An instrument for measurements of BrO with LED-based Cavity-Enhanced Differential Optical Absorption Spectroscopy

The chemistry of the troposphere and specifically the global tropospheric ozone budget is affected by reactive halogen species such as bromine monoxide (BrO) or chlorine monoxide (ClO). Especially BrO plays an important role in the processes of ozone destruction, disturbance of NO x and HOx chemistry, oxidation of dimethyl sulfide (DMS), and the deposition of elementary mercury. In the troposphere BrO has been detected in polar regions, at salt lakes, in volcanic plumes, and in the marine boundary layer. For a better understanding of these processes, field measurements as well as reaction chamber studies are performed. In both cases instruments with high spatial resolution and high sensitivity are necessary. A Cavity-Enhanced Differential Optical Absorption Spectroscopy (CE-DOAS) instrument with an open path measurement cell was designed and applied. For the first time, a CE-DOAS instrument is presented using an UV LED in the 325–365 nm wavelength range. In laboratory studies, BrO as well as HONO, HCHO, O 3, and O4 could be reliably determined at detection limits of 20 ppt for BrO, 9.1 ppb for HCHO, 970 ppt for HONO, and 91 ppb for O 3, for five minutes integration time. The best detection limits were achieved for BrO (11 ppt), HCHO (5.1 ppb), HONO (490 ppt), and O 3 (59 ppb) for integration times of 81 minutes or less. Comparison with established White system (WS) DOAS and O 3 monitor measurements demonstrate the reliability of the instrument.


Introduction
Besides their well known effects on stratospheric ozone, reactive halogen species (RHS) are responsible for a series of phenomena in various compartments of the troposphere that includes volcanic plumes, coastal areas, salt lakes, and the polar boundary layer.For instance, RHS, and especially bromine, are of scientific interest because they are mainly responsible for ozone depletion in the arctic boundary layer during polar spring.This phenomenon was first observed by Bottenheim et al. (1986), and several other studies followed that established a link between O 3 depletion and RHS (e.g., Barrie et al., 1988Barrie et al., , 1994;;Hausmann and Platt, 1994;Bottenheim et al., 1990;Sturges et al., 1993;Platt and Lehrer, 1996;Kreher et al., 1997;Tuckermann et al., 1997).The halogen catalyzed ozone depletion events occur only during polar sunrise.The release mechanism for reactive bromine is described by the "bromine explosion" mechanism (Platt and Lehrer, 1996).The phenomenon is still not completely understood and is the subject of current research (Saiz-Lopez et al., 2007;Pöhler et al., 2010;Friess et al., 2011).
Also outside polar regions RHS strongly affect the ozone budget and the oxidation capacity of the atmosphere.Bromine monoxide (BrO) was detected in the mid-latitude marine boundary layer at different locations (Leser et al., 2003;Saiz-Lopez et al., 2004;Read et al., 2008;Mahajan et al., 2010).At salt lakes high mixing ratios of BrO could be detected, up to 20 ppt at the Salar de Uyuni, Bolivia, (Hönninger et al., 2004) and up to 220 ppt at the Dead Sea (Hebestreit et al., 1999;Matveev et al., 2001;Tas et al., 2008).Most of these observations were performed with long-path or multi-axial differential optical absorption spectroscopy (LP-DOAS or MAX-DOAS, respectively) measurement techniques averaging the concentration over several kilometers.For a better understanding of the involved processes, field measurements with high spatial resolution are essential.Cavity-enhanced (CE) DOAS offers the opportunity of mobile in situ measurements of RHS without the risk of inlet and wall losses.In addition, RHS release, transformation, and loss processes can be studied in reaction chambers (e.g., Buxmann et al., 2012;Ofner et al., 2012).For these studies, compact yet highly sensitive instruments for RHS measurements are also required.
O' Keefe and Deacon (1988) started to use optical cavities with highly reflective mirrors to observe absorbers by applying cavity ring-down spectroscopy (CRDS).With this approach, the decay of light intensity leaking from the resonator after switching off the pump light source is measured.Since then the technique has been continuously refined (Zalicki and Zare, 1995;Paldus et al., 2000;Ball et al., 2001;Fawcett et al., 2002;Simpson, 2003;Ball and Jones, 2003;Schuster et al., 2009).A limitation of most CRDS instruments is that any extinction (absorptions or scattering) beside that of the investigated trace gas at the applied wavelength leads to interferences and must be avoided.Thus, measurements at spectral ranges where several molecules absorb cannot be used (e.g., the UV spectral range due to O 3 and NO 2 absorptions).Additionally this requires the removal of aerosols and thus typically demands closed resonators with aerosol inlet filters.Moreover, this approach introduces the risk of trace gas loss as well as possible chemical reactions on the inlet and filter (e.g., conversion of HOBr to Br 2 , Neuman et al., 2010; or BrO self-reaction, Liao et al., 2011).A second approach, the so-called cavity-enhanced absorption spectroscopy (CEAS), was developed, which relies on measuring the transmittance of the resonator by recording the leak-out intensity (Engeln et al., 1998;Peeters et al., 2000).In contrast to CRDS, the CEAS approach uses a continuous wave (CW) source of radiation.Based on CEAS (e.g., Fiedler et al., 2003;Langridge et al., 2006;Venables et al., 2006;Orphal and Ruth, 2008;Vaughan et al., 2008;Washenfelder et al., 2008;Langridge et al., 2008;Varma et al., 2009;Chen and Venables, 2011;Chen et al., 2011), Kahan et al. (2012) applied the broadband (BB) CEAS using broadband light sources (rather than a single wavelength as in the traditional CRDS or CEAS approaches).Ball et al. (2004) combined BBCEAS for the data evaluation with the established DOAS technique (Platt and Stutz, 2008), creating the CE-DOAS.With this technique, different trace gases can be detected simultaneously and at high spatial resolution (Ball et al., 2004;Gherman et al., 2008;Platt et al., 2009;Meinen et al., 2010;Thalman and Volkamer, 2010).We note that the BBCEAS retrieval algorithm published by Fiedler et al. (2003) relies -like DOAS -on the characteristic narrow band absorption features of the molecules for their quantification and is thus in principle equivalent to DOAS.The theoretical description of CE-DOAS can be found in Platt et al. (2009).To date, to our knowledge, only two cavity-based measurement systems to detect BrO in the UV wavelength range exist (Chen and Venables, 2011;Grilli et al., 2012), and neither system applies the DOAS technique.Both of them use a heavy and highly power consuming light source (e.g., tunable laser and xenon arc lamp).This means that there is a lack of in situ instruments allowing mobile measurements of BrO radicals in the field without requiring laboratory infrastructure.
The mode-locked frequency-doubled cavity-enhanced spectrometer described by Grilli et al. (2012) has a 7 m Teflon inlet tube.Although the instrument provides BrO detection with remarkable good detection limits of 1.7 ppt at a time resolution of 30 s, there is a possible effect of inlet losses (Neuman et al., 2010;Liao et al., 2012;Buys et al., 2013).The authors state undetectable BrO losses within the relatively large concentration fluctuations of up to ±12 ppt in an artificial air sample.This result requires further validation.
Besides cavity instruments, chemical ionization mass spectrometers (CIMS) are capable of indirect BrO detection, but complex sample preparation and calibration is needed (e.g., Neuman et al., 2010;Liao et al., 2011).This limits the application of these instruments for field studies, where low power consumption, simple operation, and a lightweight setup is required.Additionally, loss or chemical conversion of highly reactive compounds (e.g., BrO or HOBr) on instrument and inlet surfaces occur and can even result in a complete loss of BrO, which can influence the indirect detection of BrO with CIMS instruments (Neuman et al., 2010).Nevertheless, it was demonstrated by Liao et al. (2011) that reliable BrO concentrations can be provided by an optimized and very short inlet CIMS instrument.These measurements were found to be in good agreement with a well established long-path DOAS system (Liao et al., 2011).
Our paper describes the design and development of a specific CE-DOAS instrument, which, for the first time, identifies BrO in the UV wavelength range (325-365 nm) by applying an UV LED.Compared to the xenon arc lamps applied as light sources in previous experiments, UV LEDs offer several advantages particularly important for cavityenhanced measurements: (1) the relatively narrow emission spectrum of LEDs reduces the amount of stray light within the instrument because there is no light emitted outside the spectral range of the high mirror reflectivity.Hence, fewer spectral filters are required, the risk of systematic measurement errors is reduced and the instrument setup becomes less complex.(2) The light output of LEDs can be easily modulated.It is thus possible with the instrument presented here to monitor the effective path length by measuring the ring-down time between measurements.(3) The lifetime of LEDs is typically longer than for arc lamps.This reduces maintenance and the costs of operation are thus lowered.(4) LEDs are much easier to handle than arc lamps and do not, for instance, require a special housing or high voltage for ignition.The power consumption is by far lower, as mentioned above, and less excess heat is produced.This is particularly useful when the light-source is applied in temperature-controlled environments like smog chambers.
Compared to instruments designed for the visible to infrared spectral range wavelength range, Rayleigh scattering and parasitic mirror losses become important in the UV and thus reduce the achievable maximum path length.Both effects are discussed in this paper.To estimate the optical path length in the cavity, the instrument applies the cavity ringdown technique (Meinen et al., 2010) and additionally the measurements of the trace gases O 4 , O 3 , and HONO.At the Atmospheric Chemistry Research Laboratory of the University of Bayreuth, Germany, first laboratory measurements were successfully performed and the detection limits for different trace gases were determined.Also, we compare the CE-DOAS results with an established White system (WS) DOAS for detection of O 3 , HCHO, NO 2 , HONO, O 4 , OClO and SO 2 and a chemiluminescence O 3 monitor.

