Towards a New MAX-DOAS Measurement Site in the Po Valley: Aerosol Optical Depth and NO₂ Tropospheric VCDs

Abstract

Pollutants information can be retrieved from visible (VIS) and ultraviolet (UV) diffuse solar spectra exploiting Multi-AXis Differential Optical Absorption Spectroscopy (MAX-DOAS) instruments. In May 2021, the Italian research institute CNR-ISAC acquired and deployed a MAX-DOAS system SkySpec-2D. It is located in the “Giorgio Fea” observatory in San Pietro Capofiume (SPC), in the middle of the Po Valley, where it has constantly acquired zenith and off-axis diffuse solar spectra since the 1st October 2021. This work presents the retrieved tropospheric NO₂ and aerosol extinction profiles (and their columns) derived from the MAX-DOAS measurements using the newly developed DEAP retrieval code. The code has been validated both using synthetic differential Slant Column Densities (dSCDs) from the Fiducial Reference Measurements for Ground-Based DOAS Air-Quality Observations (FRM4DOAS) project and real measured data. For this purpose, DEAP results are compared with the ones obtained with three state-of-the-art retrieval codes. In addition, an inter-comparison with satellite products from Sentinel-5P TROPOMI, for the tropospheric NO₂ Vertical Column Densities (VCDs), and MODIS-MAIAC for the tropospheric Aerosol Optical Depth (AOD), is performed. We find a bias of −0.6 × 10¹⁵ molec/cm² with a standard deviation of 1.8 × 10¹⁵ molec/cm² with respect to Sentinel-5P TROPOMI for NO₂ tropospheric VCDs and of 0.04 ± 0.08 for AOD with respect to MODIS-MAIAC data. The retrieved data show that the SPC measurement site is representative of the background pollution conditions of the Po Valley. For this reason, it is a good candidate for satellite validation and scientific studies over the Po Valley.

1. Introduction

Atmospheric constituents such as nitrogen oxides (NO_x) can influence air quality and climate [1]. Nitrogen oxides are formed during the combustion process, partly from nitrogen compounds in fuel, but mainly from the direct combination of atmospheric oxygen and nitrogen in flames. Nitrogen oxides are also produced naturally by lightning and, to a small extent, by soil microbial processes [2]. Emissions by automobiles, trucks, various non-road vehicles (e.g., construction equipment, boats and other vehicles), as well as power plants, cement kilns, turbines and industrial boilers, are the primary anthropogenic sources of NO_x [3]. The tropospheric NO₂ plays an essential role in the formation of aerosol [4] and its photolysis is the most common pathway for ozone formation [5]. Its monitoring is therefore crucial, in particular in polluted regions.

Satellite instruments are widely used for monitoring the distribution of such pollutants on a global scale. They allow the investigation of phenomena in remote regions where ground-based monitoring cannot be performed. Ultraviolet-visible (UV-VIS) sensors (e.g., TROPOMI, [6]) are used to retrieve the Vertical Column Density (VCD), i.e., the vertically integrated concentration of NO₂. This variable, listed by the Global Climate Observing System (GCOS) as a precursor of ozone and aerosols, is crucial to understand anthropogenic and natural production processes of NO_x.

Despite their global coverage, satellite retrievals need to be validated due to the complex nature of the retrieval problem [7]. In particular, they require a priori information on the vertical profile of the retrieved trace gas, and the presence of clouds limits their measurement capability. Comparison with ground-based measurements allows quantification of the impact of assumptions and approximations on observed processes and scenarios that are part of the retrieved satellite products. The comparison of MAX-DOAS or Pandora and satellite results in several sites has been thoroughly investigated in previous works [8,9,10,11]. Multi-AXis DOAS (MAX-DOAS) instruments, exploiting the Differential Optical Absorption Spectroscopy (DOAS) technique [12,13], are widely used in polluted regions to retrieve vertical profiles of aerosol extinction, NO₂ and other trace gases (e.g., refs. [14,15,16,17]). MAX-DOAS instruments can participate in the Fiducial Reference Measurements for Ground-Based DOAS Air-Quality Observations (FRM4DOAS, https://frm4doas.aeronomie.be (accessed on 5 March 2025), ref. [18]) project funded by the European Space Agency (ESA) for the homogenization of MAX-DOAS measurements and processing practices [19,20].

In Italy, the first MAX-DOAS instrument fully compliant with FRM4DOAS requirements is operated by the Consiglio Nazionale delle Ricerche-Istituto di Scienze dell’Atmosfera e del Clima (CNR-ISAC), and has been located at the “Giorgio Fea” meteorological observatory at San Pietro Capofiume (herein SPC), Bologna (Lat: 44.65°N, Lon: 11.62°E, Altitude: 11 m a.s.l.) since the 1 October 2021. The “Giorgio Fea” observatory is managed by the Agenzia Regionale per la Prevenzione, l’ Ambiente e l’ Energia (ARPAE, https://www.arpae.it/it/arpae/arpae, accessed on 5 March 2025) of Emilia Romagna. CNR-ISAC operates at SPC in the frame of a long-term agreement with ARPAE. At this site, ARPAE runs radar measurements and radio soundings. The site is equipped for in situ monitoring of trace gases and particulate matter. As the Mt. Cimone-Po Valley facility, it is part of the Aerosol, Clouds and Trace gases Research InfraStructure (ACTRIS). The description of the SkySpec-2D installed at SPC and its performance compared with MAX-DOAS and Pandora instruments are reported in [21], where the authors mainly focused on presenting the first NO₂ total VCDs in the Po Valley retrieved from zenith sky spectra.

This paper focuses on analysing MAX-DOAS SPC measurements for retrieving NO₂ tropospheric concentrations and aerosol extinction profiles and their corresponding tropospheric columns. To perform the retrieval, in the frame of the Instrument Data Evaluation and Analysis Service (IDEAS+) Quality Assurance for Earth Observation (QA4EO) service DOAS-BO ESA project, we developed and validated a dedicated code, named DOAS optimal Estimation Atmospheric Profile retrieval (DEAP). The code validation has been performed by exploiting both synthetic differential Slant Column Densities (dSCDs) provided by the FRM4DOAS community, and comparing the results obtained with DEAP on actual data with those obtained applying other retrieval codes such as the Bremen Optimal estimation REtrieval for Aerosols and trace gaseS (BOREAS) [22], the Mexican MAX-DOAS Fit (MMF) [23] and the MAinz Profile Algorithm (MAPA) [24]. MMF and MAPA are the codes selected for the centralized FRM4DOAS processing. The DEAP code is then applied to the first two years of SPC MAX-DOAS measurements. The resulting NO₂ tropospheric VCDs and AODs are finally compared to collocated data from the Sentinel-5P TROPOspheric Monitoring Instrument (TROPOMI) [6] and the Moderate Resolution Imaging Spectroradiometer (MODIS) Multi-Angle Implementation of Atmospheric Correction (MAIAC) [25] satellite product.

