The main purpose of the Pandora head sensor is to collect light within a specific field of view, transmit light through optical filters (e.g., U340 to block visible part of solar spectrum), and focus it onto the fiber-optics patch cable for transmission to the spectrometer. The Pandora head sensor consists of several components: sealed aluminum cylindrical housing, a wedged fused-silica entrance window (25 mm in diameter), two filter wheels with motors, baffle holding tubes, a lens, a fiber-optics cable, an electronics board (Fig. ), and a desiccator bag. The baffle holding tube, the two filter wheels, and the dark filter wheel plug were machined from POM-H Delrin, a trade name for polyoxymethylene, engineering thermoplastic up to March of 2019. The desiccant bags (McMaster-CARR model 2189K76, manufacturer Multisorb Technologies, model name MINIPAX) contain activated carbon (43 %–48 % by weight) and silica gel (43 %–48 % by weight) enclosed in Tyvec material (high-density polyethylene fiber, 5 %–15 % by weight) and are designed to remove moisture as well as some VOCs including HCHO.

Polyoxymethylene (POM) has a wide range of applications due to (1) excellent mechanical (high tensile strength, rigidity, and toughness) and electrical properties at temperatures from -40 to 130 ∘C (short-term); (2) low reactivity with and low permeability to many chemicals including organic solvents, fuels, and gases at room temperature; and (3) ease of processing on standard thermoplastics equipment . Six major manufacturers produce about 70 % of POM worldwide, and each company has its own trade name (e.g., Ticona GmbH, Germany: Celcon^®; Polyplastics Co., Ltd., Japan: POM-C Duracon^®, Tepcon^®; E.I. Du Pont de Nemour & Co., USA: POM-H Delrin^®).

POM-H Delrin used for Pandora head sensor parts is a homopolymer POM (POM-H, Ensinger Hyde: black Delrin ecetal resin II150ebk602sheet 3/4×6×6) purchased from McMaster Carr (part nos. 8575K145 and 8576K21). POM-H is produced by polymerization of purified gaseous formaldehyde in an organic liquid reaction medium in the presence of polymerization catalysts. The resulting polymer has a crystalline granular structure with macromolecules ending in at least one unstable hydroxyl group. These hydroxyl groups are responsible for POM-H thermal instability. POM deterioration occurs mainly due to the following processes:

depolymerization (unzipping),

We hypothesized that the thermal instability of POM-H Delrin resulted in HCHO release at higher temperature and was the source of the temperature-dependent formaldehyde interference in Pandora direct-sun HCHO. To test this hypothesis, we added an internal temperature sensor in April 2019 to monitor the internal head sensor temperature in a few instruments. We have evaluated the range of internal head sensor temperatures measured at various sites: Pandora 2 (Greenbelt, MD), 148 (Blacksburg, VA; Rotterdam and Cabauw, the Netherlands) and 155 (Boston, MA) (with an emphasis on the US East Coast where several intergovernmental field campaigns took place; see Table ). Figure  shows that internal head sensor temperatures ranged between 20–25 ∘C during the nighttime hours and up to 45–50 ∘C during the daytime hours in summer months. During colder months, the temperature ranged between 0 to 25 ∘C. These data suggested that HCHO generation was potentially relevant at internal to head sensor temperatures between 20 and 50 ∘C.

Internal to head sensor temperatures are determined by the following heat transfer processes between the head sensor and the surrounding environment: (a) convective heat transfer (natural and forced) due to wind; (b) radiant heat transfer due to shortwave absorption, longwave emission, and longwave absorption; (c) conduction between the head sensor and the tracker brackets; and (d) energy generation inside the head sensor.

The head sensor collimator was protruded through the enclosure window to avoid measuring any potential HCHO outgassing inside the enclosure itself (e.g., paint). The laboratory hosting the measurements was temperature controlled (20–23 ∘C) and had both an air supply and air intake, and the door to the room was open to improve ventilation.

The collimator was pointed at the Gooch and Housego 1000 W QHL controlled by the current precision source (OL 410-1000). The QHL was operated at 8 A. The distance between the QHL and the collimator was 50 cm.