Method: cavity-enhanced differential absorption spectroscopy (CE-DOAS)
The basic idea of the CE-DOAS technique is to introduce incoherent broadband radiation with the intensity I L into an optical resonator (see Fig. 1), consisting of two mirrors with reflectivity R (for simplicity we assume both mirrors to have the same reflectivity; if the mirrors have different reflectivities R 1 , R 2 , an averaged reflectivity R = √ R 1 R 2 is used).The distance between the mirrors is d 0 (see Fig. 1).Initially only the fraction ρ = 1 − R of the radiation I L enters the resonator.Here for simplicity I L means only the fraction of the radiation from the LED which can be imaged into the resonator.(It is obvious that not all radiation emitted by the LED is passing through the lens and also passes the aperture defined by mirror M 2 .)Once inside the resonator, the radiation is reflected on average 1/(1 − R) times (neglecting other losses).Finally, in the absence of any extinction in the resonator, half of the radiation leaves the resonator through each mirror.The radiation (intensity I tot ) leaving the cavity through the mirror opposite the light source (see Fig. 1) is available for further analysis, as described below.
The DOAS techniques derive the trace gas concentration by the use of the optical density D(λ), which requires the measurement of the intensities without (I 0 (λ)) and with (I (λ)) trace gas absorption (Platt and Stutz, 2008).In a CE-DOAS setup the optical density D CE (λ) is determined from intensity measurements with absorber free filled (e.g., synthetic air) resonator I tot0 and with sample air filled resonator I tot (Platt et al., 2009): The light which leaves the resonator through M 2 is focused by the lens L 2 to a fiber bundle attached to an x-y stage.This fiber bundle transmits the light to a temperature stabilized spectrograph (Ocean Optics QE 65000).Alternatively, the movable mirror "K" can redirect the light to a photomultiplier (PMT) to optionally record a ring-down signal and thus check the path length of the resonator.F 1 and F 2 are band-pass filters (Schott UG11).F 1 protects the PMT and F 2 the fibers from scattered light and LED radiation outside of the reflectivity range of the mirrors.The shutter is used to perform background measurements.The electronic controller receives a constant current and creates the pulses for the UV LED.It also controls the shutter and the mirror K.
where I in0 (n) is the intensity after n passes through the cavity in the pure air filled cavity (without any absorbers) and I in (n) is the intensity after n passes through the cavity also including absorptions.n is the number of traverses of the photons through the cavity.However, according to Platt et al. (2009), for strong absorptions this is not equal to the optical density D(λ) used in traditional DOAS applications if using the average light path from Sect.2.1.Thus, the corrections described in Sect.2.3 have to be applied.