The paper is structured as follows: Section 2 briefly describes the measurements used during this work, the code developed for the retrieval and the reference algorithms used for the code validation. The validation results as well as the comparisons between AODs and NO₂ tropospheric VCDs from ground-based and satellite observations are given in Section 3. Discussions are reported in Section 4 and Conclusions in Section 5.

2. Materials and Methods

2.1. The SkySpec-2D MAX-DOAS Measurements at San Pietro Capofiume

The MAX-DOAS system installed at the “Giorgio Fea” meteorological station in SPC is a SkySpec-2D system developed by Airyx GmbH (before EnviMes) (https://airyx.de/wp-content/uploads/2024/09/SkySpec-2D.pdf (accessed on 5 March 2025)). This paper exploits its visible (VIS) MAX-DOAS spectra, acquired in the spectral region from approximately 410 nm to 550 nm. The system automatically starts to acquire atmospheric spectra every morning when the sun is 4° below the horizon, which represents a Solar Zenith Angle (SZA) position of 94°, according to our convention. Initially, it acquires only zenith-sky spectra. SkySpec-2D starts to perform MAX-DOAS measurements when the SZA becomes lower than 85°. In MAX-DOAS acquisition mode, the SkySpec-2D measures in three different azimuth directions (in our convention 0° corresponds to the North and then positive angles are measured moving clockwise up to 360°): 120°, 225° and 300° from the 1 October 2021 to the 23 March 2022, and 135°, 250° and 315° afterwards (Figure 1). The change in azimuth viewing directions was motivated by the fact that, having moved the telescope from the ground-level tripod to the top of the shelter, where the spectrometer is located, the availability of free lines of sight changed.

Following the minimum requirements of the FRM4DOAS network, the spectra are acquired at the following elevation angles: 1°, 2°, 3°, 5°, 10°, 30° and 90°. On average, the acquisition of one scan takes about 3 min and the number of scans measured in each azimuth direction during a day varies between about 50 in winter and 80 in summer. The VIS spectra are then analysed using the QDOAS version 3.4.6 (https://uv-vis.aeronomie.be/software/QDOAS/ (accessed on 5 March 2025)) software to retrieve the O₄ and NO₂ dSCDs and their errors due to the fit. For the QDOAS analysis of the MAX-DOAS spectra, we use the corresponding zenith sky measurement as the reference spectrum for each scan. The QDOAS set-up used in the analysis is reported in Table A2 of [21]. The O₄ and NO₂ dSCDs of each scan, relative to the mentioned elevation angles, are the input of the DEAP code. The SkySpec-2D spectra measured from SPC are delivered to the FRM4DOAS central processing.

2.2. Retrieval of Tropospheric Profiles

The DOAS international community has developed several codes for retrieving tropospheric vertical profiles from MAX-DOAS measurements. In the frame of the FRM4DOAS project, an effort was made to evaluate the performances of different retrieval codes and to select the ones more suitable for centralized processing [20]. The BOREAS, MMF and MAPA codes participated in this round-robin exercise, which included 8 codes (BOREAS, MMF, MAPA, BePRO, HEIPRO, PRIAM, MARK and NASA). The codes chosen were MMF and MAPA. The work presented in this paper used BOREAS, MMF and MAPA to validate the DEAP code. All of these codes are described in the following subsections.

2.2.1. The DEAP Retrieval Code and Set-Up

The DEAP code is an Optimal Estimation (OE) algorithm [27] developed at CNR-ISAC that, exploiting the SCIATRAN code [28] as a forward model, can retrieve tropospheric vertical profiles from MAX-DOAS measurements. The DEAP code does not need any spectral intensity input. It uses only the dSCDs of O₄ and NO₂ (or other gases) retrieved from spectra measured at the given elevation angles together with information on vertical elevation angles and relative azimuth angles. The retrieval is performed in two steps: one for aerosol extinction and one for gaseous profiles retrieval. The aerosol extinction retrieval is an iterative procedure that requires the calculation of the derivatives of O₄ dSCDs with respect to the aerosol extinction profile. This calculation is performed through the use of the SCIATRAN version 4.1 Radiative Transfer Model (RTM).

The gaseous part requires only the simulation of Box-Air Mass Factors (box-AMFs, defined as the ratio of the partial slant column to the partial vertical column [13]) without the need for iterations since the retrieval of the gas profile is treated as a linear problem. For this reason, the gaseous part is faster than the aerosol one. First, DEAP retrieves the aerosol extinction profile from O₄ dSCDs by exploiting the following iterative procedure (see Equation 5.35 of [27]):

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill {\mathit{x}}_{i+1}& ={\mathit{x}}_i+\hfill \\ \hfill & {({\mathit{K}}_i^T{\mathit{S}}_y^{-1}{\mathit{K}}_i+{\mathit{S}}_0^{-1}+\lambda {\mathit{D}}_i)}^{-1}\hfill \\ \hfill & ({\mathit{K}}_i^T{\mathit{S}}_y^{-1}(\mathit{y}-{\mathit{y}}_i)-{\mathit{S}}_0^{-1}({\mathit{x}}_i-{\mathit{x}}_0))\hfill \end{array}\end{array}$$

(1)

where ${\mathbf{x}}_i$ is the retrieved extinction profile at iteration i, ${\mathbf{x}}_{i+1}$ is the retrieved profile at iteration $i+1$, ${\mathbf{x}}_0$ is the a priori profile, ${\mathbf{K}}_i$ is the Jacobian matrix, ${\mathbf{D}}_i$ is a diagonal matrix, which contains the diagonal values of ${\mathbf{K}}_i^T$${\mathbf{S}}_y^{-1}$K_i, S₀ is the a priori variance-covariance matrix (VCM), $\lambda$ is the lambda of Marquardt damping factor, S_y is the dSCDs VCM matrix, y_i is the vector containing the simulated O₄ dSCDs and y is the vector of measured O₄ dSCDs at different elevation angles. The derivatives of O₄ dSCDs with respect to the aerosol profiles that compose the Jacobian matrix are calculated numerically with SCIATRAN.

