The main objective of this work is to investigate how correcting errors in simulated RH affects model outputs for AOT and extinction. NAAPS-RA AOT and extinction simulations were only evaluated for the 1∘ × 1∘ grid cells encompassing dropsonde release points (Fig. S2). To establish the “ground truth” dataset, HSRL-2 retrievals were extracted if they occurred anywhere within a 1∘ × 1∘ grid cell containing a dropsonde release. These retrievals were then filtered using the cloud buffer product to ensure the P-3 was not flying in cloudy conditions, while the HSRL-2 simultaneously retrieved data from below the plane. The remaining retrievals were filtered further to isolate retrievals obtained only when the P-3 was flying at a level altitude so that data were eliminated if the aircraft was either ascending or descending. Retrievals of total AOT remaining after these steps comprised the ground truth AOT dataset. Remaining retrievals of cumulative AOT were filtered one last time to eliminate cumulative AOT values with anomalously high absolute values.

NAAPS-RA reports simulated values at the center of each 1∘ × 1∘ grid cell, and these simulations are intended to represent the average conditions within that grid cell. This is problematic as the P-3 often only flew through sections of a 1∘ × 1∘ grid cell, which means the in situ data are not representative of the entire grid cell. To promote a fair comparison, NAAPS-RA model data for speciated mass concentrations and RH were interpolated to each location corresponding to a single HSRL-2 retrieval as well as to each location of a dropsonde release.

NAAPS-RA 532 nm extinction and AOT values were calculated using Eqs. (1)–(4) and (1)–(6), respectively, using the interpolated mass concentrations, interpolated RH values, and speciated 532 nm optical properties. For each AOT calculation, the lower bound of the integral in Eq. (5) corresponded to the lower range of the HSRL-2 cumulative AOT product (∼ 40 m) while the upper bound corresponded to the highest altitude at which the HSRL-2 cumulative AOT product was reported at that location. The P-3 did not typically fly above 8 km, and HSRL-2 retrievals of cumulative AOT were typically unavailable for altitudes above 6 km. Thus, these calculated NAAPS-RA AOTs and extinction coefficients are only representative of the model's 2nd–16th pressure layers.

Remotely sensed data were averaged first vertically and then horizontally to match the resolution of the NAAPS-RA model. HSRL-2 cumulative AOT values falling within the altitude bounds of each pressure layer were grouped. The cumulative AOT at the top of the pressure layer was subtracted from the value reported at the bottom of the pressure layer to establish a “slab” AOT for that pressure layer. This slab AOT was divided by the thickness of the pressure layer to achieve an extinction coefficient representative of that layer and for that specific vertical column. Calculated extinction coefficients for the same pressure layer were combined across all available column retrievals within the same 1∘ × 1∘ grid cell and averaged to arrive at a single vertical profile of extinction representing that 1∘ × 1∘ grid cell. Interpolated NAAPS-RA extinction coefficients within this same grid cell were horizontally averaged in an identical fashion (e.g., all interpolated coefficients for the first pressure layer were combined and averaged). The result was an average ground truth HSRL-2 extinction profile and NAAPS-RA extinction profile for the same 1∘ × 1∘ grid cell that could then be compared. All HSRL-2 total AOT values available within a grid cell were averaged to produce a ground truth AOT for that grid cell. Calculated NAAPS-RA AOT values were also averaged to arrive at a single simulated AOT value representative of the same portion of the grid cell.

The number of dropsondes released within a single 1∘ × 1∘ grid cell ranged from one to six. All dropsonde data available within a grid cell were combined, grouped, and averaged to produce a single RH value corresponding to each NAAPS-RA pressure layer. NAAPS-RA RH profiles interpolated to each dropsonde release were also combined and averaged to arrive at a single RH profile that could then be compared to the averaged in situ profile.

To explore NAAPS-RA performance as a function of altitude, we compare extinction coefficients within three altitude layers: (i) 40–500 m, (ii) 500–1500 m, and (iii) above 1500 m. The first altitude layer indicates how well NAAPS-RA simulates extinction within the ML (as discussed in Sect. 2.3), the second informs how well NAAPS-RA simulates the transition from the ML to the free troposphere (FT), and the third altitude layer focuses on model performance exclusively in the FT.