Determination of the average path length
A theoretical description of the average path length can be found in Platt et al. (2009).In practice, this average path length L 0 (λ) can be determined from the known Rayleigh scattering of two different gases, e.g., helium and dry zero air (Washenfelder et al., 2008), from a CRD decay time measurement (Meinen et al., 2010) or from the differential absorption of a trace gas of known concentration.In this work L 0 (λ) was determined from the absorption bands of O 4 at 343.5 and 360.4 nm, making use of the known O 2 concentration (Ball et al., 2004).The following subsections describe how this average path length has to be corrected.Research Laboratory, University of Bayreuth, Germany.One of three air-tight chambers made of either Teflon or glass is placed between the two mirrors.Because this chamber can be filled via the gas inlet with different trace gases, a purge flow of nitrogen in front of the mirrors avoids aerosols from the sample air to contaminate the mirrors.The distance between the mirrors is d 0 .The effective absorption path length is d because the sample gas only fills the volume between the ends of the mirror protection tubes between the purge gas outlets.

Correction of the base path length
The length of the absorption medium in the resonator d is not necessarily equal to the distance between the mirrors d 0 (see Fig. 2), since at either end of the resonator usually a few centimeters of the path are not filled with sample air due to a purge flow close to the mirrors.In this case, the effective light path has to be corrected by the geometric factor d/d 0 : For measurements of a small sample volume (small chamber), the dilution of the air due to the purge flow has also been taken into account.

Correction for light path reduction due to extinction in the cavity
Platt et al. ( 2009) described in detail how the light path is reduced by extinction in the resonator.Here two ways to correct this effect presented by Platt et al. (2009) are applied.A more detailed explanation of method 1 is described by Horbanski (2010) and Horbanski et al. (2014).For both ways the optical density D CE and the average path length L(λ) needs to be known.
1. Correction of the trace gas absorption cross section σ to σ eff : This σ eff is used by a DOAS-fit to obtain a corrected column density.From this column density the correct concentration is determined by dividing with L(λ).As the absolute optical density D CE is difficult to measure precisely, this method is used iteratively, where the optical density D i of the ith iteration is calculated.
During the first iteration a DOAS fit determines an optical density D 0 and σ 0 is corrected to σ 1 .In the next iteration step σ 1 determines a D 1 and σ 0 is corrected to σ 2 , etc.: ith iteration: We applied this correction in four iteration steps for measurements with only one trace gas and without aerosols.The same method can also be applied if several absorbers are present by extending Eq. ( 4).This method has the important advantage that it does not rely on the absolute optical density D CE and thus intensity I tot and is thus not sensitive to any possible intensity changes of, e.g., the LED between reference and sample measurement.
2. Method 1 with Eq. ( 3) becomes very complex when using more than one absorber and is not applicable if aerosols are present.In this case, a second method is useful in order to correct the optical density before the fit and convert it to D eff , which is equal to an absorption where the light path is not reduced: In this work, the second method applying Eq. ( 5) is used whenever there is more than one absorber.It needs to be considered that for this correction the intensities I 0tot and I tot have to be successively measured.This requires an absolutely stable optical resonator and light source intensity, similar to the requirements of the BBCEAS approach.

Correction of the wavelength dependence of the average light path
As described in Sect.2.1, the average light path L(λ) is also dependent on λ because R(λ) is dependent on λ.As described by Venables et al. (2006), it is possible to use differential absorption structures for the determination of the wavelength dependence of the mirror reflectivity and therefore for the calibration of the path length.This method requires an absorber of known concentration, which has several distinct absorption bands in the wavelength range of interest.However, the relative shape of the mirror reflectivity curve can be determined from any absorber having several absorption bands within the region of high mirror reflectivity.In the present study, we used O 3 and HONO to estimate the shape of the reflectivity curve.Since the concentration of these gases was unknown, the absorption bands of O 4 at 343.5 nm and 360.4 nm were additionally used to derive the absolute reflectivity curve of the mirrors.

Instrument
Figure 1 shows a schematic diagram of the CE-DOAS instrument (more details can be found in Hoch, 2010).The radiation of a pulsed UV LED is imaged by lens L 1 into the resonator, consisting of two highly reflective dielectric mirrors M 1 and M 2 (focal length = 1 m).The radiation, which leaves the resonator through M 2 , is focused by the lens L 2 into a fiber bundle which can be adjusted by a x-y translation stage.This fiber bundle transmits the light to a temperaturestabilized spectrograph (Sect.3.1.3).The movable mirror "K" can redirect the light to a photomultiplier (PMT) to record a ring-down signal and thus check the quality of the resonator.F 1 and F 2 are band-pass filters (for details see Sect.3.1.2).F 1 protects the PMT and F 2 shields the fibers from scattered light and LED light outside of the reflectivity range of the mirrors.The shutter is used to perform background measurements without radiation of the UV LED.The electronic controller receives a constant current and creates the pulses for the UV LED.It also controls the shutter and the mirror "K".

LED radiation source
The wavelength range (325 to 365 nm) of the LED UVTOP340 from Sensor Electronic Technology was chosen to optimize the sensitivity to BrO while minimizing the interference due to O 3 , but still being appropriate to detect O 3 .
In the order to minimize the known LED etalon structures, the surface-normal of the LED-chip is inclined against the optical axis of the resonator at an angle of 15 • (Sihler et al., 2009).For CRD measurements the LED has to be operated in a pulsed mode.Extensive noise characterizations of the LED showed that permanent pulsing of the LED creates less spectral noise than a periodic change between pulsing and continuous operation, while the spectrum of the LED does not change.Therefore the LED is permanently pulsed with 750 µs (on) and 50 µs (off) for also recording spectra with the spectrometer.The time required for the cavity to reach optical equilibrium is approximately 3.7 µs (for mirrors of reflectivity 0.9991 and 1 m cavity length), leading to a maximum light path of approximately 1111 m.This is small relative to the 750 µs when the LED is powered.The UVTOP340 delivers 400 µW of optical output power (SET, 2010).This figure appears incredibly small now compared to the optical output power of a 75 W xenon arc lamp as described by Kern et al. (2006).However, the relevant quantity needed to compare the optical output of different light sources is the spectral radiance integrated over the wavelength interval used to evaluate the trace gas absorption features.On the one hand, the radiance of the UVTOP340 between 325 and 365 nm calculates to 4.2 × 10 3 Wm −2 sr −1 , assuming a Lambertian emission pattern, an emitter area of 0.045 mm 2 as derived from microscopic measurements, and a Gaussian emission spectrum (340 nm peak, 15 nm FWHM).On the other hand, the emission spectrum of a xenon arc lamp (e.g., XBO 75 W/2 xenon arc lamp manufactured by OSRAM) is approximated by that of a Planckian gray body with an emissivity of 0.17 derived from the photometric properties given in Osram (2012).The radiance of the xenon arc lamp between 325 and 365 nm calculates to 1.9 × 10 6 Wm −2 sr −1 .Thus, the arc lamp has a radiance in this wavelength interval that is about 450 times higher compared to the UVTOP340.However, the optical output power in the 325 to 365 nm range per electrical input power for both devices is almost equal since the UVTOP340 consumes only 180 mW.