The associated total error matrix is the sum of the retrieval and smoothing error matrices. The retrieval error matrix is defined as:

$$\begin{array}{cc}\hfill {\mathit{xerr}}_{retr}& ={\mathit{G}}_i{\mathit{S}}_y{\mathit{G}}_i^T\hfill \end{array}$$

(2)

where ${\mathit{G}}_i$ is the gain matrix defined as:

$$\begin{array}{cc}\hfill {\mathit{G}}_i& ={({\mathit{K}}_i^T{\mathit{S}}_y^{-1}{\mathit{K}}_i+{\mathit{S}}_0^{-1}+\lambda {\mathit{D}}_i)}^{-1}{\mathit{K}}_i^T{\mathit{S}}_y^{-1}\hfill \end{array}$$

(3)

while the smoothing error is:

$$\begin{array}{cc}\hfill {\mathit{xerr}}_{smooth}& =({\mathit{A}}_{\mathit{k}}-\mathit{I}){\mathit{S}}_0{({\mathit{A}}_{\mathit{k}}-\mathit{I})}^T\hfill \end{array}$$

(4)

where ${\mathbf{A}}_k$ is the Averaging Kernel (AK) matrix defined as:

$$\begin{array}{cc}\hfill {\mathit{A}}_{\mathit{k}}& ={({\mathit{K}}^T{\mathit{S}}_y^{-1}{\mathit{K}}_i+{\mathit{S}}_0^{-1}+\lambda {\mathit{D}}_i)}^{-1}{\mathit{K}}_i^T{\mathit{S}}_y^{-1}{\mathit{K}}_i\hfill \end{array}$$

(5)

and I is the identity matrix.

The iterative method aims at minimizing the chi-square value, defined as:

$$\begin{array}{c}\hfill {\chi }^2={(\mathit{y}-{\mathit{y}}_i)}^T{\mathit{S}}_y^{-1}(\mathit{y}-{\mathit{y}}_i)+{(\mathit{x}-{\mathit{x}}_0)}^T{\mathit{S}}_0^{-1}(\mathit{x}-{\mathit{x}}_0)\end{array}$$

(6)

After some trade-off considerations between speed and accuracy, we set a maximum number of 20 iterations. However, the iterative procedure can stop earlier if the ${\chi }^2$ stops decreasing and the differences between the measured and simulated dSCDs, for each elevation angle, are lower than three times the dSCDs errors.

In the second step, the retrieved aerosol profile is used to calculate the trace gas box-AMFs, relative to each scanned elevation angle and to each vertical layer of the retrieval grid. Finally, the gaseous profiles are obtained using the following inversion scheme:

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill {\mathit{x}}_g& ={\mathit{x}}_0+\hfill \\ \hfill & {({\mathit{K}}^T{\mathit{S}}_y^{-1}\mathit{K}+{\mathit{S}}_0^{-1})}^{-1}\hfill \\ \hfill & ({\mathit{K}}^T{\mathit{S}}_y^{-1}(\mathit{y}-{\mathit{y}}_g)-{\mathit{S}}_0^{-1}({\mathit{x}}_i-{\mathit{x}}_0))\hfill \end{array}\end{array}$$

(7)

with the retrieval error computed as for the aerosol retrieval, with the only difference being that $\lambda$ is null.

In Equation (7), x_g is the retrieved profile, x₀ is the a priori profile, K is the box-AMF matrix, S₀ is the an priori VCM matrix, S_y is the dSCDs VCM matrix, y_g are the simulated dSCDs and y is the vector containing the measured dSCDs of the gas at different elevation angles.

The a priori VCM matrix is composed using a percentage of the a priori profile in the diagonal elements and a correlation length for extra-diagonal terms (in this case 0.4 km for the aerosol and gas). We use 100% of the a priori extinction profile as a priori error for the retrieval of aerosol extinction and 100% of the a priori NO₂ profile for the NO₂ retrieval. The aerosol extinction a priori profile has an AOD value of 0.1 and decreases exponentially with altitude with a scale height of 1 km. The simulations for the retrieval are performed using a constant single scattering albedo of $0.92$ and the Henyey–Greenstein phase function with an asymmetry factor of $0.68$ [29].

For the NO₂ a priori, we use an exponentially decreasing profile with a scale height of 1 km. The VCD of the NO₂ a priori profile is 8 × 10¹⁵ molec/cm². This value is within the ones that can be observed in the Po Valley, as can also be seen from TROPOMI data. A variation of ±25% in the NO₂ a priori profile leads to a variation of only about ±1% in the retrieved VCDs.

The altitude retrieval grid, customizable by the user, is composed of equally spaced steps of 200 m from 0 to 3 km, for both the extinction and NO₂ retrievals. The Marquardt $\lambda$ is initially set to 0.1 and is updated at every iteration. If the iteration leads to a lower ${\chi }^2$, $\lambda$ is divided by 1.2, and is increased by a factor of 2 otherwise.

These settings are used to retrieve aerosol extinction and NO₂ profiles from SkySpec-2D MAX-DOAS VIS spectra from the 1 October 2021 to 30 September 2023. All the scans measured by the SkySpec-2D are processed with the DEAP code. The retrieved results are then filtered according to a two-step process. The first step relates to comparing the measured dSCDs and those simulated with the retrieved aerosol extinction and NO₂ profiles. In particular, a profile is discarded if the difference between the simulated and measured dSCD is more than three times the dSCD error coming from the fit and if the percentage difference is more than 30% of the measured dSCD (for at least one of the elevation angles). Moreover, the NO₂ profile is automatically discarded if the aerosol profile of the same scan is discarded.

Since the cloud impact on the spectra is not negligible [30], the second step has the purpose of filtering out cloudy measurements. Cloud filtering is based on a color index approach [31]. The color index approach is widely used for flagging cloud-contaminated MAX-DOAS measurements (e.g., refs. [18,30,32,33,34,35]). Other methods involve the effects of clouds on, e.g., radiance, O₄ absorption and Ring effect [30]. The use of color index (CI) allows filtering of the cloudy data without the use of external information, using measured spectra only. In fact, the particles scattering effect on CI makes this indicator ideal for investigating cloud contamination. The color index is estimated as the ratio between the average radiance computed in the wavelength interval 410–415 nm and the one in the interval 545–550 nm. The retrieved profiles are flagged as cloudy if the zenith measurement of the relative MAX-DOAS scan, used as a reference for the QDOAS analysis, is cloudy or if clouds are present in the off-axis line of sights of the considered direction. The zenith reference spectrum is considered cloudy if the computed color index value is below the threshold of 1.2. We chose this threshold because the estimated distribution of the zenith color, shown in Figure 2, is bimodal with a transition region between the two distributions around the color index value of 1.2. We empirically verified that the left distribution, relative to the lower color indexes, represents the cloudy spectra. The threshold value of 1.2 has been chosen in order to filter out the heavily contaminated spectra without removing too much data from the dataset.

Setting a threshold color index for the off-axis spectra is a challenging task. Color indexes calculated from off-axis spectra present complex dependencies on solar zenith and azimuth angles and the aerosol load [32]. For this reason, we decided to exploit a qualitative method inferred from the simulations: clouds are present in the off-axis line of sights if the values of the color index derived from the spectra measured at the elevation angles of 3°, 5°, 10° and 30° are not sorted in ascending order.

Figure 3 shows an example of retrieved aerosol extinction and NO₂ profiles for the 23 March 2023. No clouds are present. The AKs for these profiles are reported in Figure 4 as an example. As can be noticed, 2.2–2.9 degrees of freedom (DOFs, calculated as the trace of the AK matrix) are obtained for both profiles over 16 retrieved altitude values.