The enclosure temperature was controlled by NesLab 7 recirculating bath and a LYTRON heat exchanger with two fans to ±0.1 ∘C. The NesLeb 7 temperature sensor was placed near the Pandora head sensor. Enclosure air temperature near the head sensor, front plate Pandora head sensor temperature, and Pandora head side temperature were recorded during all the measurements using fast response stick-on surface thin film Platinum Resistance Temperature Detectors (Pt100 RTDs; 3-Wire, The Sensor Connection). An ADAM-4015 six-channel RTD Module with Modbus digitized the RTD signal. The temperature inside the newer-generation head sensor (since April 2019) was measured using the preinstalled Bosch BME280 digital humidity, pressure, and temperature sensor on a SparkFun Atmospheric Sensor Breakout Board. PT100 RTD elements were inter-calibrated. They agreed within the manufacturer specifications (<0.15 ∘C). The accuracy of BME280 was harder to verify due to Pandora head internal power generation of 2 W (manufacturer reported accuracy is ±1.0 ∘C between 0 and 65 ∘C).

Since only one of the tested Pandora head sensors was equipped with the internal temperature sensor, we determined an outside measurement that is the most representative of the internal temperature. This was done by comparing surface temperature measurements by the PT100 RTD elements at various locations on the Pandora head sensor versus the internal to head sensor temperature. As expected, there is some time delay in response between the surface measurements and the internal temperature. This delay is rate specific. Strong linear correlation between the surface measurements and internal temperature measurements (accounting for transient heat transfer) was observed for both front plate (slope = 0.970; intercept = 4.74 ∘C; RMSE = 0.059 ∘C) and side (slope = 0.999; intercept = 6.79 ∘C; RMSE = 0.087 ∘C). Since the electronics board heat sink is connected to the front, we use the front plate surface temperature as the proxy for the internal temperature.

The Pandora spectrometer temperature was controlled using a Pandora thermoelectric controller at the set temperature of 15 ∘C. The measurements were averaged over 40 s sequentially switching between open, plug, U340, plug filter wheel positions to simulate direct-sun measurements.

In addition to the QHL one head sensor (Pandora 118) was also analyzed using a 300 nm LED (Thorlabs M300L4) controlled by a high-precision LED driver (Thorlabs DC2200). In the case of the LED source, the Pandora collimator pointed into an 8.3 cm Labsphere Spectralon^® reflectance material integrating sphere illuminated by the LED.

Since summer 2019, new Pandora head sensors are POM-H Delrin free. POM-H Delrin was replaced with molybdenum disulfide (MoS2)-filled easy-to-machine wear-resistant cast nylon 6/6 also purchased from McMaster (Tecamid 66 MO, polyamide >90 % by weight, MoS2<10 % by weight, manufactured by ENSINGER INC). To evaluate potential thermal oxidation of polyamide and MoS2 by air oxygen, three new head sensors (Pandora 165, 167 and 168, manufactured at the end of 2019) upgraded with the nylon parts were tested using QHL (1000 W) and 300 nm LED sources. The enclosure temperature varied from 10 to 55 ∘C over 8 h, with 1 h at 55∘ and 8 h cooling from 55 to 10 ∘C. This translated to internal temperatures from 17 to 60 ∘C. The QHL current was set at 7.5 A. Pandora spectra were binned within 40 s for measurements with no filters (open, single-spectrum integration time 2.4 ms, about 12 550 cycles per measurement and dark 2320 cycles); 240 s with a U340 filter (integration time 12.9 ms, about 15335 cycles per measurement and dark 2835 cycles); and 240 s with a BP300 filter (integration time 117 ms, about 1730 cycles per measurement and dark 320 cycles). The spectrometer electronics board temperature was maintained at 12.9 ∘C (controller set temperature 5 ∘C). We also repeated the test with 300 nm LED at a constant current of 350 mA with no filters for 100 s total integration time. The experiments were designed to ensure low noise in case of small emissions of HCHO or presence of other species.