Mirrors and optics
The main component of the setup is the resonator, which consists of two highly reflective mirrors (Fig. 1, M 1 and M 2 ).The reflectivity of the mirrors from LAYERTEC GmbH, (Mellingen, Germany) was chosen to achieve a good compromise between optical light path length and signal-to-noise ratio (Fiedler et al., 2007).The radiation of the UV LED is coupled into the resonator by lens L 1 (F 1 = 25 mm) to use as much of the LED output as possible.The radiation that passes mirror M 2 is focused by lens L 2 (F 2 = 100 mm) on a fiber coupler to match the numerical aperture (NA 1 ) of the spectrometer.The fiber bundle (Loptek GmbH & Co, Berlin, Germany) consists of seven 200 µm fibers (NA of 0.22, UV solarization stabilized, 6 m long), arranged in a circular configuration on the resonators side and column configuration at the spectrometers side.The band-pass filters F 1 and F 2 , both UG11 from Schott, (Mainz, Germany), have a transmission range of 270-380 nm and a thickness of 1 mm.F 1 was placed directly in front of the PMT in order to protect the device as much as possible against ambient light (e.g., if cover is opened).The reflectivity of the mirrors cannot readily be derived from the manufacturer's data: Layertec provided transmission values (T ) but not reflectivities.Since R + T + A = 1 (A = absorption), R cannot be calculated from T unless A in the dielectric mirror is negligible, which, unfortunately, is not the case in this wavelength range.We used two sets of mirrors with different values of stated transmissivity at 340 nm: M 1 : T = 0.048 %, M 2 : T = 0.11 %.

Spectrograph and detector
The fiber transmits the light into a temperature stabilized ((20.0 • ± 0.1) C) spectrograph QE65000 from Ocean Optics (Dunedin, Florida, USA).The spectrometer has a focal length of 101 mm and is equipped with a 2400 lines mm −1 grating, resulting in 0.62 nm spectral resolution with the column of seven 200 µm fibers (see Sect. 3.1.2)acting as entrance slit.

Photomultiplier (PMT) for CRD
For the ring down measurement the radiation intensity is measured by a PMT (H578-01 from Hamamatsu Photonics (Naka-ku, Hamamatsu City, Shizuoka Pref., Japan)).As current amplifier a DLPCA-100 from FEMTO Messtechnik GmbH, (Berlin, Germany) is applied.A PC oscilloscope DAQSCOPE PCI-5102 from National Instruments (Austin, Texas, USA) records the ring-down signal.

Data recording and cavity ring-down calibration
The program DOASIS (Kraus, 2004) is used to control the measurements and to analyze the CE-DOAS and WS DOAS spectra.
To acquire the CRD signals, a dedicated program was developed with LabVIEW from National Instruments.This program receives the ring-down signal from the PC oscilloscope, calculates the mean of a given number of measurements and fits a double exponential decay f (t) (Eq.6): τ 1 denotes the electro-optical decay time from electronics and LED and τ 2 for the CRD decay time.A typical value for τ 1 is 100 ns and it is determined by the rise time of the pulse electronics of the LED and the rise time of the LED output itself.Also, A denotes the initial signal at t = 0 and b denotes any offset signal at t = ∞.From this fit and known τ 1 , the CRD decay time can be determined and converted to an optical light path or mirrors reflectivity according to Meinen et al. (2010).However, in our case it is not possible to derive with this method the accurate absorption light path for the data evaluation, as the mirror reflectivity varies too much over the LED spectral range (see Sect. 2.4).Nevertheless, τ 2 just gives an indication of the wavelength averaged light path, which is used to monitor the contamination of the mirrors as well as the optical alignment of the resonator.This allows quick action to optimize instrument performance.In fact, during chamber studies the UV CE-DOAS system performed with excellent stability during a time period of several days without mirror cleaning and readjustment.