We begin by comparing HSRL-2 and NAAPS-RA extinction coefficients and AOT when NAAPS-RA values were calculated with modeled RHs to establish a basic understanding of model performance without any substitutions of in situ data. The average dropsonde RH profile for each grid cell was then used to recalculate all NAAPS-RA extinction coefficients and AOT values within that same 1∘ × 1∘ grid cell. The recalculated values are compared to the same HSRL-2 retrievals for that grid cell to understand how correcting errors in modeled RH affected NAAPS-RA simulations for AOT and extinction. The coefficient of determination (R2), bias, relative bias, root mean square error (RMSE), and normalized RMSE (NRMSE) are used to evaluate all NAAPS-RA simulations using the following formulations:

7R2=1N-1∑i=1NXi-X‾(Xi-X‾)2Yi-Y‾(Yi-Y‾)22 8bias=∑i=1NYi-XiN 9relativebias=biasX‾ 10RMSE=∑i=1NYi-Xi2N 11NRMSE=RMSEX‾, where X and Y are a set of in situ observations and NAAPS-RA simulations, respectively, for the same parameter (e.g., AOT, extinction, RH); N is the total number of points for a given comparison; and X‾ and Y‾ are the mean of sets X and Y, respectively.

Case I: background marine (RF19: 5 October 2019). The location and relatively low observed and simulated aerosol particle loadings indicate the P-3 sampled a relatively clean marine environment as compared to the rest of the campaign during this flight and within the selected grid cell (Fig. 1).

Mixed-layer AOT (AOTML) is the metric used to evaluate NAAPS-RA performance for the case study analysis. The HSRL-2 cumulative AOT value at the altitude closest to the average MLH for each case study flight was subtracted from the cumulative AOT value at the lowest altitude (40 m) to determine AOTML for each retrieval available within the case study 1∘ × 1∘ grid cell. The average of all retrieved AOTML values became the ground truth AOTML for a given case study. NAAPS-RA extinction coefficients calculated with modeled RHs (Sect. 2.5.2) were used in Eq. (5) to calculate AOTML at all locations coinciding with HSRL-2 retrievals. In these integrals, the lower bound was again 40 m, while the upper bound was the average MLH for a given case study flight. The calculated AOTML values were averaged to produce a single NAAPS-RA AOTML when only modeled parameters were used. This procedure was repeated with the NAAPS-RA extinction coefficients calculated using dropsonde RHs (Sect. 2.5.4) to arrive at a NAAPS-RA AOTML when only errors in model RH had been corrected.

Next, observed γ values and modeled RHs were used in Eq. (2) to calculate NAAPS-RA AOTML values when only γ was corrected. To account for the range of in situ γ values observed during a given case study, the mean γ as well as γ values 1 standard deviation above and below the mean were used in Eq. (2), resulting in a range of NAAPS-RA AOTML outputs. Normally, NAAPS-RA uses a species-dependent γi value in Eq. (2) to calculate f(RH) for each of the four aerosol types. Here, we use the same in situ γ in Eq. (2) for all four aerosol types. A mean mass-weighted NAAPS-RA γ was calculated for each case study using average mass fractions of ABF species, dust, smoke, and sea salt particles in the ML multiplied by their respective γi value. Comparing the NAAPS-RA mean mass-weighted γ to statistics for the in situ γ observations provides insight into how accurately the model simulated particle hygroscopicity for each case study.

Observed γ and dropsonde RH values were then both used in Eq. (2) to produce NAAPS-RA AOTML values when the entire f(RH) term had been corrected. After correcting this term, remaining discrepancies between modeled and retrieved AOTML values are presumably due to errors in simulated particle mass concentrations and/or the mass scattering and absorption efficiencies assigned to each particle type.