The reference spectrum was collected at the lowest internal head sensor temperature (about 17 ∘C). DOAS fitting included only HCHO and sulfur dioxide, SO2, molecular absorption cross sections , the polynomial order for the broadband attenuation (PL) was set to 5, and a 0-order polynomial represented the offset. Only the dark current spectrum was subtracted from the data since all other parameters are expected to be constant (e.g., no wavelength shift in the temperature-controlled lab, no need for nonlinearity correction, no need for pixel response non-uniformity correction, etc.). The DOAS fitting was performed using QDOAS v3.2 program. SO2 was fitted as a precaution since sulfur contained in MoS2 has been reported to oxidize to SO2 at temperatures higher than 140 ∘C (https://core.ac.uk/download/pdf/10884897.pdf, last access: 21 January 2021).

Analysis of Pandora 148 data over a 9 h period showed that the equilibrium between HCHO generation and removal processes inside the head sensor is reached almost instantaneously (at the DOAS fitting accuracy). Investigation of the actual process mechanisms are outside of the scope of this paper. However, it is probably also controlled by desiccant-activated carbon adsorption as a function of temperature, in addition to “pure” solid POM-H vapor-phase processes (Sect. ).

Figure  shows that HCHO columns inside Pandora head sensors follow exponential dependence on temperature irrelevant of heating or cooling rates for all four tested head sensors (Eq. ). This temperature dependence does not show any hysteresis at the timescales relevant to this study. HCHO columns in all four sensors had the same exponential function damping (b=0.0911±0.0024 ∘C-1) but different amplitudes (and most likely absolute offsets). 2ΔSΔThs=aexp⁡b⋅Ths-exp⁡b⋅Thsref

The newest head sensor (Pandora 148) produced the lowest amount of HCHO (about 0.3 DU) at an internal temperature of about 50 ∘C (front panel external temperature 45 ∘C). Pandora 118 generated the largest – about 3 DU at the same temperature. Pandora 46 and 21 were in between. No clear trend was observed between the age of the instruments and the HCHO amount produced in the head sensor. Pandora 148 head sensor was evaluated for temperature dependence of HCHO several times over 5 months and did not show any difference in the HCHO generation during that period.

Initially we conducted DOAS fitting of HCHO and SO2 absorption within their standard fitting windows 332–359 and 307–328 nm , respectively. No HCHO or SO2 was detected above the optical depth rms noise level of 5×10-5 from the new “POM-H Delrin free” head sensors as a function of internal head sensor temperature relative to 17 ∘C. To consolidate HCHO/SO2 results for both species and to evaluate residuals we have done DOAS fitting at a broader fitting window: 300–350 nm. Figure  shows a retrieval example for Pandora 167 (300–350 nm fitting window from spectra collected with U340 filter). The spectra were also evaluated for any absorption across the entire instrument wavelength range from 300 to 530 nm by only taking the radiance ratio but not fitting any trace gases. We did not see any signatures above the instrumental noise level.

Since direct measurements of HCHO in the head sensor are harder to perform during routine direct-sun observations, we evaluate the head sensor HCHO production effect using two instruments: Pandora 32 and Pandora 2 (neither was evaluated in the laboratory according to Sect. ) during outdoor operation. Both instruments operated side-by-side at the NASA Goddard Space Flight Center in Greenbelt, MD (38.9926∘ N, 76.8396∘ W, 90 m a.s.l.), in direct-sun mode during July 2019–January 2020. Pandora 2, originally built in 2009 with the standard POM-H Delrin components, was upgraded in June 2019 with POM-H Delrin free components and internal temperature sensor (see Sect. ). The Pandora 32 head sensor, originally built in 2012, still contains the original POM-H Delrin components and does not have an internal temperature sensor. To evaluate the effect of internally generated HCHO on the direct-sun total column measurements during a “typical” field campaign study, we used 1.5 months of data from 30 August to 15 October 2019, when both instruments had minimal instrumental issues.

The evaluation consists of several steps.

Use Pandora 32 and 2 data to estimate the exponential HCHO production amplitude inside Pandora 32 during the selected 1.5 months:5a=medianΔSμ,ΔThs,t-ΔS(μ,t)atmexp⁡b⋅Ths(t)-exp⁡b⋅Thsref=medianΔSμ,ΔThs,tP32-ΔS(μ,t)P2exp⁡b⋅Ths(t)-exp⁡b⋅Thsref.

