme and the assessment of the O3 impacts on vegetation (Figure 1b), which are roughly consistent with the 2016 records from the US Department of Agriculture National Agricultural Statistics Service for several major crops such as maize, soybean and wheat 135 (https://nassgeodata.gmu.edu/CropScape, lass access: 8 November 2021). In Section 3 of this paper, model results are summarized and/or discussed by groups of grid-dominant LULC and soil type that are shown in Figure 1(a;d). The original 20 LULC types were grouped into urban and non-urban areas, and for vegetation-dominant areas, into forests, croplands, and shrub/grasslands, following the criteria introduced in Table S1. The grid-dominant LULC groups for vegetated regions used in our analysis are vastly similar to independently-developed data products, e.g.: a dataset derived from the European Space 140 Agency-Climate Change Initiative Land Cover project (https://gwis.jrc.ec.europa.eu/apps/country.profile/overview/USA, lass access: 8 November 2021), and the 2016 National Land Cover Database. Urban-dominant grid cells are well aligned with dense population areas ( Figure 1c) based on the Gridded Population of the World version 4.11 (NASA Socioeconomic Data and Applications Center, 2018). Grid-scale discrepancies exist between the used LULC input and independent LULC products, which, however, are not anticipated to considerably impact the results averaged by LULC groups. Three groups of 145 soil are highlighted, namely sand/loamy sand, loam and clay. The original sand and loamy sand categories are combined because of their high sand fractions (http://www.soilinfo.psu.edu/index.cgi?soil_data&conus&data_cov&fract&methods, lass access: 10 December 2021).

Physics and configurations of the Noah-MP LSM
The Noah-MP LSM includes a number of improvements from Noah, and one of the enhanced features in Noah-MP is that it 150 contains a separate canopy layer that explicitly computes photosynthetically active radiation, canopy temperature, and related energy, water, and carbon fluxes so that it facilities a dynamic vegetation model. A modified two-stream radiation transfer scheme was used to compute fractions of sunlit and shaded leaves and their absorbed solar radiation. The Ball-Berry type of ! scheme (e.g., Ball et al., 1987) was applied as required by the dynamic vegetation option. When this option is used, green vegetation fraction (GVF) does not come from an input dataset as in Huang et al. (2021) but is related to LAI 155 based on (1): Niyogi and Raman (1997) concluded that Ball-Berry along with two other physiological schemes, performed better on ! than the multiplicative Jarvis type which has been frequently used with the prescribed vegetation option. Specifically, it helps better capture the variance in ! and is more responsive to environmental changes. As described in Appendix B of Niu 160 et al. (2011), this scheme relates stomatal resistance rs,i of sunlit and shaded leaves i to the photosynthesis rates (Ai) per unit LAI of sunlit and shaded leaves i separately:  (3)- (6): where Igs is a TV-dependent growing season index, AC, AL,i, and AS are carboxylase-limited, light-limited, and export-limited photosynthesis rates per unit LAI, respectively; ci and oi are CO2 concentrations inside leaf cavity which is about 0.7 times of 175 the atmospheric CO2 concentration and atmospheric O2 concentration, respectively. PAR represents the photosynthetically active radiation per unit LAI. ccp is the CO2 compensation point and it equals to 0.5 : ( : , 0.21 3 , where Kc and Ko are the Michaelis-Menton constants for CO2 and O2, respectively, varying with TV; is the quantum efficiency.
Vmax represents the maximum rate of carboxylation, expressed as: where Vmax is maximum carboxylation rate at 25°C; f(TV) is a function that mimics thermal breakdown of metabolic processes; f(N) is a foliage nitrogen factor; and β is the SM factor controlling Rs, which presents strong dependencies on soil type and hydrological regime. In this study model results based on the Noah and the Community Land Model (CLM) types of β schemes are compared (Table 1), the latter of which is known to often result in sharper and narrower ranges of variation with SM than the former does. 185 Other Noah-MP configurations include: the three-layer snowpack physics and the CLASS snow surface albedo; the Jordan scheme for partitioning precipitation into rainfall and snowfall; the Niu-Yang-2006 frozen soil permeability and supercooled liquid water option; the Simple Groundwater Model runoff scheme; and the Monin-Obukhov CH scheme, which, unlike Noah's default Chen97 scheme, accounts for the zero-displacement height. Being affected by stability correction and 190 additional effects of planetary boundary layer height on friction velocity, it is likely that the Monin-Obukhov scheme can result in either weaker or greater CH (i.e., less or more efficient ventilation of the land surface) than the Chen97 scheme during the daytime in summer (Niu et al., 2011;Yang et al., 2011). Irrigation process was included in all Noah-MP based simulations in this study. The benefit of including irrigation relies on 195 the choice and parameterization of the irrigation scheme, as well as the LSM model's inputs (Lawston et al., 2015). Sprinkler scheme was chosen as it was reported as the prevalent irrigation method in 2015 across the US and many of the states within our model domain (Dieter et al., 2018). Irrigation was triggered over irrigated land in growing season within local morning times (6-10 am) when rootzone SM drops below 50% of the soil field capacity. The irrigated land was determined by the model's LULC input and irrigation intensity information in Salmon et al. (2015), and the rootzone area was derived from the 200 maximum root depth, which varies by crop type and GVF.

Wesely and dynamic O3 dry deposition schemes
Dry deposition velocity D is estimated based on the resistance analogy approach: 2 and G are aerodynamic resistance and quasi-laminar sublayer resistance, respectively, sensitive to surface properties 205 such as surface roughness. Over the land, surface resistance 7 , the major component of D , is classified into stomatalmesophyll resistance ( ! -4 ), cuticular resistance ( HI ), in-canopy resistance ( D7 and 7H ), and ground resistance ( 27 and where D7 is resistance for gas-phase transfer affected by buoyant convection in the canopy when sunlight heats the (near-210 )surface, 7H is resistance for leaves, twigs, bark, and others in the lower canopy, 27 is resistance for transfer that depends mostly on canopy structure, and 6! is resistance for soil, leaf litter, snow, and others at the ground surface.