The measurement setups
The instrument was developed and characterized at the Institute for Environmental Physics (IUP) in Heidelberg, Germany.After the first successful measurements, it was applied for the observation of different trace gases at the Atmospheric Chemistry Research Laboratory, University of Bayreuth, Germany.For these measurements the instrument was used as it is described in Sect.3.1 and shown in Fig. 1.Two hermetically sealed plastic boxes were placed around the optical component assemblies in order to protect the sensitive optics against dust and other extreme conditions for planned measurements in a climate chamber.Furthermore, the hermetically sealed containment guarantees a stable nitrogen purge-flow (1 L min −1 ) of the resonator mirrors to prevent contaminations from the measurement air.Otherwise a slight overpressure in the reaction chamber could cause a flow of unfiltered air from the chamber towards the mirrors and thus cause contaminations.For the measurements described below, an air-tight chamber was placed between the two boxes containing the mirrors and associated optics (Fig. 2).This chamber could be filled through three gas inlets/outlets with different gases.Three different setups with various base-path lengths were used: one with a glass chamber (setup A) and two with Teflon chambers (setups B and C).The glass chamber setup was the first to be used because it could be built easier than Teflon, but the cavity length is much shorter.The Teflon/glass chambers A, B and C were kept in pressure equilibrium with the laboratory with the help of two open outlets, as indicated in Fig. 2. The mirror purge flow compensates for any leakage of the chamber.Table 1 shows the sizes of these three chambers and Table 2 gives an overview of the different trace gas measurement setups.With these setups, different trace gas measurements were carried out: O 3 and HONO were measured mainly in order to determine the wavelength dependency of the mirror reflectivity and the optical path length, O 4 measurements were performed in order to determine the absolute optical path length at two particular wavelengths (343.4 and 360.5 nm), and BrO and HCHO measurements were made in order to determine the detection limits for these species.Before each trace gas measurement, offset (20 000 scans and 8 ms integration time) and dark current (1 scan and 30 000 ms integration time) spectra were taken with blocked LED.If any background radiation intensity would be present, it would be included in the dark current spectrum.In the next step a CRD measurement with 1000 average determinations was performed to determine possible mirror contamination.A spectrum with 85 % of full saturation could be obtained in 9.50 s with mirror set M 1 and 3.50 s with mirror set M 2 .Based on noise tests, 30 spectral scans were taken and co-added for mirror set M 1 and 50 spectral scans were taken and co-added for mirror set M 2 to achieve sufficient measurement accuracy.This corresponds to a temporal resolution of 4.8 min for M 1 and 2.9 min for M 2 .The measurements shown within this study were undertaken without the presence of aerosols.

Ozone measurements
For the O 3 measurements the setups A, B, and C were used.
For setup A both mirror sets, for setup B mirror set M 2 , and for setup C mirror set M 1 were used (Table 2).To determine an I 0 spectrum the chamber was filled with nitrogen and the mirrors were purged with nitrogen, too.Afterwards O 3 was produced and filled into the chamber through the gas inlet (up to 2 ppm).The gas outlets were connected to an active charcoal trap.In setup A and B the ozone was created by ionization of O 2 (purity = 99.995%) using a corona discharge ozonizer and in setup C by passing O 2 (purity = 99.995%) along four Pen-Ray mercury discharge lamps.

Nitrous acid measurements
For the nitrous acid (HONO) measurements, setup A with both mirror sets, setup B with mirror set M 2 , and setup C with mirror set M 1 were used (see Table 2), similar to the setups for O 3 measurements.The generation of the nitrous acid is based on the reaction of a sodium nitrite solution NaNO 2 with sulfuric acid H 2 SO 4 as, e.g., described by Taira and Kanda (1990): The reaction takes place in a glass vessel with a glass frit at its base.The diluted solutions of sodium nitrite and sulfuric acid are continuously pumped into a reaction vessel and the mixture is then drawn off to waste.A carrier gas, here nitrogen, is introduced into the reagent mixture through the frit and continuously purges nitrous acid vapor from the mixture.The HONO generation depends on temperature, gas flow, NaNO 2 concentration and fluid flow.Typical mixing ratios up to 30 ppb were accomplished.

Oxygen dimer measurements
In order to determine the light path in the cavity, O 4 measurements were carried out at two particular wavelengths (343.4 and 360.5 nm).For this purpose the resonator was first completely filled with nitrogen and afterwards completely (including mirror purge gas) with pure oxygen.Additionally, barometric pressure (with a mercury barometer) and temperature were measured next to the chamber (the same values assumed as in the chamber) to determine the O 2 concentration in the resonator.With the absorption of O 4 at 343.4 and 360.5 nm, the light path at these wavelengths could be determined.

Bromine monoxide measurements
BrO measurements were carried out using setup B with mirror set M 2 and in setup C with mirror set M 1 (see Table 2).For these measurements the mirror protection tubes were purged with nitrogen.The gas outlets are connected to a pump producing a slight underpressure.First O 3 was admitted through the gas inlet and afterwards Br 2 .A fluorescent tube (Philips TL/12, 40 W, UV A and UV B) set up next (at 20 cm distance) to the chamber was switched on to photolyze Br 2 .A typical concentration up to 1000 ppt BrO was measured.Evaluation and results of these measurements are described in Sect. 4.

Comparison of the CE-DOAS with a White system DOAS and an ozone monitor
During experiments with a simulated salt pan (Buxmann et al., 2012) at the Atmospheric Chemistry Research Laboratory, University of Bayreuth, Germany, in the lowtemperature aerosol simulation chamber (LOTASC) (http: //www.eurochamp.org/chambers/lotasc/), a comparison between the CE-DOAS and a DOAS WS (White, 1942(White, , 1976;;Ritz et al., 1993;Volkamer, 1996;Volkamer et al., 2002;Hak et al., 2005) as well as the CE-DOAS and a chemiluminescence (CL) ozone monitor (Bendix-UPK8002) was performed.The cylindrical chamber, made of Teflon foil (FEP 200 A, DuPont, thickness 0.05 mm) with a diameter of 1.33 m and a volume of 3500 L, has been described in detail by Siekmann (2008); Bleicher (2012); Buxmann et al. (2012).The LOTASC was kept at a pressure slightly above atmospheric pressure in order to prevent leaking of room air into the system.The dilution rate, replenishing the loss by sampling and other outlets, was of the order of 3 × 10 −5 s −1 , ).An exponential fit to the concentration decrease directly yields the dilution rate (Buxmann et al., 2012;Bleicher, 2012).
During the intercomparison experiments the WS had a base path of 2 m and an aperture of F/25, with a total light path of 288 m diagonally arranged through the chamber, whereas the CE-DOAS light path was horizontally arranged with a mirror distance of 1.33 m.In order to determine the absolute light path in the cavity, O 4 measurements were carried out, as described above.For that purpose an additional tube was mounted in between the resonator inside the smog chamber.The maximal path length at 346.3 nm was 816 m.A sketch of the experiment setup is shown in Fig. 9.These experiments were described in detail by Buxmann (2012) and the results are described in Sect.4.3.