The retrieved aerosol extinction and NO₂ profiles are then integrated along the vertical domain to obtain the AODs and the tropospheric NO₂ VCDs.

2.2.2. MMF Description and Set-Up

The Mexican MAX-DOAS Fit (MMF, [23]) is an OE-based profiling algorithm. As a forward model, it uses VLIDORT version 2.7 [36]; however, only the intensity part of the Stokes vector is used. The layer input parameters for each atmospheric layer are calculated from temperature and pressure profiles, the trace gas concentration in each layer, and the aerosol properties. The latter are assumed to be equal in each layer, except for the partial AOD. The aerosol phase function uses the Henyey–Greenstein [29] phase function parametrization. Additional input parameters are the wavelength of the retrieval and the surface albedo.

As in DEAP, the retrieval algorithm comprises an aerosol extinction profile retrieval and a trace gas profile retrieval. The former constrains the aerosol extinction profile in the forward model of the trace gas retrieval. Since both retrievals work in logarithmic retrieval space, the problem is no longer linear, and therefore the inversion uses constrained damped least-square fitting with an optimal estimation regularization for both retrievals. Both the a priori and the covariance matrix are constructed; the latter uses 50% of the former on the diagonal and applies a correlation length of 200 m to calculate the off-diagonal entries, as described in [37]. The a priori AODs are set to 0.18, and the NO₂ retrieval uses an a priori VCD of 9 × ${10}^{15}$ molec/cm². Both a priori profiles are exponentially decreasing with a scale height of 1 km. These choices are the default values within FRM4DOAS; no adaption for SPC has been performed.

The outputs of MMF consist of the aerosol extinction profiles, trace gas partial column profiles, their integrated quantities, the corresponding noise and smoothing errors, the averaging kernel, the degrees of freedom, and a quality flag for the retrieval. The quality flagging is based on the convergence of the algorithm, the root mean square of the difference between measured and simulated dSCDs, the degrees of freedom, and the stability of the retrieval.

2.2.3. MAPA Description and Set-Up

The MAPA v0.98 [24] is a profiling algorithm developed by the Max Planck Institute for Chemistry (MPIC). MAPA does not use OE, but is based on a parameterization approach: vertical profiles are parameterized by (1) the column (i.e., AOD for aerosols and VCD for trace gases), (2) the layer height, and (3) a shape parameter. Best matching parameters are determined by a least-squares fit of the observed vs. modelled dSCDs for an elevation angle sequence. The MAPA setup within the FRM4DOAS processing chain is the same as in [24].

2.2.4. BOREAS Description and Set-Up

The MAX-DOAS profile inversion algorithm BOREAS was developed at the Institute for Environmental Physics in Bremen, Germany [22]. The algorithm uses trace gas and O₄ dSCD scans to retrieve vertical concentration and extinction profiles, respectively. The aerosol retrieval utilizes the Newton–Gauss algorithm including a Tikhonov smoothing term and minimizes the difference between measured and simulated differential slant optical thicknesses of the oxygen collision complex O₄. The trace gas retrieval uses the iterative Levenberg–Marquardt algorithm. The retrieval grid is defined from 0 to 4 km in 100 m steps. The aerosol a priori profiles decrease exponentially with the altitude, with a scale height of 1.25 km and a ground extinction value of 0.18. The NO₂ a priori profiles decrease exponentially with the altitude, with a scale height of 0.75 km and a ground concentration of 9 × 10¹⁰ molec/cm³. The a priori profile is scaled by the vertical column derived from the geometric approximation for the 30° elevation measurement. A constant single scattering albedo of 0.95 and a Henyey–Greenstein asymmetry factor of 0.7 are used.

2.3. Correlative Data

Satellite products validation is one of the principal applications of MAX-DOAS measurements [38]. In this work, we used satellite datasets for inter-comparisons with our retrievals. MODIS-MAIAC data were used to compare the AODs at 477 nm, while Sentinel-5P TROPOMI products were used to compare the NO₂ tropospheric VCDs. Both satellite datasets are described in the following subsections.

2.3.1. MODIS-MAIAC

MODIS has been flying onboard the NASA Terra (descending node, about 10:30 equatorial crossing time) and Aqua (ascending node, about 13:30 equatorial crossing time) satellites since 2000 and 2002, respectively. MODIS observes the Earth, acquiring data in 36 spectral bands (from 0.4 $\mathsf{\mu }$m to 14.4 $\mathsf{\mu }$m), with a spatial resolution of about 1 km at nadir. AODs over land are retrieved through the Multi-Angle Implementation of the Atmospheric Correction (MAIAC) algorithm [39]. The MODIS-MAIAC AOD retrievals allow determination of the characteristics of fine aerosols and distinction between aerosol sources [40]. In this work, we use the MCD19A2 Version 6 Level 2 data product. This product contains, among other variables, the AODs at 470 nm together with AODs uncertainty and a quality flag (QA). Since MODIS-MAIAC AOD products are provided on a 1 × 1 km² regular grid, for this analysis, we considered best quality (e.g., CloudMask = Clear and AdjacencyMask = Clear) AOD products corresponding to the grid cell containing the SPC site and maximum allowed time differences between MODIS and SkySpec-2D observations of ±15 min.

2.3.2. Sentinel-5P TROPOMI

TROPOMI has been flying onboard the Sentinel-5P satellite (ascending node 13:30) since October 2017. TROPOMI measures from the UV to the short-wave infrared (SWIR) spectral region at a spatial resolution of 3.5 × 7 (5.5) km², exploiting four separate spectrometers. The Sentinel-5P TROPOMI products are described in [6,41]. As in [21], we use the OFFL NO₂ products [42,43] having a combined quality assurance value (qa_value) higher than 0.75 [43]. A reprocessed version of the NO₂ dataset named RPRO that has been produced using v2.4 is available up to 25 July 2022. From 17 July 2022 we used version 2.4, up to 12 March 2023, when v2.5 started.

As coincidence criteria, we considered (averaging them) Sentinel-5P TROPOMI data in a radius of 5 km around the SPC site and MAX-DOAS data within ±15 min from the Sentinel-5P TROPOMI overpass.

Sentinel-5P TROPOMI tropospheric NO₂ VCDs are retrieved using the retrieval algorithm developed at KNMI [44]. The algorithm has three steps: (1) retrieval of the total NO₂ slant column density, (2) separation of stratospheric and tropospheric contribution and (3) conversion of the tropospheric and stratospheric NO₂ SCDs to VCDs. This conversion exploits vertically resolved AMFs calculated using NO₂ vertical profiles from the TM5-MP model [45]. The TM5-MP model reports the NO₂ profiles on a 1 × 1 latitude-longitude degree resolution. This resolution can smear out some fine structures. Several studies (e.g., ref. [46]) apply a correction to the Sentinel-5P TROPOMI columns, using more realistic NO₂ profiles instead of the TM5-MP model ones. For example, they show that the use of daily median MAX-DOAS profiles, as Sentinel-5P TROPOMI a priori, improves the agreement between satellite and MAX-DOAS data.