Two deposition schemes, namely the Wesely and a dynamic scheme, were applied in this study, in which ! and HI are treated differently. In the Wesely scheme, ! and HI are calculated based on (10) and (11): 215 *# 02 N9O 4 + 1000 "/ ! "= , for dry surfaces according to humidity and precipitation fields Where ! and HI are LULC-and season-dependent constants subject to uncertainty; G and ! are radiation and surface temperature, respectively, whose definitions are different than those of PAR and TV in equations (2)-(7); N 1 P and B are molecular diffusivities for water vapor and trace gas x (e.g., O3), respectively; H, which is sensitive to surface temperature, 220 represents the Henry's law constant for the focused trace gas; and # is a reactivity factor for oxidation. The Wesely-scheme related results new from this study and Huang et al. (2021) are compared (Table 1). As expressed in equations (12), in the dynamic scheme, ! used in dry deposition modeling was taken from ! calculated from Noah-MP's dynamic vegetation model, and thus considers the physiological process of leaf responses to photosynthesis 225 rate, humidity and CO2 concentrations. The direct effects of SM, as reflected in the β formula, as well as other environmental variables, are included in this method, and this work quantifies the impact of the β factor configurations in Noah-MP (Table   1) on the dynamic-scheme-related results.
In the dynamic scheme, HI for dry surfaces is modified from the Wesely formula by considering its LAI dependency: In both the Wesely and the dynamic schemes, D7 is sensitive to surface radiation, and 4 is expressed as: Similar to the HI calculations in equations (11) and (13), to approximate an effect that coldness sometimes reduces the uptake, 1000 "/ ! "= is added to LULC-and season-dependent constants to derive 6! and 7H . It is worth mentioning that the direct effects of water stress on mesophyll resistance have been recognized (e.g., Egea et al., 2011). Yet, in neither scheme we applied, such effects have been incorporated into the 4 formula as a part of the D calculation.

Model evaluation, analysis, and O3 impact assessments 240
For the cases listed in Table 1, we quantify the impacts of SM DA on the modeled SM, vegetation dynamics, surface fluxes (i.e., gross primary productivity, GPP, which is integrated by LAI from A in equations (2)-(3), energy fluxes and their partitioning in the format of evaporative fraction, dry deposition flux and individual Vd terms for O3 particularly ! and HI related), meteorological and surface O3 fields during the 16-28 August 2016 period. The SM DA impacts on most of these model fields are expressed as daily or/and daytime (~13:00-24:00 UTC) averaged absolute or relative changes referring to 245 the results from the no-DA cases. For O3 dry deposition fluxes, we also conducted linear regression analyses to determine the relationships between the relative flux changes versus the relative changes in column-averaged initial SM due to the DA.
Results of O3 dry deposition fluxes and the regression analyses (i.e., slopes, correlation coefficient r values, and p values) are summarized by grouped LULC types defined in Figure 1a. Agency Clean Air Status and Trends Network (CASTNET) and Air Quality System (AQS) sites; and 3) the 0.5°×0.5°, daily 255 FLUXCOM latent and sensible heat fluxes. New evaluation datasets used in this work include: 1) the 9 km VOD retrievals from the enhanced SMAP product, which indicates the attenuation of microwave signals by vegetation, proportional to above-ground canopy biomass, and was used together with a 10-day average Copernicus Global Land Service GVF product to derive GVF for the focused 13-day period; 2) the daily GPP estimates from the 9 km SMAP level 4 carbon (L4C) product version 6, developed based on the SMAP L4 surface (0-5 cm) and rootzone (0-100 cm) SM together with satellite LULC 260 and vegetation datasets, which was supplemented by two independent GPP proxies (Whelan et al., 2020) of satellite-derived solar-induced chlorophyll fluorescence (SIF) data (Yu et al., 2019) and the Portable Flask Package (Sweeney et al., 2015) carbonyl sulfide (OCS) measurements collected onboard the B-200 and C-130 aircraft during the ACT-America campaign, and other airborne trace gas (e.g., benzene) measurements during this campaign were analyzed together with the OCS data to help distinguish the influences of combustion sources from plant CO2 uptake on the observed OCS distributions; and 3) Vd 265 data from selected CASTNET sites, estimated using a multilayer model (MLM, not supported by CASTNET as of 2017) version 3.0 which has known limitations and biases against eddy covariance flux measurements as well as Vd estimated using other methods (e.g., Finkelstein et al., 2000;Saylor et al., 2014;Wu et al., 2018). The known limitations of MLM and how they may affect our model comparisons with the CASTNET Vd data are discussed. Our O3 dry deposition results are also compared with eddy covariance measurements reported in independent works for similar climate or/and LULC types 270 during other time periods.
This study also evaluates how the SM DA affected the assessments of surface O3 impacts on human and ecosystem health.
Specifically: 1) MDA8 O3 fields over unban and nonurban regions were investigated linked to their respective population ranges; and 2) LULC-specific O3 stomatal uptake ! in the format of Phytotoxic Ozone Dose above the critical level of 275 nmol m -2 s -1 (PODy), as well as crop-specific AOT40, were evaluated based on equations (15) and (16).