The assumption about the same internal temperature for Pandora 32 and 2 is based on almost identical head sensor designs, collocation and the same mode of operation. The derived HCHO production amplitude for Pandora 32 is 0.0133 DU. DOAS analysis to calculate HCHO columns was performed in the fitting window 332–359 nm with PL=4 and an offset and wavelength shift of polynomial order 1. In addition to HCHO at 298 K (Meller and Moortgat, 2000), absorption by ozone (O3, at 223 and 243 K, Malicet et al., 1995), nitrogen dioxide (NO2, a linear temperature model, Vandaele et al., 1998), an oxygen collision complex (O2O2, at 294 K; ), and bromine monoxide (BrO, at 223 K; ) were fitted. Their high-resolution molecular absorption cross sections were convolved with the Pandora instrument transfer function prior to DOAS fitting (for convolution details, see ). The reference spectrum was created by averaging all spectra within ±5∘ of the minimum solar zenith angle (SZA) on a cloud-free day 15 October 2020 with an average internal head sensor temperature of ≈29 ∘C.

Figure  shows a linear correlation between ΔSHCHO measured by Pandora 32 and differential columns estimated from Pandora 2 measurements and HCHO produced by the Pandora 32 head sensor. The linear regression analysis between these data sets shows that the exponential function represents a reasonable estimation of the internally generated HCHO by Pandora 32 head sensor measurement during direct-sun measurements (slope = 1.00, intercept = -0.03 DU and R2 = 0.92). Deviations between the true Pandora 32 measurements and simulated from Pandora 2 measurements and internally produced HCHO are also due to small differences in Pandora 32 and 2 fields of view, diffusers, and pointing accuracy.

Figure  shows estimated HCHO column density inside Pandora 32 head sensor (red) based on the exponential function coefficients (a = 0.0133 DU; b = 0.0911 ∘C-1) and colocated Pandora 2 internal temperature. Internally generated HCHO amount is smaller during the winter months (<0.2 DU) and reaches up to 1.15 DU during hot summer days for the Pandora 32 head sensor. Since the total SHCHO is divided by the direct-sun air mass factor, the head sensor contribution to the total vertical column is also solar zenith angle dependent in addition to the internal head sensor temperature (this should not be confused with the actual amount in the head sensor). Its contribution is the largest during the middle of the day near the summer solstice and smallest at large solar zenith angles (80∘ in this study). Due to colder temperatures and larger AMFs during winter months over non-tropical regions, head-sensor-generated HCHO contribution to the vertical column is small (<0.1 DU for this instrument). Figure  shows that during the cooler and windy summer days (e.g., 23 August 2019), head sensor HCHO can result in a relatively small amount contributing to the total direct-sun column. Similar behavior was observed during the KORUS-AQ field campaign when altitude integrated in situ aircraft measurements mostly agreed with Pandora columns in the morning at higher solar zenith angles and disagreed during the middle of the day. Figure a shows that warm-season measurements at solar zenith angle <60∘ are most impacted. Figure b shows that a significant number of measurements have a contribution to the order of background level (≈0.5 DU) or higher, which significantly impacts the accuracy of direct-sun observations.

In this section we evaluate the variability of internally generated HCHO and its effect on MAX-DOAS retrievals using “closest” in time (<10 min) zenith reference spectra during the TROLIX'19 campaign. Pandora instruments 148 and 118 participated in the TROLIX'19 campaign. The main goal of TROLIX'19 was validation of TROPOMI L2 main data products including UV aerosol index (UVAI), aerosol layer height, NO2, O3, and HCHO under a wide range of atmospheric conditions. Pandora 148 has been tested in the laboratory (see Sect. ) three times over the period of 5 months and showed no changes in internally produced HCHO as a function of temperature. We use Pandora 148 data collected during the TROLIX'19 campaign to estimate the effect of internally produced HCHO on the multi-axis ΔSHCHO retrieved with individual scan reference. Pandora 148 was equipped with an internal temperature sensor and had well characterized internal HCHO temperature dependence before deployment in western Rotterdam metropolitan area (51.9172∘ N, 4.4066∘ E, 7 m a.s.l.) during September 2019. Pandora 118 was characterized for temperature-dependent HCHO production in December 2019, 3 months after the TROLIX'19 deployment.