Mirror reflectivity and path length calibration
As described in Sect.2.4, it is possible to use differential absorption structures to derive the wavelength dependence of the mirror reflectivity and therefore the determination of the wavelength-dependent path length.We used O 3 and HONO to estimate the relative shape of the mirror reflectivity, not the absolute, as the actual concentrations of O 3 and HONO were unknown.The used fit windows are shown in Table 4.The relative shape was derived using a second-order polynomial fit, and the sensitivity of the O 3 and HONO measurements result in a maximum uncertainty of the relative shape of the reflectivity within the fitting range of 0.05 %.With the additional absorption bands of O 4 at 343.4 and 360.5 nm, the absolute reflectivity curve  Stutz and Platt (1996) to 2 %.Thus the total error of the absolute path length is estimated at 2 %.
The following measurements were corrected with those reflectivity curves.The absorption A of the mirrors can be derived from the data in Fig. 3 as The absorptions of both mirror sets are of the order of A = 4 × 10 −4 .According to Fiedler et al. (2007), the relation between optical light path length and signal-to-noise ratio Q can be determined with Figure 4 shows the enhancement factor (quality factor) Q dependency on (1 − R) assuming A = 4 × 10 −4 .Based on this theory, we expected a lower detection limit for a reflectivity of R 2 = 0.9984 of M 2 than for M 1 (R 1 = 0.9991), if the measurement accuracy is purely dominated by photon noise.

Detection limits
The true 1σ error of the measurement of the different trace gases is obtained from the statistical 1σ error of the DOAS fit by multiplying with a factor of about 2.5 (depending on the ratio between width of the spectral features and the spectral resolution, etc.), as the fit error underestimates the true measurement error (Stutz and Platt, 1996), depending on the number of absorption bands, the number of channels used for evaluation, the intensity of the recorded spectrum, uncertainties in the wavelength-pixel mapping, and the scale and structures in the residual spectrum.The values are derived for measurements with relatively high trace gas concentration.Thus, measurement error and detection limit may actually be lower for lower concentrations due to weaker structures in the residual spectrum.The relative error decreases with 1 E -3 0 .0 1 0 . 1 1 1 0 L = 8 E -5 ( e q u a t e s t o R a y l e i g h s c a t t e r i n g a t 3 4 0 n m ) Fig. 4. Calculation of the enhancement factor (quality factor) Q from Eq. ( 8) (Fiedler et al., 2007) as a function of (1 − R) for different losses L and a mirror absorption A = 4 × 10 −4 .The red line indicates the theoretical upper limit for Q for the minimal losses due to Rayleigh scattering in the resonator at 340 nm.
-2 0 -1 0 0 1 0 higher concentrations.But at higher concentrations larger residual structures arise due to imperfect descriptions of the absorption bands.Thus, the absolute error and detection limit increase.The detection limit is given as twice the 1σ measurement error.
Example fits of different trace gases are shown in Figs. 5, 6, 7 and 8.The fit result in Fig. 5 shows an example where ozone and BrO were detected simultaneously, whereas only one absorber was present in the other examples    absorbers were evaluated, the first example (Fig. 5) shows an approximately twice as large residual.This is typical for DOAS measurements.However, simultaneous detection and identification of the different absorbers is possible with sufficient sensitivity.The larger measurement error in these cases is considered in all data reported in this paper.
In Table 5 the detection limits for the measured trace gases for the two different mirror sets and different integration times are given.An estimated detection limit for NO 2 of 3.6 ppb at an integration time of 3 min is given as well.The differential NO 2 absorption (3 × 10 −20 cm 2 molec −1 ) in the wavelength range 325 nm-360 nm is relatively weak compared to BrO × −17 cm 2 molec −1 ), but concentrations could differ by 3 orders of magnitude in nature and thus be of the same magnitude.Nevertheless, other studies have also shown that NO 2 and BrO can be detected simultaneously and reliably by DOAS (Platt and Stutz, 2008;Mahajan et al., 2009;Pöhler et al., 2010).Based on the theory of Fiedler et al. (2007), we expected a better detection limit for the lower reflectivity of R 2 = 0.99840 ± 0.00005 for M 2 than for M 1 (R 1 = 0.99910 ± 0.00009).However, we observed a better detection limit for M 1 compared to M 2 .The reason is that in practice the detection limit of a DOAS measurement is not only dependent on the photon noise but also on spectral structures, which are not constant in time.Other sources increasing the noise are contributions from electronic noise and optical noise, e.g., additional noise introduced by the LED light source, optical fibers and spatial inhomogeneity of the optical grating of the spectrograph.

Instrument comparison
The BrO correlation plot (Fig. 10) shows very good agreement between the CE-DOAS and the WS-DOAS systems for mixing ratios from below the respective detection limits up to 400 ppt.The slope of the BrO mixing ratios measured by each instrument is 1.06 ± 0.07, using a bivariate weighted fit as suggested by Cantrell (2008) to obtain a correlation coefficient of r 2 = 0.95.The agreement of HCHO of the two multi-reflection systems is remarkably good as well, with r 2 = 0.998 and a slope of 0.97 ± 0.03 for values up to 1 ppb, using again the bivariate fit (Fig. 11).The correlation plot for ozone furthermore shows (Fig. 12) a very good agreement between the CE-DOAS and the CL ozone monitor with a slope of 1.11 ± 0.08 and r 2 = 0.92, even though the CE-DOAS measurement spectral range was not optimized for ozone measurements.The excellent agreement between the two multi-reflection systems (WS and CE-DOAS), measuring at different light paths, demonstrates the accuracy and reliability of the newly developed CE-DOAS instrument.Integrity of operation is evident by the good agreement with the ozone monitor, which takes in air from a single location close to the chamber wall.This intercomparison of the instruments also demonstrates the rapid mixing of the chamber air within a timescale of 120 s (average time resolution of the instruments).A well mixed chamber is crucial for an understanding of the mechanistic of halogen release processes using chamber studies.A time series of a bromine explosion event with BrO mixing ratios of up to 360 ± 10 ppt is shown in Fig. 13.In the presence of high ozone mixing ratios of up to 350 ppb within the initial 30 min of the experiments, before the light was switched on, the BrO mixing ratio measured by the two DOAS instruments drops to slightly negative values (around −25 ppt on average).This negative offset is probably due to interference effects, which might not be captured by the DOAS evaluation, as well as an imperfect ozone literature reference spectrum.However, the values of BrO are not significant within the 2σ measurement error of 54 ppt.The heterogeneous formation of bromine caused a total ozone depletion of 320 ppb within 150 min.The very good detection limit of the CE-DOAS compared to the White system will allow performance of chamber studies close to natural BrO levels, which are typically in the range of several ppt up to more than 200 ppt (Saiz-Lopez and von Glasow, 2012).