The calculated PODy and AOT40 were used to estimate the Relative Biomass Loss (RBL) or Relative Yield Loss (RYL) for several types of vegetation or crops based on dose-response functions reported in literature (  Mills et al., 2007Mills et al., , 2018a. Our 13-day WRF-Chem model results were linearly-extrapolated to ~three months to derive the PODy and AOT40 fields. We focus on qualitatively interpreting the results and discussing their implications. The outcome from this analysis is also compared with the findings from several independent O3 impact assessment studies for different time periods.  Table 1. At the surface layer (0-10 cm belowground), both cases produced SM horizontal gradients that resemble the Noah-based results presented in Huang et al. (2021). They are moderately correlated with the column-averaged 290 SM fields (r=0.875 and 0.871, respectively), and the mean differences in column-averaged and surface SM from the Noah_D and CLM_D cases are 0.003 and -0.006 m 3 m -3 , respectively. Kumar et al. (2009) have found that, compared to other LSMs such as the Catchment (based on which the SMAP L4 datasets are produced), the 4-soil-layer Noah and 10-soil-layer CLM LSMs display successively weaker surface-subsurface coupling strengths, and the weakest coupling strength of CLM was primarily attributed to its significantly larger number of soil layers. The slightly weaker surface-subsurface correlations in 295 the CLM_D case than in the Noah_D from this work, both are based on a 4-soil-layer Noah-MP modeling system, indicate the minor role of the LSM physics, in particular the β factor scheme, in controling the vertical coupling strength of SM conditions.
The modeled SM fields from the Noah_D and CLM_D differ on grid scale, particularly in the subsurface zones ( Figure 2a-300 b). For example, in sand-dominant regions that were experiencing drought conditions during this period (e.g., Florida and the Texas-Oklahoma border regions, where simulated SM is mostly under 0.2 m 3 m -3 ), column-averaged SM values from the CLM_D case are notably smaller than those from the Noah_D case. These results contrast with those reported by Niu et al. (2011), in which cases Noah-MP with the CLM-type β factor consumed less soil water, resulting in smaller SM variability than did the Noah-type β factor during drought periods. In their cases focusing on loam and clay soil that have higher wilting 305 points when the CLM-type β factor scheme was applied, plant transpiration ceased to save soil water under drought conditions. Our results can be explained by the steeper CLM-type β-SM curve than the Noah-type β-SM curve for low SM, sand-dominant areas, as illustrated in Figure 3a of Niu et al. (2011). For such conditions, Noah-MP with the CLM-type β factor produces stronger evapotranspiration (ET) and consumes more soil water, resulting in drier soil. For wet regions where SM values exceed 0.4 m 3 m -3 , such as Louisiana and Arkansas, the CLM-and Noah-type β values are close to 1.0 and 310 insensitive to soil type and SM variations; therefore, SM and ET produced from the Noah_D and CLM_D cases do not diverge. These findings corroborate the conclusions by Yang et al. (2011) that the degree of the β impacts on the SM-ET relationship should depend on the soil type and hydrological regime, and they are important for understanding the vegetation and surface flux results to be presented in the later parts of this paper.

315
Referring to the SMAP SM data, in general, surface SM produced by the no-DA modeling systems show wet biases in nonforested regions and dry biases over the forests for the study period. These SMAP-model discrepancies were successfully reduced by the DA for all vegetated LULC groups ( Figure S1), leading to overall slightly drier soil in DA-enabled simulations. For both the Noah_D and CLM_D cases, the DA adjusted the modeled SM fields across the entire soil columns, demonstrating that observational information at the surface was propagated into deep soil layers. The SM responses to the 320 DA as a function of soil layer from the Noah_D and CLM_D cases are roughly similar but different at small spatial scales, which reflect the controls of the β factor scheme on the surface-subsurface coupling strengths of the used modeling/DA system. With the SMAP DA enabled, the r values between column-averaged and surface SM from the Noah_D and CLM_D cases increased to 0.902 and 0.897, respectively.

325
The satellite-derived GVF fields (methods introduced in Figure S2 caption) transition from low-to-moderate (<0.6) to high (>0.8) values from the western (mostly shrub/grasslands) to the central and eastern parts (forests/croplands dominant) of the study region, and such spatial gradients are highly correlated with the SMAP VOD retrievals (Figure 3a;d). The Noah_D and CLM_D cases both moderately well reproduced these spatial patterns. Major differences between these cases are found in dry sandy regions, where, as discussed in previous paragraphs, more soil water was consumed for ET and plant growth in the 330 CLM_D case and therefore higher GVF values are given. The DA adjusted the modeled GVF and SM fields toward similar directions, with the relative changes in GVF overall smaller. While the SM changes in the Noah_D and CLM_D cases are of close magnitudes, GVF responded more strongly in the CLM_D case except for sandy regions. Referring to the satellitederived GVF fields which are also subject to large uncertainty, the modeled vegetation fields are more effectively improved by the DA over sparsely vegetated regions such as the South-Central Plains. The DA also remarkably reduced the model-335 satellite mismatches over some of the dense vegetation regions such as the southwestern Ohio. The likely degraded model performance over certain dense vegetation areas can be partially explained by weaknesses related to SM-vegetation growth feedbacks in the dynamic vegetation model parameterizations which need to be identified and addressed in future work. It is also suggested that jointly assimilating satellite SM and vegetation phenology products such as the VOD retrievals needs to be experimented which may maximize the positive DA impacts on multiple land variables and their atmospheric feedbacks. 340 Figure 4 compares the spatial distributions of the period-mean WRF-Chem carbon and energy fluxes with SMAP L4C and FLUXCOM products which contain observation information, and Table 3 summarizes WRF-Chem and observation-derived flux results by three LULC groups. The observation-derived products indicate the highest GPP and evaporative fraction (EF 345 = daily latent heat/(daily latent heat + daily sensible heat)) over croplands. Without the DA, the Noah-MP related cases outperformed the Noah related P1_W case on simulating EF, especially over shrub/grassland and cropland regions. This indicates that, from Noah to Noah-MP, the multiple updates in LSM physics related to Rs, irrigation and CH, are beneficial.