The effect of internally generated HCHO on multi-axis ΔSHCHO using the reference zenith spectrum that was closest in time collected a maximum of 10 min apart from the rest of the scan spectra is evaluated according to the following steps:

calculating the head-sensor-produced amount, SHCHO(Ths(t)), based on the head sensor temperature and Pandora 148 exponential function coefficients: a = 0.0041 DU and b = 0.0911 ∘C-1;

Figure  shows that Pandora 148 internally generated HCHO contribution to the multi-axis ΔSHCHO while using single scan reference is very small (<0.005 DU). As expected, this small contribution is mostly due to lower internal temperature variations within 10 min period and partially due to small generation rates inside Pandora 148 head sensor (a = 0.0041 DU, see Fig. ). Since Pandora 118 was only characterized once for temperature dependence of HCHO production we do not have high confidence in its temperature dependence “stability”. If we assume that the exponential function amplitude was the same in September as in December (a = 0.049 DU), Pandora 118 head sensor contributed almost 10 times more than Pandora 148 to multi-axis ΔSHCHO. Even in this case, the resulting amount is smaller than 0.05 DU, which is lower than the DOAS fitting noise for most DOAS instruments (0.3 DU = 8×1015 molecules cm-2, Table 7 in .)

Five Pandora instruments participated in the Second Cabauw Intercomparison campaign for Nitrogen Dioxide measuring Instruments (CINDI-2) that took place at Cabauw, The Netherlands (51.97∘ N, 4.93∘ E, September 2016) . A formal semi-blind intercomparison exercise was performed to compare ΔSgas of NO2, HCHO, O2O2, and O3 measured by 36 spectroscopic systems from 24 institutes during 17 d in September 2016. To limit any variability due to differences in temporal sampling by each instrument for semi-blind intercomparison exercise, all multi-axis daily scans were analyzed using that day's local noon spectra. This type of analysis results in a stronger contribution of the internally generated HCHO on the Pandora ΔSHCHO that were compared with the rest of DOAS instruments. Since none of the Pandora instruments in September 2016 were equipped with an internal temperature sensor, we use Pandora 148 data during TROLIX'19 measurements as a surrogate for the CINDI-2 campaign. Pandora 148 was deployed at a location about 38 km southwest of the CINDI-2 site during the same month of the year as CINDI-2. While differences in atmospheric conditions are expected between the sites and years (2019 versus 2016), we assume that general trends in internal Pandora head sensor temperature are similar over the measurement periods. Only one Pandora (Pandora 118) was tested for internal HCHO generation, but this was done more than 3 years later. We assume that Pandora 32 exponential amplitude is more representative of a “typical” Pandora rate than Pandora 118.

To evaluate the effect of internally generated HCHO on ΔSHCHO used for semi-blind intercomparison during CINDI-2, a single reference spectrum was used to analyze the entire day of multi-axis data during TROLIX'19 using the following steps:

calculating the head sensor produced amount based on Pandora 148 head sensor temperature and Pandora 32 exponential function coefficients: a = 0.013 DU; b = 0.0911 ∘C-1.

Figure  shows that the internally generated HCHO contribution to ΔS is negative before local noon, positive during early afternoon, and negative again during late afternoon. Since the internal temperatures did not vary by more than 15 ∘C during daily measurements and maximum did not exceed 40 ∘C the overall effect is in general small <0.1 DU with slightly negative bias (Fig. ) for an instrument similar to Pandora 32, 21 and 46. While we do not know the exact HCHO internal generation rates for the Pandora instruments deployed during CINDI-2 we can assume that the minimum corresponds to Pandora 148 and maximum to Pandora 118, which is about 3.2 times smaller or 3.7 times larger than in Fig.  and .

Note, that DOAS analysis using daily noon zenith reference spectra was implemented only for the formal semi-blind intercomparison of ΔSCD exercise . Full data processing and inversion to the final products, tropospheric columns and profiles, was done using individual scan zenith spectra and not daily noon zenith spectra .