It is challenging to evaluate simulated mass concentrations of ABF species, dust, smoke, and sea salt particles and their respective optical properties because these particle type categories do not align with what was measured on the aircraft. For example, the AMS can quantify mass concentrations of organics, but it is difficult to determine the fraction of these organics associated with smoke versus the fraction associated with anthropogenic and/or biogenic emissions (which NAAPS-RA would place in the ABF category). To bypass this issue, we only compared simulated fine and coarse particle mass concentrations to in situ observations. NAAPS-RA fine mass was calculated as the sum of ABF and smoke mass, while coarse mass was calculated as the sum of dust and sea salt mass. The method to derive in situ fine and coarse particle mass concentrations is described below.

Eqs. (1) and (3) show that dry particle mass concentrations are an important component in simulating particle light extinction. Fine and coarse in situ mass concentrations were calculated to compare to those simulated by NAAPS-RA in the ML for each case study. In situ fine mass was characterized as the sum of AMS mass concentrations for OA, SO42-, NO3-, NH4+, and Cl-. Previous studies have examined the ability of the AMS to capture total fine particle mass by comparing to fine mass concentrations derived with other instruments, such as particle-into-liquid samplers (PILSs; e.g., Takegawa et al., 2005), optical particle counters (OPCs; e.g., Middlebrook et al., 2012, and references therein), and tapered element oscillating microbalances (TEOMs; e.g., Salcedo et al., 2006). AMS collection efficiency (CE) is adjusted to reach mass closure with the aforementioned and related instruments, with a CE of 0.5 being most common (Middlebrook et al., 2012, and references therein). AMS CE was set to 1 for the campaign based on comparison with coincident PILS measurements. This, in conjunction with the instrument's insensitivity to submicron dust and sea salt, indicates AMS mass concentrations represent a lower limit of true dry fine mass.

Coarse particle mass concentrations were calculated using FCDP size distributions and assuming all coarse particles were sea salt. In support of this, Hilario et al. (2020a) found crustal-marine particles to contribute 57 % of the coarse particle mass (1.15–10 µm) in the South China Sea in late September. As the FCDP sampled particles under ambient conditions, the dry particle diameter (Dp) range was calculated for each bin using the procedure described in Lewis and Schwartz (2004; see pages 54–55). Specifically, equations modeling the deliquescence growth curve for sea salt were used to determine relationships between the radii of sea salt particles at ambient RH (r), at 80 % RH (r80), and in dry conditions (rdry):

12rr80=0.541-RH13forRH>93%13rr80=0.671-RH14forRH<93%14rdry=r802.

For each FCDP size distribution, radii marking the edges of each size bin were set equal to r, while airborne meteorological data provided temporally coincident ambient RH values. The dry size distributions were then integrated using the density of sea salt (2.20 g cm-3; Seinfeld and Pandis, 2016) to arrive at total dry coarse particle mass concentration.

There are several uncertainties associated with quantifying coarse mass this way. First, this correction is very sensitive at RH >∼ 90 %, where sea salt exhibits large growth factors (e.g., Lewis and Schwartz, 2004). RHs above this threshold were common in the ML throughout the campaign (as will be shown in Sect. 3.2). The resulting large differences between ambient and dry particle radii corresponded to even larger corrections for dry particle volume and, therefore, dry particle mass. Additionally, there are known challenges in using OPCs (such as the FCDP) to accurately quantify coarse particle mass concentrations. The FCDP assumes the refractive index of water to derive sizes for all particles it samples, which introduces error when particles are not predominantly liquid. However, it is inconclusive as to whether coarse mass concentrations derived from OPCs tend to be negatively or positively biased. For example, Reid et al. (2003, 2006) found coarse-mode OPCs to overestimate the size of coarse particles (e.g., sea salt and dust), while other works have found OPCs to underestimate coarse mass concentrations (Kulkarni and Baron, 2011; Burkart et al., 2010). Our derived fine and coarse masses are still useful in roughly evaluating the corresponding NAAPS-RA simulations, but this analysis is highly preliminary and requires further investigation.