Conclusions
We developed the first LED-based CE-DOAS for detection of BrO in the UV wavelength range from 325 to 365 nm.Due to the low power consumption of the light source, the instrument can be applied to field measurements with limited power supply.The achieved BrO detection limit is 11 ppt using a time resolution of 81 min per measurement.Besides   losses or reactions on inlets, filters or walls.The detection limits are more than adequate for simulation chamber measurements, where BrO levels up to 6000 ppt were observed (Buxmann et al., 2012;Ofner et al., 2012).Additionally, the instrument can improve BrO measurements of chamber experiments due to its better accuracy than state-of-theart DOAS White systems, which have BrO detection limits around 50 ppt for a time resolution of 4 min (Buxmann et al., 2012).The relative dependence of the path length on the wavelength was determined using the absorption structures of O 3 and HONO.With an additional O 4 measurement at a known O 2 concentration, we calculated the absolute optical path length.Combining these data with the relative path length, the absolute wavelength-dependent mirror reflectivity and path length over the whole wavelength range was derived.We used two sets of mirrors with different maximal reflectivity values at 347 nm, M 1 with R 1 = 0.99910 ± 0.00009 (T = 0.05 %) and M 2 with R 2 = 0.99840 ± 0.00005 (T = 0.12 %).We determined the absorption for both mirror sets to be in the range of 4 × 10 −4 at 347 nm.The absorption increases towards shorter wavelengths, so that the peak of the reflectivity is shifted towards larger wavelengths compared to the transmission of the mirrors.
In contrast to expectations based on the theory of Fiedler et al. (2007), we found a better detection limit for the higher reflectivity of M 1 than of M 2 .The reason is that the detection limit of a DOAS measurement is not only dependent on the photon noise but also on spectral structures, which are not constant in time.Other sources for increasing noise are electronic noise and optical noise originating, e.g., from the LED light source, optical fibers and spatial inhomogeneity of the optical grating of the spectrograph.
Several options are possible for future improvement of the system: mirrors with higher reflectivity would lead to lower detection limits if the absorption of the mirrors could be reduced (e.g., by using different materials during the production process, which is currently under development by different companies).With the current absorptions, a higher reflectivity will not result in much lower detection limits.The base path length could also be expanded, but at the expense of compactness and ease of handling of the instrument setup.For most measurements this is not applicable.
Another aim is to increase the light intensity for a better time resolution and therefore improved detection limit.The intensity of UV LEDs is still very poor, but its improvement is a topic of current research at all LED manufacturers.
Cooling of the LED also looks promising, since cooling of 30 K down to −10 • C was found to increase the radiation intensity up to 50 % (for an LED operation current of 1 mA), as expected from the data sheet (SET, 2010).While this work demonstrates that CE-DOAS measurements are feasible, applying UV LEDs as a light source, it is noted that xenon arc lamps may provide a 450-times higher input radiance.When compared to the electrical input, however, both devices deliver a similar radiance per input power.Therefore, UV LEDs may present a competitive alternative light source to xenon arc lamps, depending on the application of the respective instrument.
Furthermore, the system could be improved with a spectrometer with higher light throughput and larger entrance slit area to use more of the radiation exiting the cavity.Finally, we expect improvement of the detection limit from reduced optical noise.Our new UV LED CE-DOAS is a prototype instrument.Applying the above improvements and assuming that UV LED development will continue in the near future, we expect that accuracy and detection limit of the instrument can be improved by an order of magnitude during the next few years.
For validation an intercomparison between the CE-DOAS and a WS DOAS as well as between the CE-DOAS and a chemiluminescence (CL) ozone monitor was performed.Excellent agreement between the two multi-reflection systems (WS and CE-DOAS), measuring at different light paths and good agreement with the CL ozone monitor was achieved.This demonstrates the reliability of the newly designed CE-DOAS instrument.
Due to its compact size and low power consumption, it is possible to apply the BrO CE-DOAS instrument -besides in reaction chamber investigations -also in field studies with batteries even if other electrical power is not available.To our knowledge this instrument is the first mobile cavity-based BrO instrument in the UV wavelength range that is applicable to field measurements.Thus, in situ measurements with UV LEDs, especially of BrO, are possible in the future.As mentioned above, two other CE instruments exist for the detection of BrO (Chen and Venables, 2011;Grilli et al., 2012), but their field application seems to be limited due to size and power consumption.
Furthermore, BrO instruments with inlets, like CIMS (e.g., Liao et al., 2011) and mode-locked frequency-doubled CE spectrometer (Grilli et al., 2012) may suffer from the problem of inlet losses.Our novel UV LED CE-DOAS instrument provides a very good alternative, as it is an open path instrument without any inlets, and self-calibrating DOAS retrieval is used.We would like to stress that there are several areas where BrO has been detected in the atmosphere at levels exceeding the detection limit of the presented version of our CE-DOAS instrument, e.g., in arctic regions 41 ppt (Pöhler et al., 2010), at the Salar de Uyuni, Bolivia, up to 20 ppt (Hönninger et al., 2004), at the Dead Sea 200 ppt (Hebestreit et al., 1999;Matveev et al., 2001;Tas et al., 2008), and in volcanic plumes where BrO mixing ratios of the order of 1000 ppt were inferred (Bobrowski et al., 2003).All these observations are averaged concentrations over few km-long light path lengths.Thus, the BrO CE-DOAS instrument could help to study the involved processes by deriving in situ BrO concentrations which are probably even higher.Moreover, it appears likely that the detection limits of our instrument will be greatly improved in the future.