In our work, to correct the tropospheric Sentinel-5P TROPOMI VCDs according to our MAX-DOAS profiles, we used the formula described in the Sentinel-5P user manual section 8.8 [43]:

$$\begin{array}{cc}\hfill VC{D}_{SAT-TROP}^{MAXDOAS-pro}& =VC{D}_{SAT-TROP}{\displaystyle \frac{{\sum }_i{c}_i}{{\sum }_{i}A{K}_i^{TROP}{c}_i}}\hfill \end{array}$$

(8)

where $VC{D}_{SAT-TROP}^{MAXDOAS-pro}$ is the TROPOMI corrected tropospheric VCD, $VC{D}_{SAT-TROP}$ is the original TROPOMI tropospheric VCD, $c_i$ contains the NO₂ concentrations from the MAX-DOAS profile and $A{K}_i^{TROP}$ is the tropospheric AK with

$$\begin{array}{c}\hfill A{K}_i^{TROP}=0\end{array}$$

(9)

above tropopause level, and

$$\begin{array}{c}\hfill A{K}_i^{TROP}={\displaystyle \frac{AM{F}^{TOTAL}}{AM{F}^{TROP}}}A{K}_i^{TOTAL}\end{array}$$

(10)

below tropopause level. In Equation (10), $A{K}_i^{TOTAL}$ is the total AK and $AM{F}^{TOTAL}$ and $AM{F}^{TROP}$ are the total and tropospheric AMFs, respectively.

$AM{F}^{TOTAL}$, $AM{F}^{TROP}$ and $A{K}_i^{TOTAL}$ are contained in the Level 2 Sentinel-5P TROPOMI files.

3. Results

3.1. DEAP Validation

3.1.1. DEAP vs. MAPA, MMF and BOREAS with Synthetic dSCDs

Before processing the SPC MAX-DOAS dataset, the DEAP code was validated with synthetic dSCDs. In the frame of the FRM4DOAS activities, a dataset of synthetic dSCDs was used for a round-robin exercise [20]. This dataset, the used reference atmosphere and set-up, are freely available, and represent a valuable tool for the development of retrieval algorithms that exploit MAX-DOAS measurements. The package consists of a dataset of O₄ (360 and 477 nm), NO₂ and HCHO dSCDs and their corresponding errors simulated at nine elevation angles for a total of 990 different combinations of sun position, trace gas profiles and aerosol profiles. In this work, we used the dataset version with 5% noise added to account for possible atmospheric variabilities (then named “v1n”).

As reported in Table 8 of [47], we used O₄ SCDs at 477 nm for aerosol retrievals and NO₂ dSCDs at 460 nm for NO₂ profiles. Eleven aerosol profiles were used to model the synthetic O₄ SCDs. HCHO and NO₂ dSCDs were modelled using the combination of 10 gas vertical profiles and 11 aerosol profiles (110 model atmospheres). For each atmospheric scenario we used the combinations of 3 SZA (40°, 60°, 80°) and 3 relative azimuth angles (RAA, 0°, 90°, 180°).

To be consistent with the results reported in Table 8 of [47], we did not use the dSCDs relative to the aerosol profiles 8, 9 and 10. All the retrieval settings that we used were the same as in reference [47]; this also means a different grid than our default grid (200 m spacing up to 4 km). To quantify the DEAP results, we calculated the correlation coefficient R, slope, intercept, RMS value of simulated (true) values and retrieved values. For the comparison, we used the MAPA (named MPIC-PAR-MC in [47]), MMF and BOREAS results taken from [47]. MAPA and MMF are the codes selected in the frame of FRM4DOAS for centralized processing. Here, we compare the aerosol extinction and NO₂ concentrations at all the retrieval altitudes (

Table 1. Slope, intercept, correlation, RMS and number of points for MAPA, MMF, BOREAS and DEAP codes applied to synthetic dSCDs. Parameters are calculated for NO₂ and aerosol extinction tropospheric profiles at 477 nm.

			Tropospheric	Profiles	– “v1n”
		Aerosol 477 nm	[km−1]			NO2 [molec/cm3]
	MAPA	MMF	BOREAS	DEAP	MAPA	MMF	BOREAS	DEAP
Slope	1.01	0.91	0.89	0.88	0.97	0.87	0.81	0.89
Intercept	−0.00	0.005	0.008	0.007	0.015 × ${10}^{11}$	0.05 × ${10}^{11}$	0.04 × ${10}^{11}$	0.05 × ${10}^{11}$
R	0.89	0.89	0.91	0.94	0.85	0.84	0.66	0.65
RMS	0.08	0.07	0.06	0.05	0.39 × ${10}^{11}$	0.33 × ${10}^{11}$	0.62 × ${10}^{11}$	0.70 × ${10}^{11}$
N	10,540	12,260	12,920	12,700	10,220	12,440	11,800	12,320

), their surface values (

Table 2. Slope, intercept, correlation, RMS and number of points for MAPA, MMF, BOREAS and DEAP codes applied to synthetic dSCDs. Parameters are calculated for NO₂ and aerosol extinction surface values at 477 nm.

			Surface	Values	– “v1n”
		Aerosol 477 nm	[km−1]			NO2 [molec/cm3]
	MAPA	MMF	BOREAS	DEAP	MAPA	MMF	BOREAS	DEAP
Slope	1.03	0.97	0.85	0.90	0.88	0.90	0.47	0.91
Intercept	0.009	0.01	0.02	−0.009	0.12 × ${10}^{11}$	0.21 × ${10}^{11}$	0.14 × ${10}^{11}$	−0.08 × ${10}^{11}$
R	0.87	0.91	0.97	0.95	0.82	0.90	0.52	0.83
RMS	0.17	0.13	0.09	0.11	0.9 × ${10}^{11}$	0.59 × ${10}^{11}$	1.58 × ${10}^{11}$	0.92 × ${10}^{11}$
N	527	613	646	635	511	622	590	616

) and their tropospheric integrated values (

Table 3. Slope, intercept, correlation, RMS and number of points for MAPA, MMF, BOREAS and DEAP codes applied to synthetic dSCDs. Parameters are calculated for NO₂ tropospheric columns and AODs at 477 nm.