Carbon/energy fluxes and weather conditions
Larger GPP and EF values are found in CLM_D than in Noah_D, most of which match better with the SMAP L4C and FLUXCOM data. The DA led to increased EF over shrub/grasslands in all model cases as well as over croplands in the 350 Noah_D case, bringing the model results closer to the FLUXCOM data. The EF values were unfavorably reduced by the DA in the CLM_D and P1_W cases over croplands and in all model cases over forests, reflecting the challenges of satellite SM DA over regions with dense vegetation and/or affected by human activities, which have also been reported in previous studies. For the Noah_D and CLM_D cases, this may also be due to the possibly degraded vegetation performance discussed in Section 3.1. The modeled GPP in the CLM_D cases were lowered by the DA overall, which helped reduce the model-355 SMAP L4C discrepancies over forests and croplands. In the Noah_D case, GPP was improved by the DA over forests and (slightly) over shrub/grasslands. Based on the evaluation statistics, for this case, the CLM type of β factor scheme is shown slightly superior to the Noah type. Note that the quality of the SMAP L4C and FLUXCOM products may also be strongly LULC dependent, e.g., it has been known that the uncertainty of SMAP L4C data is generally larger for highly productive plant functional types (Kimball et al., 2020). Such evaluation, therefore, has demonstrated the critical role of LULC type in 360 understanding the model performance of carbon and energy fluxes and its responses to satellite SM DA.
Additional datasets were also utilized to help understand terrestrial carbon uptake, including satellite SIF and ACT-America aircraft OCS as well as its vertical gradients ( Figure S3). Consistent with the SMAP L4C and WRF-Chem based results, the largest SIF values are shown over croplands, especially maize and soybean fields in Illinois and Indiana, 2-3 times as high as 365 those over shrub/grasslands in the South-Central Plains. Free-tropospheric (>2 km, above ground level, a.g.l.) OCS mixing ratios are mostly higher than those near the surface (≤2 km, a.g.l.), except for locations that may be strongly influenced by oceanic and anthropogenic combustion sources according to independent studies (Lennartz et al., 2017;Zumkehr et al., 2018) and other chemical tracers (e.g., benzene) measured onboard. The maximum OCS mixing ratios are higher than 550 pptv, and the OCS drawdowns (i.e., free-tropospheric minus near-surface concentrations) far exceed 60 pptv around the 370 Lower Mississippi cropland regions and the Texas-Oklahoma border where soil was wet and likely an OCS source (Bunk et al., 2017). These values are much larger than those observed in summer 2004 over the eastern US (Campbell et al., 2008), indicating possible higher OCS emissions and stronger terrestrial carbon uptake in summer 2016 than in summer 2004.
In general, the modeled EF fields as well as their directions of changes due to the DA resemble those of latent heat flux and 375 relative humidity (RH), which are opposite to those of sensible heat and surface temperatures (Figures 5 and S4). The model overall well reproduced the observed spatiotemporal variability of 2 m air temperature (T2) and RH, as well as FLUXCOM latent and sensible heat fluxes. The diagnostic 2 m weather fields and their responses to the DA strongly correlate with the model results at its surface level. The Noah-MP related cases reacted more strongly to the DA than the Noah-related cases, with the responses in the CLM_D case larger than in the Noah_D case except for dry, sandy regions, which can be attributed 380 to combined effects of the used CH and stomatal resistance schemes. It is important to note that diagnostic temperature and humidity variables are represented differently in Noah and Noah-MP and thus are not directly comparable. Specifically, in Noah, T2 is an explicit function of skin temperature, air density, specific heat of dry air at constant pressure, and 2 m surface exchange coefficient for heat, and 2 m specific humidity is a function of surface specific humidity, upward moisture flux at the surface, air density and 2 m surface exchange coefficient for moisture; whereas in Noah-MP, they are expressed as 385 functions of temperatures and water vapor for vegetated land and bare soil being weighed by their respective fractions. We therefore focus on quantitatively evaluating and intercomparing prognostic model weather variables (i.e., the model-level air temperature and humidity) against ACT-America aircraft observations ( Figure 6). For air temperature, at all altitudes and near the surface (i.e., ≥800 hPa), the CLM_D case responded most strongly to the DA, and the DA-enabled CLM_D case outperformed the Noah_D and P1_W cases. This performance is qualitatively consistent with the model's sensible heat 390 performance referring to the FLUXCOM data. As for humidity, despite the most significant DA improvements in CLM_D, the Noah-MP related cases did not perform as well as the Noah related cases, which is also found in the model's latent heat performance in comparison with the FLUXCOM data. However, note that the model's humidity performance is more strongly related to that of Rs and Vd in the Noah-MP based cases via the direct impacts of humidity on Rs calculations (equation 2). The solar radiation fields from all model cases, which play vital roles in controlling the land-atmosphere 395 exchanges of water and trace gases, do not differ dramatically and their responses to the DA are negligible (e.g., Figure 5gl). This indicates that the DA impacts on the modeled surface fluxes resulted primarily from the changes in the modeled SM, humidity, surface/canopy temperatures, as well as vegetation fields. Figure 7 presents the period-mean, daily-averaged Vd and dry deposition flux Ft (i.e., Vd multiplied by concentration at the 400 surface level, Wesely, 1989) for O3 from all model cases, along with their responses to the SMAP DA. The daytime averages of these fields have similar spatial gradients but of larger magnitudes (not shown in figures). Table 3 summarizes for three LULC groups the daily-and daytime-averaged results. The modeled stomatal-mesophyll and cuticular conductances, as well as their diurnal variability are indicated in Figure 8. All model cases produced lower Vd and Ft values over shrub/grasslands than over forests and croplands, qualitatively consistent with results from many existing model-and measurement-based 405 studies. The results from Noah_W and P1_W, both of which are based on the same scheme (Wesely), are generally similar, with minor differences largely attributed to different surface temperature fields (Figures 5 and S4). The WRF-Chem modeled Vd and Ft fluxes were more strongly affected by the upgrade from the Wesely to the dynamic scheme: i.e., with the updated scheme, they show enhanced magnitudes, stronger spatial variability, as well