Fig. 1 .
Fig. 1.Schematic diagram of the CE-DOAS instrument.The radiation output of a pulsed UV LED is imaged by lens L 1 into the resonator consisting of two highly reflective dielectric mirrors M 1 and M 2 .The light which leaves the resonator through M 2 is focused by the lens L 2 to a fiber bundle attached to an x-y stage.This fiber bundle transmits the light to a temperature stabilized spectrograph (Ocean Optics QE 65000).Alternatively, the movable mirror "K" can redirect the light to a photomultiplier (PMT) to optionally record a ring-down signal and thus check the path length of the resonator.F 1 and F 2 are band-pass filters (Schott UG11).F 1 protects the PMT and F 2 the fibers from scattered light and LED radiation outside of the reflectivity range of the mirrors.The shutter is used to perform background measurements.The electronic controller receives a constant current and creates the pulses for the UV LED.It also controls the shutter and the mirror K.

Fig. 2 .
Fig. 2. Setup for the measurements at the Atmospheric ChemistryResearch Laboratory, University of Bayreuth, Germany.One of three air-tight chambers made of either Teflon or glass is placed between the two mirrors.Because this chamber can be filled via the gas inlet with different trace gases, a purge flow of nitrogen in front of the mirrors avoids aerosols from the sample air to contaminate the mirrors.The distance between the mirrors is d 0 .The effective absorption path length is d because the sample gas only fills the volume between the ends of the mirror protection tubes between the purge gas outlets.

Fig. 3 .
Fig.3.Reflectivity curves of mirror sets M 1 and M 2 : the black squares indicate the (1-transmission)-curve from the manufacturer's data sheet.The red and blue lines are the reflectivities calculated from measurements with setup A and C, respectively.The dashed lines are the accuracies of this determination, which are maximal 0.05 % within the second order polynomial fit range.The maximal reflectivities are R 1 = 0.99910 ± 0.00009 for M 1 and R 2 = 0.99840 ± 0.00005 for M 2 at 347 nm.The green lines between the black dots and the reflectivity curves indicate the losses due to absorption in the reflective layer.It can be seen that the absorption is small at 370 nm and increases towards shorter wavelengths.

Fig. 5 .
Fig. 5. Example of the spectral identification of BrO with the CE-DOAS setup.The thick red line indicates the fit result and the thin blue line the sum of the fit result and the residual.The corresponding mixing ratio for BrO is (507 ± 11) ppt and for O 3 (1984 ± 51) ppb.The residual is 4 × 10 −3 peak to peak.Derived detection limits are based on measurements of gas mixtures (cf.Sect.4.2).

Fig. 6 .
Fig. 6.Spectral DOAS fit of HCHO with (88 ± 6) ppb.Upper panel: the thick red line indicates the fit result and the thin blue line the sum of the fit result and the residual.Lower panel: fit residual with peak to peak 3 × 10 −3 .

Fig. 7 .
Fig. 7. Spectral DOAS fit of HONO with (16.0 ± 0.7) ppb.Upper panel: the thick red line indicates the fit result and the blue thin line the sum of the fit result and the residual.Lower panel: fit residual with peak to peak 4 × 10 −3 .

Fig. 8 .
Fig. 8. Spectral DOAS fit of O 3 with (1766 ± 59) ppb.Upper panel: the thick red line indicates the fit result and the thin blue line the sum of the fit result and the residual.Lower panel: fit residual with peak to peak 2 × 10 −3 .

Fig. 9 .
Fig.9.Sketch of the multi-reflection cells used for DOAS in the Teflon chamber.A White system with a base path of 2 m was installed on a diagonal frame.The mirrors and prisms of the White system were mounted inside the chamber with additional Teflon bags.A maximum total light path of 320 m was achieved.The transfer optics adapts the opening aperture of the White system to the fiber and spectrograph.The CE-DOAS is mounted on a different frame, with a 1.3 m base path, resulting in an maximal light path of 816 m within these measurements (depending on the mirror reflectivity, here mirror set M 2 was used).The mirrors of the BrO cavity were constantly purged with clean air.

Fig. 10 .
Fig.10.Intercomparison of BrO detected by the White system and CE-DOAS simultaneously during a smog chamber experiment above a simulated salt pan, without injection of aerosols.The slope of the BrO mixing ratios measured by each instrument is 1.06 ± 0.07 using a bivariate weighted fit.The correlation coefficient is r 2 = 0.95.

Fig. 12 .
Fig. 12. Intercomparison of O 3 detected by the ozone monitor and CE-DOAS simultaneously during a smog chamber experiment above a simulated salt pan, without injection of aerosols.The slope of the O 3 mixing ratios measured by each instrument is 1.11 ± 0.08 using a bivariate weighted fit.The correlation coefficient is r 2 = 0.92.

Fig. 13 .
Fig. 13.Intercomparison of a time series of BrO (top panel) detected by the WS (black squares) and CE-DOAS (red open cycles) and O 3 (bottom panel) detected by the CE-DOAS (red open cycles)and O 3 monitor (blue line) simultaneously during a smog chamber experiment above a simulated salt pan.BrO is formed under irradiation (360 ppt within 17 min) and decreases to less than 30 ppt after 2 h.A corresponding total depletion of 320 ppb O 3 after 150 min is observed during the irradiation by the solar simulator.

Table 1 .
Chamber sizes and base-path lengths for setups A, B and C.

Table 2 .
Measurement setups for detection of different trace gases.

Table 3 .
Literature absorption cross sections used in the DOAS evaluation.The cross sections were corrected with a Pre-shift and linked together for the following measurements.

Table 4 .
DOAS fit windows of O 3 and HONO for the determination of the relative wavelength-dependent path length.

Table 5 .
Mean detection limits, including the presence of high ozone concentrations for different measurement configurations.Mirror set M 1 with R 1 = 0.99910 ± 0.00009 and M 2 with R 2 = 0.99840 ± 0.00005.The path length L eff (at 347 nm) is for setup B = 839 m and for setup C = 1472 m, with an accuracy of 2 %.An estimate for the NO 2 detecion limit is given.Derived detection limits are based on measurements of gas mixtures (cf.Sect.4.2).
Intercomparison of formaldehyde detected by the White system and CE-DOAS simultaneously during a smog chamber experiment above a simulated salt pan, without injection of aerosols.
multaneously and with little interference, although it is limited to areas with high concentrations and laboratory studies at the current state of development.Additionally the instrument measures the trace gases in comparison to most other in-situ instruments directly in the atmosphere minimizing