			Tropospheric	Columns	– “v1n”
		Aerosol 477 nm	[km−1]			NO2 [molec/cm3]
	MAPA	MMF	BOREAS	DEAP	MAPA	MMF	BOREAS	DEAP
Slope	1.06	0.62	0.82	0.82	1.02	0.93	0.90	1.08
Intercept	−0.024	0.11	0.05	0.05	0.03 × 1016	0.12 × 1016	0.07 × 1016	−0.04 × 1016
R	0.90	0.92	0.88	0.98	0.96	0.98	0.97	0.99
RMS	0.14	0.13	0.13	0.07	0.65 × 1016	0.31 × 1016	0.18 × 1016	0.20 × 1016
N	527	613	646	635	511	622	590	616

). The DOFs retrieved with DEAP compare well with respect to the MMF results in Figures 6 and 8 of [20], ranging from 1.01 to 3.08 for aerosol extinction to 1.16 to 2.88 for NO₂ profiles.

Regarding the profile retrievals, DEAP shows quite good slope values, similar to MAPA, MMF and BOREAS. The correlation is always high, sometimes the highest among the codes (e.g., for aerosols). A good RMS is obtained, especially for the aerosol retrievals. Small differences with respect to the other codes are present in the number of valid data. DEAP, as BOREAS, presents worse correlations when comparing NO₂ vertical profiles. Considering surface results (Table 2), we obtained slope values in between the ones of the other retrieval codes and a high correlation (0.95 for aerosols and 0.83 for NO₂). The resulting tropospheric columns, in Table 3, from DEAP show a very high correlation (0.98, 0.99) and small RMS.

To better visualize the DEAP performances in retrieving tropospheric profiles, in the Appendix A, we report an overview of DEAP extinction and NO₂ profiles obtained from synthetic dSCDs. For this exercise we repeat what was performed in [20], and Figure A1 and Figure A2 reproduce, as much as possible, Figures 12 and 14 of [20]. In our representation, we interpolate all the profiles on a common vertical grid of 0.1 km spacing that ranges from 0.1 to 3.9 km. For these examples, we do not apply any filtering. As can be seen, the DEAP code correctly reproduces the reference profiles under the majority of cases with a correlation of 0.93 for aerosols and 0.82 for NO₂ profiles.

In conclusion, DEAP performs well and its results are entirely consistent with the state-of-the-art codes. Given that this is the first version of the code, the obtained results can be considered as satisfactory.

3.1.2. DEAP vs. MAPA, MMF and BOREAS with Real dSCDs

The validation of the DEAP code with synthetic dSCDs allowed the removal of possible code bugs. However, applying a retrieval code to actual measurements is always challenging because real measurement scenarios may contain features that are more difficult to reproduce than the simulated ones. As reported in Section 2, we routinely provide the SPC SkySpec-2D spectra to the FRM4DOAS project for centralized processing. In addition, the SkySpec-2D spectra have also been processed with the BOREAS code. For this reason, the SPC MAX-DOAS data were processed with all four retrieval codes in the period from the 1 October 2021 to the 9 March 2022. It must be taken into account that the dSCDs used as input for the four retrieval codes are not identical because the ones used for MAPA and MMF were produced by the FRM4DOAS central processing, which presents some small differences in the QDOAS settings compared to our home-made processing.

For the comparison, we considered the tropospheric AODs and NO₂ VCDs measured at the azimuth direction 300°. Regarding the centralized processing data, we used MAPA version v0991 with free O₄ scaling factor and MMF version 1.0. Moreover, for MAPA we considered only data having a quality flag equal to 0 (we removed the warning data), while for MMF we decided to also use data flagged as warning in order to have a significant population.

The results of the comparisons between the reference codes and DEAP, applied to the measured SCDs, are reported in

Table 4. Slope, intercept, correlation, RMS and number of points for the comparison of DEAP results with the ones obtained with MAPA, MMF and BOREAS on real SCDs measured in SPC at the azimuth direction 300°. Parameters are calculated for NO₂ tropospheric columns and AODs at 477 nm.

	Reference Codes vs. DEAP
	AOD			NO2 Tropospheric VCD [molec/cm2]
	MAPA	MMF	BOREAS	MAPA	MMF	BOREAS
Bias	0.02	0.03	0.01	−0.29 × 1015	0.14 × 1015	−0.008 × 1015
RMS	0.08	0.02	0.03	0.67 × 1015	0.75 × 1015	0.61 × 1015
R	0.70	0.98	0.97	0.99	0.99	0.99
N	1987	2796	2588	1899	2681	2500

. The correlations between DEAP and the other codes are always high (0.97–0.99) apart from the comparison of AOD with MAPA (0.70). This suggests that DEAP AODs are in good agreement with MMF and BOREAS and that MAPA AODs present very different behavior. Although the biases are minimal, DEAP underestimates the AODs compared to all the other codes by about 0.02. On the other hand, the DEAP NO₂ VCDs are in good agreement with BOREAS and range between the MAPA and MMF values. Figure 5 and Figure 6 show the time series of AODs and NO₂ tropospheric VCDs obtained from the SPC MAX-DOAS data with the four codes. The middle panel shows the differences between the other codes and DEAP, while the lower one reports the value of the UV color index (UV-CI) calculated as the ratio of the zenith sky spectra averaged in the 5 nm region around 330 nm divided by the one around the 390 nm. The color indexes calculated in the UV region are more sensitive to particles with respect to the one in the VIS, which we used for the cloud filtering, and thus to aerosol presence. The time series confirm the good performances of DEAP code in catching diurnal variation in AODs or NO₂ tropospheric VCDs distributions. In particular, during clear sky days (e.g., on the 1, 2 and 9 March 2022 as shown by the UV-CI values) the agreement is very strong, especially for AODs. Additionally, lower AODs values are also retrieved on those days.

An example of the profiles retrieved from DEAP with respect to the other codes is given in Figure 7 for aerosol extinction and in Figure 8 for NO₂ retrieved on the 9 March 2022 using measurement at 300° azimuth. The same filtering adopted in Figure 5 and Figure 6 is applied here. Although this is a qualitative comparison, it still highlights the performances of the DEAP code with respect to other algorithms: the main features and behaviour of aerosol extinction and NO₂ distribution are well captured both in the vertical and in the time range. The aerosol profiles have a maximum near the ground from 9 to 12 a.m., while for NO₂ the maximum values near the ground are reached slightly before or from 6 to 8 a.m. It is worth noticing that, in the case of NO₂, similar features in the vertical structure of the profiles is found in all the codes at around 8 a.m., with higher values extending up to 0.4–0.6 km. The concentration retrieved from all the four codes is similar, with an exception being major differences found in the MAPA results for aerosol extinction at 0.4–0.6 km from 8 to 12 a.m.

3.2. San Pietro Capofiume Dataset of NO₂ Tropospheric VCDs and AODs

The DEAP code was applied to the first two years of SPC SkySpec-2D MAX-DOAS dSCDs, obtained with the QDOAS (settings in Table A2 of [21]) from the VIS spectra. A global overview of the dataset, as a function of months and time of the day, is given in Figure 9 for AODs at 477 nm, and in Figure 10 for NO₂ tropospheric VCDs. The average values, errors, standard deviations and number of observations are reported in the four panels of each figure. We show the filtered results, even though the effect of residual clouds can still be present, particularly in the case of fog or very low clouds.