as more intensive responses to the DA, especially over forests and croplands. These results can be mainly explained by the fact that the stomatal-mesophyll and 410 cuticular resistances in the dynamic scheme are sensitive to more environmental and biophysical variables, accounting for both the direct and indirect (i.e., via influencing the weather fields and plants' physiology) effects of SM on Vd. Vd from the Noah_D and CLM_D cases, as well as its major term stomatal-mesophyll conductance, show strong correlations with the modeled GPP, latent heat, and EF fields which have been discussed in earlier sections. Comparing the cases that implemented the CLM-and Noah-like β schemes, O3-related fluxes resulting from the former configuration are of notably 415 larger magnitude, spatial variability and absolute changes due to the DA. The SM impacts on the modeled Vd and Ft were further quantified using linear regression analyses between the relative changes in the modeled O3 fluxes due to the DA versus those in column-averaged SM initial conditions. All regression models yielded low p values (i.e., <<0.01), suggesting good ΔVd~ΔSM and ΔFt~ΔSM relationships. The regression slopes are summarized in barplots (Figure 9) by three LULC groups for all model cases in Table 1. For all LULC groups, the slopes based on the two cases that implemented the dynamic 420 scheme are 2-3 times larger than those from the two cases using the Wesely scheme, and the slopes differ most strongly among cases over forests and croplands. The low r values associated with several regression models (denoted by red "L"s in Figure 9) reflect the stronger nonlinear relationships between the changes in the studied O3 fluxes and SM. These results emphasize the importance of better understanding and realistically representing in models the SM control on plants' stomatal behaviors which regulate the land-atmosphere exchanges of water, energy, and trace gases. The earlier evaluation of the 425 period-mean GPP and ET across the domain have demonstrated some advantages of using the CLM-like β scheme, and that the DA more effectively improved the model performance in sparsely vegetated shrub/grassland regions. These conclusions are likely also applicable to the modeled O3 dry deposition process, particularly its stomatal-mesophyll pathway.

Ozone dry deposition velocities and fluxes
In all no-DA and DA cases, the diurnal variability of O3-related surface fluxes shows clear LULC dependency. Over the 430 shrub/grassland and forests/croplands regions, the daytime averaged Vd values are 24-31% and 35-50% higher than the 24 h mean, respectively, while the daytime averaged Ft results are 40-50% and 42-63% higher than the 24 h mean, respectively (Table 3). Such Vd diurnal cycles are a result of the strongest diurnal variability in stomatal-mesophyll conductance (i.e., its daytime mean values are ~twice as high as the 24 h mean for all LULC types) being balanced out by weak diurnal variability associated with other Vd terms. As the most diurnally variable Vd component, stomatal-mesophyll conductance, on average, 435 contribute less substantially to Vd for shrub/grassland areas (24 h/daytime: up to ~30%/40%) than for forests/croplands (24 h/daytime: up to ~50%/66%), which helps explain the weaker diurnal variability in the modeled Vd over shrub/grasslands. The stronger diurnal cycles in Ft than in Vd reflect the impacts of higher daytime O3 surface concentrations used in the Ft calculations. The DA did not dominantly intensify or dampen the diurnal cycles of these fluxes for any given grouped LULC type. Whether the DA improved the estimated diurnal cycles of fluxes for various LULC types remains to be evaluated, 440 which can benefit from independent observation-constrained flux products of broad spatial coverage and subdaily variability.
A detailed analysis was then conducted at two forest CASTNET sites with different soil types and hydrological regimes. The were produced from the CLM_D case, followed by the Noah_D and Noah_W cases, which are 2-3 times as high as the MLM-estimated. The fluxes from all WRF-Chem cases during the nighttime are close, up to >80% lower than their daytime maxima, but still dramatically higher than the MLM-based results which are nearly zero. Our nighttime Vd results are close to flux observations at European forest sites during both dry and wet periods in the past decades (Lin et al., 2020). Wu et al. 455 (2018) compared Vd observations with model calculations based on the operational MLM, Wesely, and a dynamic scheme Noah-GEM, at a Canadian mixed forest site dominated by sand-like soil. Their diverse model results are qualitatively consistent with our findings at these two CASTNET sites. The remarkably lower Vd values from the operational MLM calculations are attributed to its simplified approaches of calculating Ra and Rb using wind speed and direction, the empirical approach of calculating Rs, and the lack of continuous, accurate model input data. Within the respective ranges of the 460 modeled SM at these two sites, β factors based on the CLM-type scheme are both larger than those based on the Noah-type β scheme (referring to Niu et al., Figure 3), which helps explain the higher model fluxes from the CLM_D case than the Noah_D case without the DA. At the SUM156 site, despite the strongest SM decrease (~0.04 m 3 m -3 ) by the DA in case CLM_D, the modeled fluxes responded least significantly to the DA, in part due to the flattened CLM-type SM-β curves in contrast to the linear Noah-type SM-β function for sand within the 0.12-0.16 m 3 m -3 SM range. At the PED108 site, the 465 modeled SM values from all model cases were lowered by the DA by ~0.02 m 3 m -3 . The stronger flux reactions to the DA from the CLM_D case than those from the Noah_D case can be partially explained by the steep CLM-type SM-β curve versus the linear Noah-type SM-β relationship for loam within the 0.18-0.22 m 3 m -3 SM range. Case studies at these two sites with the same type of LULC emphasize the importance of soil type and hydrological regimes for understanding SM controls on dry deposition, which was often omitted or underdiscussed in previous dry deposition studies. 470 Figure 11 illustrates the impacts of the choice of dry deposition scheme and SM DA on WRF-Chem modeled surface MDA8 O3. During the study period, several warmer-and drier-than-normal Atlantic states experienced high MDA8 at times (i.e., ≥60 ppbv, which can negatively affect lung function, and at ≥70 ppbv, cause respiratory symptoms and other adverse effects, 475 Fleming et al., 2018, and references therein). Numerous populated urban centers reside in these areas. The levels of MDA8 are shown to be much lower (i.e., <40 ppbv) over the southern part of the domain, including several major urban/suburban regions such as the Texas Triangle, which was frequently influenced by passing cold fronts and tropical systems from the Gulf of Mexico.