RF19 sampled the cleanest conditions for the entire campaign and provided an opportunity to evaluate NAAPS-RA when AOT was very low (HSRL-2 AOTML and AOT ranged from 0.01–0.04 and 0.03–0.09, respectively, for the 1∘ × 1∘ grid cell considered for this flight). This period was associated with a mild tropical disturbance and advection of clean marine air from the northern subtropical western Pacific east of the Philippines (Hilario et al., 2021).

NAAPS-RA underestimates AOTML and AOT (MLH = 579 m; AOTML bias =-0.01; AOT bias =-0.05; Table 3) and underestimates extinction throughout the column (Fig. 7). Both the AOTML bias and shape of the NAAPS-RA extinction profile are largely unchanged when dropsonde RHs are used, which is unsurprising because vertically resolved model and dropsonde RHs are similar in the ML. A mean in situ γ value of 0.20 ± 0.16 indicates particles were less hygroscopic than has been observed for other clean marine environments (0.38≤γ≤0.73; Titos et al., 2016, and references therein). NAAPS-RA overestimates particle hygroscopicity, but correcting model γ values induces negligible changes in AOTML biases because extinction coefficients are already very low in magnitude for this case study. Slight increases in simulated particle mass concentrations and/or mass scattering and absorption efficiencies will likely allow NAAPS-RA to reach full agreement with the HSRL-2 extinction profile. NAAPS-RA appears to accurately simulate fine mass and overestimate coarse mass in our preliminary comparison of simulated and observed fine and coarse particle mass concentrations. However, we acknowledge there is great uncertainty in our method to derive coarse mass concentrations (as described in Sect. 2.6.3), especially considering ambient RHs do not fall below ∼ 80 % in the ML for this case study. We leave a more thorough closure analysis between simulated and observed mass concentrations to a future study.

In contrast to the background marine case study, RF9 sampled the most polluted conditions for the campaign as smoke from the MC advected into the Sulu Sea (Hilario et al., 2021; HSRL-2 AOTML and AOT ranged from 0.45–0.68 and 1.26–1.58, respectively, for the 1∘ × 1∘ grid cell considered for this flight). Like the background marine case study, NAAPS-RA underestimates AOTML and AOT (MLH = 638 m; AOTML bias =-0.11; AOT bias =-0.29), but the model does capture the general shape of the extinction profile correctly in the ML (Fig. 8) so that the largest extinction coefficients are just below the MLH. Biases become more negative when dropsonde RHs are used (AOTML bias =-0.13; AOT bias =-0.31) because NAAPS-RA underestimates RH at altitudes up to ∼ 250 m and overestimates RH from ∼ 250 to the MLH. The decrease in modeled RH to measured RH at altitudes where extinction coefficients are highest causes the recalculated NAAPS-RA extinction profile to fall even further behind the HSRL-2 profile at these same altitudes, and agreement worsens.

The observation of negative γ values in smoke plumes advecting towards the Philippines from the southwest is arguably one of the more interesting preliminary results from the CAMP2Ex field campaign. However, errors in these observations may still exist given the nature of the particle chemistry. Nevertheless, f(RH) was low, and negative in situ γ values (-0.06 ± 0.02) observed on this flight may imply that a majority of the smoke particles were nonspherical and collapsed into spherical morphology upon humidification (Shingler et al., 2016). In contrast, NAAPS-RA assigns a slightly positive γ value to smoke particles based on a global average. Thus, when in situ γ values are used, biases in AOTML become even larger (from -0.25 to -0.27), and the HSRL-2 extinction profile cannot even be seen in the frame of Fig. 8b. This implies NAAPS-RA is underestimating either fine mass, coarse mass, or scattering and absorption efficiencies (or some combination of these parameters). Our preliminary assessment of simulated versus observed fine and coarse particle mass concentrations suggests that NAAPS-RA is overestimating both fine and coarse mass, but we report this result with caution. The large discrepancies in extinction between NAAPS-RA and HSRL-2 retrievals are likely due in some part to errors in simulated particle mass concentrations, and we encourage future work to investigate this more deeply.