AOD values are generally higher during the summer with higher values usually present in the middle of the day. The average standard deviations are slightly larger in winter, probably due to residual cloud effects under fog conditions. The number of points is generally higher during spring/summer and lower during winter. This behavior is mainly due to the number of sunny days which is higher during summer. However, it is also evident that a lower number of points is present in May. This is due to an instrument failure in May 2022.

NO₂ tropospheric VCDs show seasonal behavior. Higher values are found all day during winter. Due to photochemistry, values in the early morning and late afternoon are higher than the ones around noon. In winter, the NO₂ distribution is more uniform due to more stable atmospheric conditions in the boundary layer, with the possible presence of a thermal inversion that inhibits the dispersion and enhances the fog probability. As for AODs, the standard deviations are higher in winter, probably due to unfiltered clouds/fog effects that produce some high values.

The weekly cycles of AODs and NO₂ tropospheric VCDs are reported in Figure 11. The median values are the orange lines, the box limits represent the 25th and 75th percentiles, the whiskers are the 5th and 95th percentiles and the green triangles are the mean values. The AODs do not show any evident variations with the weekdays, with average and median values of about 0.14 and 0.13, respectively. On the other hand, the NO₂ tropospheric VCDs show an evident weekly trend. They increase during the first days of the week, reaching the peak on Wednesday for the mean and on Thursday for the median, and tend to decrease during the second half of the week, reaching the minimum on Sunday. Indeed, the NO₂ VCDs on Sunday are on average about 20% lower than the mean weekly VCDs.

3.3. Inter-Comparison with Satellite Data

Validation of satellite data is one of the primary uses of MAX-DOAS instruments. While retrieved MAX-DOAS NO₂ products (tropospheric and total columns) are considered a reference for satellite validation (e.g., refs. [9,48,49,50]), aerosol extinction retrievals are not. However, the quality of aerosol extinction retrievals also affects the quality of subsequent gaseous retrievals. For this reason, we compare the AODs obtained from the SkySpec-2D retrieved extinction profiles at 477 nm with the MODIS-MAIAC retrievals at 470 nm. As reported in Section 2.3, MODIS Terra equatorial overpass is at 10:30 while for Aqua it is at 13:30. The scatter plot between MODIS-MAIAC AODs in 1 × 1 km² cell, containing the SPC site, and MAX-DOAS AODs, averaged within ±15 min from MODIS-MAIAC sensing time and considering all the three azimuth directions, is reported in Figure 12. The results show a good correlation of about 0.73, a bias of 0.037 and a standard deviation of about 0.08. About 69% of the points are in the range of MODIS MAIAC Expected Error (EE) of ±(0.05 + 0.15 × AOD). In general, MODIS-MAIAC AODs are larger than the DEAP ones.

Figure 13 shows the monthly behaviour of SkySpec-2D AODs compared to MODIS-MAIAC Aqua and Terra ones. In general, the seasonal behaviours of SkySpec-2D and MODIS-MAIAC are quite similar, with more significant differences in January, April and May. As shown in Figure 14, the hourly behaviours of SkySpec-2D and MODIS AODs are very similar with slightly higher values at the beginning of the day. It is interesting to note that, by using the two satellites (with different overpasses), part of the information on hourly AOD distribution can be recovered.

The validation of Sentinel-5P TROPOMI NO₂ tropospheric VCDs is one of the targets of the FRM4DOAS network. Figure 15 shows the time series of SkySpec-2D NO₂ tropospheric VCDs from MAX-DOAS scans in coincidence with Sentinel-5P TROPOMI data for the first two years of SkySpec-2D measurements. Their differences are also reported. For the coincidence, we considered the Sentinel-5P TROPOMI data averaged in a circle of 5 km radius around SPC and the MAX-DOAS cloud free data in the three azimuth directions averaged within $\pm 15$ min around the Sentinel-5P TROPOMI overpass time. The Sentinel-5P TROPOMI and MAX-DOAS error bars represent the standard deviations of the averaged data. SkySpec-2D and Sentinel-5P TROPOMI show similar behaviour with higher values in winter and lower values in summer, as already reported in Figure 10. The bias is of the order of −14% (−6 × 10¹⁴ molec/cm²).

Similarly to Figure 12 and Figure 13, Figure 16 and Figure 17 presents the scatterplot of Sentinel-5P TROPOMI versus SkySpec-2D NO₂ tropospheric VCDs and their monthly behaviour. NO₂ VCDs are higher in autumn-winter and lower in spring-summer. SkySpec-2D VCDs are higher than the Sentinel-5P TROPOMI ones in autumn-winter. However, the correlation with Sentinel-5P TROPOMI is good (around 0.9) and the major differences are found for higher NO₂ VCDs (above 0.5 × 10¹⁶ molec/cm²).

The underestimation of NO₂ tropospheric VCDs in polluted regions by Sentinel-5P TROPOMI has already been documented. As reported in Section 2.3, Sentinel-5P TROPOMI tropospheric NO₂ VCDs are retrieved using the retrieval algorithm developed at KNMI [44] and NO₂ profiles from the TM5-MP model.

To correct the Sentinel-5P TROPOMI values according to Equation (8), we need the values of the SkySpec-2D NO₂ retrieved profiles interpolated to Sentinel-5P TROPOMI AK altitude levels. For this work, we decided to average the SkySpec-2D profiles in a temporal range of ±15 min around the Sentinel-5P TROPOMI overpasses to reduce the standard deviations of the averaged points. We filtered out MAX-DOAS profiles as already explained in Section 2.2.1.

Figure 18 reports the comparison of NO₂ tropospheric VCDs from SkySpec-2D and Sentinel-5P TROPOMI before and after the correction. As can be seen, the bias moves from slightly negative to slightly positive and the correlation slightly decreases from 0.90 to 0.87. The positive bias introduced by the correction is mainly due to the low values overestimated by Sentinel-5P TROPOMI. However, a better agreement exists for the high values with the ordinary least square slope increasing from 0.75 to 0.89.