480
All model cases reproduced these MDA8 spatial patterns moderately well. Referring to observations at AQS and CASTNET sites, their domain wide mean RMSEs all fall within 6-8.5 ppbv (Figure 11m). We first intercompare the MDA8 levels from all no-DA cases. Positive and negative differences between the results from Noah_W and P1_W, both of which implemented https://doi.org/10.5194/acp-2021-1068 Preprint. Discussion started: 20 January 2022 c Author(s) 2022. CC BY 4.0 License. the Wesely scheme, are almost equally distributed across the domain, with the MDA8 from the former case associated with negligibly lower RMSEs (i.e., <0.02 ppbv on average) referring to AQS and CASTNET observations (Figure 11l-m). The 485 differences between these two cases are largely due to the impact of the chosen LSM on the model's meteorological fields, particularly temperatures, which affected the simulations of various O3-related processes including dry deposition. As Figure   11 (j;k;m) show, replacing Wesely with the dynamic dry deposition scheme considerably lowered the calculated MDA8 levels in majority of the model grids, as well as their associated RMSEs (i.e., by >0.5 ppbv on average) relative to surface observations. These reductions in MDA8 are of comparable magnitudes with those due to updating anthropogenic emissions 490 from the National Emission Inventory 2014 to 2016beta (Huang et al., 2021). Comparing the implementations of CLM-and Noah-types of β schemes, the former led to stronger reductions in the modeled MDA8 fields and their associated uncertainty.
These results reflect the impacts of the faster O3 removal via dry deposition in the dynamic scheme related cases, as well as the different model meteorology. Our findings are qualitatively consistent with the conclusions from several global-scale modeling experiments that compared the Wesely and dynamic schemes (e.g., Val Martin et al., 2014;Lin et al., 2019). 495 In all model cases, the DA reduced surface and subsurface SM in many of the grids, leading to enhanced MDA8 (Figure 11fi). The responses of the period-mean MDA8 to the DA from the Noah_W and P1_W cases are mostly within ±4 ppbv. When the dynamic dry deposition scheme was applied, the modeled MDA8 responded several times more strongly to the DA (i.e., by up to 6 ppbv and 8 ppbv in the Noah_D and CLM_D cases, respectively), especially over nonurban regions where surface 500 MDA8 on average are several ppbv lower than in urban grids. In urban grids where population densities are ~25 times higher than in nonurban grids (Figure 1c), the DA impacts on MDA8 reach 3-4 ppbv in places, under the controls of the local-toregional circulation patterns (Figure 12a;e). As the no-DA cases are positively biased against surface observations in many places, corresponding to the DA-induced surface O3 changes, the overall model performance of MDA8 was not improved, or much degraded, by the DA. Over limited areas such as the South-Central Plains, the modeled MDA8 decreased due to the 505 DA by up to >2 ppbv, corresponding to improved performance. The no-DA and DA results based on different LSMs and dry deposition schemes confirm that drier soil conditions exacerbate O3 air pollution, which, together with heat stress, threatens human health. Such O3-SM relationships have also been demonstrated by Falk and Søvde Haslerud (2019) and Anav et al. (2018) using other chemical transport models and multiplicative dry deposition schemes. Our Noah_W and P1_W related results indicate the influences of SM on air quality via its feedbacks to weather; and results from the Noah_D and CLM_D 510 cases provide valuable information regarding both the indirect (i.e., via adjusting vegetation phenology and weather conditions) and direct SM effects on O3. The complex SM impacts on O3 dry deposition as well as surface O3 concentrations based on the coupled photosynthesis-Rs calculations rely heavily on the application of water stress function (β scheme), soil properties and hydrological regime. The WRF-Chem results from this case indicate that, to more accurately simulate MDA8, improving land DA must be combined with aggressive efforts to identify other sources of uncertainty in O3 modeling (e.g., 515 emissions, chemistry, and extra-regional pollution contributions) and reduce their negative impacts on model performance.