RFs 9 and 10 were coordinated such that biomass burning emissions from the MC were sampled on subsequent days, which provided an opportunity to learn how smoke particle composition and hygroscopicity (among other air mass properties) changed with ∼ 24 h of additional aging. Aged smoke was the dominant air mass in the ML for the 1∘ × 1∘ grid cell selected for this case study, but conditions were considerably less polluted compared to RF9 (MLH = 593 m; HSRL-2 AOTML and AOT ranged from 0.08–0.11 and 0.20–0.23, respectively, for the grid cell considered in this case study).

AMS data indicate the smoke plume sampled during RF9 and RF10 had very similar chemical composition (Fig. S9), yet in situ γ values for RF10 suggest the air mass entering the Philippine Sea was more hygroscopic (γ= 0.04 ± 0.10) than the air mass sampled in the Sulu Sea during RF9 (γ=-0.06±0.02). However, the hygroscopic properties of this smoke mass are not straightforward as both positive and negative γ values were observed. More work is needed to fully explain this phenomenon. Nonetheless, NAAPS-RA treats all smoke particles the same, no matter their age, motivating interest in how such an assumption could lead to errors in simulated extinction.

For this case study, NAAPS-RA greatly overestimates AOTML and AOT (biases of 0.16 and 0.54 respectively; Fig. 9). NAAPS-RA slightly underestimates RH throughout the ML (and up to ∼ 2100 m), so agreement only worsens when dropsonde RHs are substituted into the model (AOTML and AOT bias = 0.17 and 0.57, respectively).

Similar to Case II, the model overestimates the hygroscopicity of smoke particles in this air mass (mean in situ and simulated γ values are 0.04±0.10 and 0.26±0.00, respectively). After adopting in situ RHs and γ values, NAAPS-RA overestimates AOTML (biases range from 0.07–0.13), suggesting there is likely a positive bias for fine and/or coarse particle mass concentrations and/or mass scattering and absorption efficiencies. NAAPS-RA appears to overestimate both fine and coarse particle mass concentrations in the ML, but additional work is needed to study agreement between in situ and simulated particle mass concentrations.

This case study provides an opportunity to assess NAAPS-RA performance for an air mass dominated by urban pollution from East Asia with moderate AOT (HSRL-2 AOTML and AOT ranged from 0.03–0.14 and 0.13–0.41, respectively). The model overestimates extinction (AOTML bias = 0.06; Fig. 10) and underestimates RH for all pressure layers within the ML (MLH = 535 m). When dropsonde RHs are substituted into the model, AOTML bias increases to 0.09, and extinction increases drastically in the pressure layer where dropsonde RH exceeds 90 %. NAAPS-RA simulates ABF as the dominant species in this air mass (average mass fraction of 0.70±0.01), which the model considers as the most hygroscopic aerosol type. The prevalence of this species in combination with relatively high dropsonde RHs within the ML makes the large increase in AOTML expected.

ABF is arguably one of the most difficult aerosol types for NAAPS-RA to accurately model as it combines organic and inorganic species, which can have very different hygroscopic and optical properties. NAAPS-RA assigns a γ value to ABF by assuming 40 % SO42- and 60 % OA. However, the composition of anthropogenic and biogenic emissions is likely to vary across different regions. For example, mean fine mass fractions of SO42- and OA (0.62±0.04 and 0.22±0.03, respectively) for this case study are largely different from what the model assumes, motivating interest in how AOTML will adjust when observed γ values are substituted into the model. The mean in situ γ value (0.23±0.04) is nearly half the mean NAAPS-RA mass-weighted γ (0.42±0.00) and the γ value assigned to ABF (0.46). When in situ γ values are used in the model, extinction agreement improves dramatically (AOTML biases drop to a range of 0.02–0.03), which suggests that the γ value assigned to ABF requires modification for this region.

Even with corrected RHs and γ values, NAAPS-RA overestimates AOTML for this 1∘ × 1∘ grid cell, which implies the model is overestimating fine and/or coarse particle mass concentrations and/or scattering and absorption efficiencies. Our preliminary comparison of simulated and observed fine and coarse mass concentrations indicates simulated mass concentrations are too high, but as we have mentioned, more work must be done before we can comment on these mass concentrations with certainty.