4. Discussion

The starting point of our work was the development and validation of the retrieval code DEAP for MAX-DOAS aerosol extinction and NO₂ tropospheric profiles. The availability of analogous results from two FRM4DOAS reference retrieval codes (MMF and MAPA) and from the state-of-the-art BOREAS algorithm was then exploited. The validation of the DEAP code with synthetic dSCDs showed that, for AODs, DEAP performs at least as good as MMF and MAPA in terms of slope, correlation and RMS, and worse in terms of intercept. Considering the NO₂ tropospheric VCDs, the DEAP correlations and RMS are better than MMF and MAPA ones. However, the slope is worse, being too high. Regarding tropospheric profiles, the DEAP shows a correlation of 0.94 for aerosol extinction at 477 nm, and a low RMS of 0.05 and a correlation of 0.65 for NO₂ tropospheric profiles. The DEAP code applied to actual data produces good results compared to MMF, MAPA and BOREAS. In particular, the DEAP AODs present low biases with respect to all three other codes but high correlations only with respect to MMF and BOREAS (higher than 0.97). The lower correlation and high standard deviation found with MAPA suggest that MAPA AODs are more scattered with respect to the other codes. Considering the NO₂ tropospheric VCDs retrieved with real data, DEAP presents similar performances with respect to MAPA, MMF and BOREAS, with high correlations (about 0.99). Once the code was validated we exploited it over the two full years of measurements at SPC. The AOD diurnal cycle is in-line with the one observed in the past at SPC [51]. The AOD values are higher during the day and in the afternoon. The patterns are similar over the months with some oscillations due to different seasons. Regarding the weekly cycle, the results reported here highlight the absence of a weekly cycle in the AODs at SPC.

The tropospheric NO₂ VCDs show higher values between October and March and lower smooth values from April to September. The observed seasonal cycle is the same as reported in [52] using Global Ozone Monitoring Experiment (GOME) data on the Lombardy from 1996 to 2002. Additionally, the observed values are similar to the ones reported in [52]. The weekly cycle we observed is weak, with Sunday values lower by about 20% with respect to Monday–Friday averages. In [52] they found average higher NO₂ values and a stronger weekly cycle, with Sunday values significantly lower than those on the other days of the week. This difference is mainly related to the different locations. Indeed, although both SPC and Milan are located in the Po Valley, the NO₂ concentrations around Milan are highly influenced by local emissions due to factories and traffic, which can present stronger weekly patterns. On the other hand, SPC is located in a rural area, where the anthropogenic emissions are lower than the background. In SPC, the accumulation of gaseous pollutants is favored by atmospheric stability, thermal inversions and weak winds, in particular during cold winter months, as in [53]. Our results at SPC show that the local NO₂ production by traffic is far from the major production sources. Hence the SPC observatory is well representative of background conditions in the Po Valley.

Since the validation of satellite data is one of the primary uses of the MAX-DOAS retrievals, we also compared our results to data from MODIS-MAIAC and Sentinel-5P TROPOMI. Although MAX-DOAS measurements of aerosol extinction are not considered a reference for satellite validation, unlike MAX-DOAS NO₂ and HCHO retrieved tropospheric and total VCDs, the quality of aerosol extinction profiles also affects the quality of the subsequent retrieval of gas profiles. For this reason, we have compared the results obtained with the DEAP code for aerosol extinction with MODIS-MAIAC results for Aqua and Terra satellites. In addition, we also compared the NO₂ tropospheric VCDs obtained by DEAP with the one measured by Sentinel-5P TROPOMI. In general, the seasonal behaviour of SkySpec-2D and MAIAC is similar, with larger differences in March, April and May. The overall agreement is good with most points (about 69%) laying between MODIS-MAIAC EE. The possibility of exploiting two satellites (Terra and Aqua) with different overpass times allows us to compare our results over different times of the day. The MODIS-MAIAC data confirm the diurnal cycle observed by the MAX-DOAS instrument. We found a bias with respect to MODIS-MAIAC that is below 0.04. As expected, the SkySpec-2D NO₂ tropospheric VCDs are higher than the Sentinel-5P TROPOMI ones, especially in autumn-winter when the NO₂ tropospheric VCDs are higher. We found a bias of −0.6 × 10¹⁵ molec/cm² with a standard deviation of 1.8 × 10¹⁵ molec/cm² with respect to Sentinel-5P TROPOMI data. The underestimation of Sentinel-5P TROPOMI values has already been documented in [54]. The bias we found is slightly smaller than the one reported in [54] for MAX-DOAS stations with similar average NO₂ VCDs. As said, the reported Sentinel-5P TROPOMI NO₂ tropospheric VCDs underestimation in polluted regions has been attributed to the TM5-MP NO₂ profiles used in Sentinel-5P TROPOMI data processing. Several works have suggested the correction of Sentinel-5P TROPOMI data through the use of Sentinel-5P TROPOMI AKs and MAX-DOAS profiles. Furthermore, in this work, the agreement between MAX-DOAS and Sentinel-5P TROPOMI was improved by applying the correction suggested in the Sentinel 5-P user manual section 8.8: the bias moved from slightly negative to slightly positive, and the slope moved from 0.72 to 0.87. In [55], the authors compared Sentinel-5P TROPOMI VCDs over Europe with the Copernicus Atmosphere Monitoring Service (CAMS) regional models, highlighting also in this case the Sentinel-5P TROPOMI NO₂ underestimation. In addition, they replaced the TM5-MP profiles with the profiles generated with CAMS. The comparison of these “corrected” Sentinel-5P TROPOMI products with ground-based results showed an improvement in Sentinel-5P TROPOMI products. In particular, in the Po Valley, they show that the NO₂ columns increase when using different a priori profiles. This result agrees with our test when using MAX-DOAS profiles as a priori. It is worth noting that replacing the a priori does not always improve the validation results [56].

5. Conclusions

This work presents the dataset of AODs and NO₂ tropospheric VCDs retrieved from MAX-DOAS measurements performed at SPC with the SkySpec-2D instrument. The retrieval has been performed with DEAP, a retrieval code specifically developed and validated against state-of-the-art codes selected by the FRM4DOAS network. The DEAP code can be easily adapted to other instruments, like Pandora or other MAX-DOAS, if information on measuring geometries and used spectral range is provided together with SCDs.

The results we present here highlight the potential of the SPC observatory in providing a valuable insight into the background Po Valley conditions. This can be seen from the weekly cycle of both AODs and NO₂ tropospheric VCDs, where no strong variations can be observed between the different days of the week and with a minimum in NO₂ tropospheric VCDs on Sunday that is 20% lower than in the rest of the week. In addition, the tropospheric NO₂ VCDs time series shows a strong seasonal cycle, with higher values in winter due to (a) the longer lifetime of NOx in winter and (b) larger emissions of NOx in winter combined with the stability of the atmosphere, and lower values in summer. The smooth behavior of the time variations reflects the absence of local sources near the observatory. Thus, the SPC MAX-DOAS site proves to be a valid candidate for validating the Sentinel-5P TROPOMI and future Sentinel-4 and 5 missions under the background pollution conditions of the Po Valley.

In 2022–2023 the “Giorgio Fea” observatory improved its remote sensing capabilities with the acquisition of AErosol RObotic NETwork (AERONET) compliant sun photometer (which started operating in February 2023) and a Raymetrics LIDAR (in the setting up phase), while an Automatic LIDAR/Ceilometer is operating on the site. The synergy of these instruments will be hopefully exploited in the future. The potentiality of this set of remote sensing and ground based measurements at SPC can play a crucial role in both satellite validation and scientific studies.