Implications for O3 vegetation impact assessments using concentration-and flux-based metrics
Both O3 flux-and concentration-based metrics have been applied to assess O3 impacts on vegetation as well as the associated economic loss. Estimating the plants' stomatal O3 uptake Fs is the basis for constructing flux-based O3 impact assessments. Figure 13 illustrates the period-mean daytime Fs fields based on all WRF-Chem no-DA cases as well as their responses to 520 the SM DA. Box-and-whisker plots in Figure 12(b;f) summarize these results by three LULC groups. The averaged Fs values for all three LULC groups exceed their respective critical levels (i.e., 1 nmol m -2 s -1 for forest and grasslands; and 3 nmol m -2 s -1 for crops). As a major contributor to O3 dry deposition flux during the daytime, Fs fields appear to be closely correlated in space and time with the surface humidity and flux fields (e.g., GPP, latent heat and EF, as well as Vd), which differ distinctly from the surface O3 concentration fields. For example, Fs hotspots are shown over some low O3 concentration areas 525 including the humid, Lower Mississippi River regions, and the lowest Fs values occur in certain high O3 concentration regions strongly affected by urban pollution (e.g., Georgia) and pollution transport from upwind US states and/or the stratosphere (e.g., western Kansas and Oklahoma, as discussed in Huang et al., 2021). The changes in Fs and surface O3 concentrations due to the DA show opposite directions, i.e., drier soil enhances surface O3 concentrations whereas slows down the plants' stomatal O3 uptake (Figures 11f-i and 13e-h). This comparison highlights how the choice of O3 metrics can 530 affect the assessment of O3 vegetation impacts under the changing climate. As emphasized by Mills et al. (2018a) and Ronan et al. (2020), flux-based metrics have evident advantages over concentration-based metrics. To conduct reliable impact assessments using these flux-based metrics, accurate information on stomatal and non-stomatal fluxes as well as the various environmental and biophysical variables that they are sensitive to become increasingly important.

535
An assessment of O3 vegetation impacts was conducted based on the results from various model cases and different metrics, namely PODy (where y is LULC-dependent critical level) and AOT40. For this demonstration, the 13-day model results were linearly extrapolated to ~three months. This also assumed similar DA adjustments to SM dynamics (driven by factors such as clouds/radiation, rainfall, and irrigation for cropland-dominant regions) at the time scale of ~three months. Based on the known seasonal variability of surface O3 and Vd in the study region, the linearly scaled PODy and AOT40 values may have 540 been overall underestimated. We therefore focus on discussing the results qualitatively and highlighting their implications for O3 impact assessments using long-term records. Statistics of the derived PODy and AOT40 fields are summarized by O3 sensitive LULC and crop types in Figure 12(c-d;g-h), and Figure 14 presents the estimated AOT40 fields as well as their responses to the SM DA for cropland-dominant grids. The highs and lows in AOT40-related results are found over maize and wheat dominant fields, respectively. Among the three focused LULC types, the highest and lowest PODy values are 545 estimated for forests and grasslands, respectively. Largely driven by daytime peak O3 concentrations, the spatial variability and biases (referring to AQS and CASTNET observations) of the model-derived AOT40 fields, as well as their responses to the DA, match the MDA8-based (Figure 11). In contrast, the spatial variability of PODy and Fs aligns well, so are their responses to the DA. Both PODy and AOT40 reacted several times more intensively in the cases that implemented the dynamic dry deposition scheme, especially the CLM_D case. 550 For selected LULC and crop types, the WRF-Chem derived PODy and AOT40 fields were used together with dose-response functions in literature to evaluate the RBL/RYL due to O3 exposure and uptake. As reported in Figure 12(c;g), with the SM DA enabled, the mean RBLs based on Noah_D and CLM_D derived PODy are 0.06-0.10, 0.02-0.03, and 0.04-0.05 for deciduous forest, grasslands and wheat, respectively, which are >33% lower than the Noah_W and P1_W based RBL 555 estimates. It is shown that, in response to the DA which lowered SM in many places, the Noah_W and P1_W based RBL estimates did not drop as strongly as the Noah_D and CLM_D based, and even increased for grasslands and wheat. For wheat, one of the most O3-sensitive crops, the estimated RYL values based on PODy and AOT40 approaches differ by up to a factor of 2-3, and the DA had contrasting effects on these estimates (Figure 12c-d;g-h). The PODy-and AOT40-based RYL values differ more significantly when the model-derived PODy and AOT40 fields came from the Noah_D and CLM_D 560 cases. Using the model-derived AOT40 and different AOT40 dose-response functions (Mills et al., 2007(Mills et al., , 2018a Table 2), the estimated RYLs and their changes due to the DA are nonnegligible (Figure 12d;h). Our estimated RBL/RYL results for various LULC and crop types mostly fall within the ranges reported in previous studies which applied model-derived O3 metrics and dose-response functions (e.g., Avnery et al., 2011;Mills et al., 2007Mills et al., , 2018a. Our results emphasize that the selected O3 impact assessment metrics for various LULC/crop types and their matching dose-response functions, as well as 565 the model results used to derive the chosen O3 metrics which are sensitive to dry deposition schemes and SM, all introduce uncertainty to the estimated O3 impacts on vegetation. The widely-used dose-response functions are considered appropriate for studying North America and Europe, but they may not be applicable to other regions (Emberson et al., 2009). Therefore, updating and developing dose-response relationships for a larger number of vegetation types in different regions of the world are needed, which may require new experiments to be conducted. Yue and Unger (2014) and Lombardozzi et al. (2015) as 570 well as follow-on investigations parameterized the O3 impacts on several types of vegetation using the relationships between cumulative O3 uptake and O3 damage factors for photosynthesis and conductance from empirical and experimental studies.
Based on multi-decadal model simulations, they reported <20% changes of biomass, GPP, and energy fluxes due to O3, which are roughly consistent with our RBL/RYL results in Figure 12. Such approaches that dynamically assess the impacts of O3 along with other factors (e.g., non-O3 pollutants and environmental stresses), as highlighted in Emberson et al. (2018), 575 will be considered in future work.
We note that, revising the dry deposition scheme and constraining the modeled SM fields with observations would not only better be combined with adding O3 injury to vegetation but also multistress impacts on biogenic emissions. Considering O3 injury to vegetation would affect more evidently longer-term climate simulations via feedbacks to biomass, surface fluxes, 580 weather and weather-driven emissions. As for biogenic emissions, Figure S5 shows SM anomalies during the study period determined by our Noah-MP modeling system as well as drought stress activity factor γd estimated from β of a multiyear, independent CLM simulation by Jiang et al. (2018). Based on these, we estimate that, depending on soil type, hydrological regime, as well as β configurations, omitting the direct impacts of water stress on biogenic emissions, may have introduced larger uncertainty to biogenic emission and O3 modeling over several states experiencing drier-than-normal conditions, such 585 as Tennessee, South Carolina, Alabama, and West Virginia. Quantitatively understanding the interplay between these processes and O3 pollution levels is recommended for more accurate air quality modeling and O3 impact assessments.

Summary and suggestions on future directions
This paper described a follow-up study of Huang et al. (2021). It presented how the choice of O3 dry deposition scheme affected our evaluation of SMAP SM DA impacts on coupled WRF-Chem modeling over the southeastern US in August 590 2016. In new Noah-MP LSM related simulations, two dry deposition schemes were implemented, namely the WRF-Chem default Wesely scheme and a dynamic scheme, in the latter of which the calculation of Vd (particularly its stomatal and cuticular terms) were modified to be coupled with photosynthesis and vegetation phenology. We showed that dry deposition parameterizations significantly affected the modeled O3 dry deposition process, as well as its response to the DA. Comparing the no-DA cases, it was found that, when the dynamic scheme was applied, overall, the modeled O3 dry deposition velocities 595 and fluxes were larger and surface O3 concentrations were lower. The modeled O3 fluxes responded 2-3 times more strongly to the SM changes due to the DA, which can be mainly explained by the fact that both the direct and indirect (i.e., via influencing weather and vegetation fields) effects of SM on O3 dry deposition modeling are considered in the dynamic scheme. Depending on soil type and hydrological regime, the selection of SM factor controlling Rs (i.e., β factor, a key variable representing the direct effects of SM on the modeled surface fluxes) scheme can strongly affect the quantitative 600 results. The Wesely-scheme derived dry deposition results driven by meteorological fields from Noah-MP and Noah (from Huang et al., 2021) LSM based WRF-Chem simulations displayed much smaller differences than those due to updating the dry deposition parameterizations. While we note that accounting for physiological effects in dry deposition modeling can be beneficial, the Ball-Berry Rs scheme applied in land surface and dry deposition modeling in this work needs to be compared with other semi-empirical Rs schemes, for a better understanding of their respective strengths and weaknesses. Alternative 605 schemes include the Medlyn scheme which has been integrated into the CLM version 5. Model intercomparison efforts such as the ongoing AQMEII4 activity (Galmarini et al., 2021) can also help determine areas for improvement in commonly-used dry deposition modeling approaches for studying 2016 and other years, over North America and other regions of the world. By analyzing the model responses to the SM DA from these various cases, we conclude that, in coupled modeling systems 610 that consider the direct and indirect influences of SM on O3 dry deposition, the accuracy of SM is particularly critical to dry deposition and O3 modeling, as well as the scientific analyses and impact assessments based on model simulations. The usefulness of SM DA for improving the modeled state and flux variables was evaluated by multiple observation (-derived) data products. Referring to in situ measurements, key meteorological variables relevant to Vd calculations such as surface temperature and humidity are shown to be improved by the DA by up to ~9%. Referring to satellite(-derived) datasets which 615 may be associated with high uncertainty, the model performance of vegetation phenology, GPP, as well as energy fluxes and their partitioning, showed mixed, LULC-dependent reactions to the DA. According to the evaluation statistics, for this case, the CLM type of β factor scheme was slightly superior to the Noah type. The modeled carbon and energy fields as well as their DA-related changes, correlated strongly with the modeled Vd fields, implying that the DA impacts on the accuracy of Vd were also possibly complicated which is difficult to verify due to the lack of high-accuracy, independent Vd evaluation 620 datasets. Observation(-derived) Vd datasets covering diverse LULC types nested in broad geographical regions and through more recent periods are in strong need. In places, the likely ineffectiveness of SM DA on vegetation and surface fluxes can not only be attributed to the quality of satellite SM retrievals and the used DA approach as discussed in previous Noah LSM based DA experiments, but also shortcomings in the Noah-MP LSM and its dynamic vegetation scheme regarding its surface-subsurface coupling and representation of SM-vegetation growth feedbacks. Continued efforts on advancing land 625 measurement/retrieval skills, identifying and addressing deficits in LSMs as well as practicing multivariate land DA are recommended in future work.
This study also demonstrated that, model-driven assessments of O3 impacts on human health and various types of vegetation can be significantly affected by the applied O3 dry deposition scheme, the implementation of land DA, the chosen O3 metrics 630 and their matching exposure-response functions. Various model cases showed that, the DA impacts on MDA8 were more evident in nonurban areas where the mean MDA8 was ~5 ppbv lower and the averaged population density is <1/25 of those in urban areas. Using concentration-and flux-based metrics AOT40 and PODy, the mean RYLs of maize, soybean, and wheat fell within ranges of 0.03-0.07, 0.10-0.17, and 0.04-0.14, respectively. While the multiple no-DA and DA cases helped us better understand the indirect or/and direct effects of SM on O3 dry deposition process, which has important 635 implications for O3 impact assessments, it is recognized that, the DA often exacerbated the positive O3 biases in free-running systems which has been a common issue shared by numerous regional and global models for this study region/season. It is necessary to combine land DA with efforts to identify and reduce other sources of uncertainty in O3 modeling. These should include reasonably representing the impacts of O3 along with other factors on vegetation as well as the direct impacts of water stress on biogenic emissions of volatile organic compounds and nitrogen species. 640