Recent improvements of Long-Path DOAS measurements: impact on accuracy and stability of short-term and automated long-term observations

Over the last decades, Differential Optical Absorption Spectroscopy (DOAS) has been used as a common technique to simultaneously measure abundances of a variety of atmospheric trace gases. Exploiting the unique differential absorption cross section of trace gas molecules, mixing ratios can be derived by measuring the optical density along a defined light path and by applying the Beer-Lambert law. Active long-path (LP-DOAS) instruments can detect trace gases along a light path of a few hundred metres up to 20 km with sensitivities for mixing ratios down to ppbv and pptv levels, depending on the trace 5 gas species. To achieve high measurement accuracy and low detection limits, it is crucial to reduce instrumental artefacts that lead to systematic structures in the residual spectra of the analysis. Spectral residual structures can be introduced by most components of a LP-DOAS measurement system, namely by the light source, in the transmission of the measurement signal between the system components or at the level of spectrometer and detector. This article focuses on recent improvements by the first application of a new type of light source and consequent changes to the optical setup to improve measurement accuracy. 10 Most state-of-the-art LP-DOAS instruments are based on fibre optics and use xenon arc lamps or light emitting diodes (LEDs) as light sources. Here we present the application of a Laser Driven Light Source (LDLS), which significantly improves the measurement quality compared to conventional light sources. In addition the lifetime of LDLS is about an order of magnitude higher than of typical Xe-arc lamps. The small and very stable plasma discharge spot of the LDLS allows the application of a modified fibre configuration. This enables a better light coupling with higher light throughput, higher transmission homo15 geneity, and a better suppression of light from disturbing wavelength regions. Furthermore, the mode mixing properties of the optical fibre are enhanced by an improved mechanical treatment. The combined effects lead to spectral residual structures in the range of 5− 10 · 10−5 RMS (in units of optical density). This represents a reduction of detection limits of typical trace gas species by a factor of 3-4 compared to previous setups. High temporal stability and reduced operational complexity of this new setup allow the operation of low-maintenance automated LP-DOAS systems as demonstrated here by more than two years of 20 continuous observations in Antarctica.


10
Most state-of-the-art LP-DOAS instruments are based on fibre optics and use xenon arc lamps or light emitting diodes (LEDs) as light sources. Here we present the application of a Laser Driven Light Source (LDLS), which significantly improves the measurement quality compared to conventional light sources. In addition the lifetime of LDLS is about an order of magnitude higher than of typical Xe-arc lamps. The small and very stable plasma discharge spot of the LDLS allows the application of a modified fibre configuration. This enables a better light coupling with higher light throughput, higher transmission homo- 15 geneity, and a better suppression of light from disturbing wavelength regions. Furthermore, the mode mixing properties of the optical fibre are enhanced by an improved mechanical treatment. The combined effects lead to spectral residual structures in the range of 5 − 10 · 10 −5 RMS (in units of optical density). This represents a reduction of detection limits of typical trace gas species by a factor of 3-4 compared to previous setups. High temporal stability and reduced operational complexity of this new setup allow the operation of low-maintenance automated LP-DOAS systems as demonstrated here by more than two years of 20 continuous observations in Antarctica.
The main advantage of DOAS in atmospheric remote sensing is that it allows the contact-free and simultaneous measurement of several trace gases. Exploiting that many molecules have unique differential absorption cross sections, mixing ratios can be derived by measuring the optical density of long light paths in the atmosphere using the DOAS principle (see e.g. Platt and Stutz (2008) for a detailed introduction).

10
In contrast to passive instruments (e.g. Multi Axis (MAX)-DOAS or satellite instruments), which use scattered or reflected light and hence rely on natural light sources such as solar (or lunar Wagner et al., 2000) radiation, active DOAS instruments use artificial light sources such as LEDs or arc lamps. The independence from natural light sources allows continuous observations of trace gases to e.g. study night-15 time chemistry. It also enables investigations of trace gases absorbing in the deep UV where no natural light sources exist. Another advantage is the well-defined light path of up to 20 km. Along this light path, a mean mixing ratio is determined. In comparison to passive instruments, this reduces the analytical effort to obtain mixing ratios and usually leads to smaller uncertainties as no radiative transport models are needed for the interpretation of the data. Furthermore, compared to point measurements, Long-Path 20 DOAS results are less sensitive to large spatial gradients yielding concentrations with a better representativeness for comparison with chemsitry models or typical footprints of air borne platforms and satellites.
Most modern LP-DOAS setups use fibre optics for light transfer between light source, telescope and spectrometer (Merten et al., 2011) and a mono-static telescope, i.e. one telescope is used for both sending 25 and receiving the light reflected from a retro-reflector array. In the following, recent improvements to this setup will be presented which, in combination, can increase accuracy and ::::::::: precision :::: and : hence reduce detection limits of LP-DOAS measurements by a factor of 3 to 4 compared to previous setups. To achieve this, a novel light source type was applied and the light coupling from the light source to the telescope was optimised to reduce stray light. Furthermore, a new configuration of the optical fibres with an improved 30 mode mixing was introduced. In addition to an enhanced measurement performance, these improvements have made the previously quite cumbersome setup of LP-DOAS instruments considerably easier and now allow the operation of low-maintenance, automated instruments for long-term observations.
In section 2 the state of the art in LP-DOAS instrument design will be described. Improvements of measurement performance and operation procedure following the introduction of the novel light source 5 and changes to the fibre configuration are presented in section 3. In section 4 the influence of residual structures due to fibre modes and a new method for mode mixing to reduce these structures is presented.
In section 5 the combined contribution to the improved instrument performance with respect to reduced stray light and reduction of total noise is quantified based both on lab measurements and field campaigns and typical detection limits for setups that incorporate the improvements are presented.
10 2 Long-path DOAS LP-DOAS instruments couple light from an artificial light source into a telescope which creates a light beam that is transmitted through the atmosphere across a distance ranging from a couple of hundred metres to several kilometres. At the end of this atmospheric path, the light is collected by a telescope 15 and analysed for spectral absorption structures -typically with a grating spectrometer. This originally bi-static setup with separate telescopes for sending and receiving was replaced by a mono-static setup with a single telescope and a retro-reflector array introduced by Axelsson et al. (1990). After reflection at the retro-reflector, the light is received again by the same telescope which reduces the complexity of the setup with regard to power supply and alignment. It also doubles the length of the light path. State of the 20 art LP-DOAS instruments mostly rely on fibre optics for light coupling between light source, telescope and spectrometer (see Fig. 1). Compared to traditional systems that use a complex system of mirrors for the light coupling between light source, telescope and spectrometer, this approach, first introduced by Merten et al. (2011), further reduces the complexity of alignment of the telescope itself and increases the transmittance compared to the coaxial Newton-type telescopes used with the mirror coupling. The crucial components in a modern fibre-based LP-DOAS setup are the light source and a Y-shaped optical fibre bundle where one end serves as sending fibre bundle that guides the light from the light source to the telescope and the other end serves as a receiving fibre bundle, leading from the telescope to the spectrometer (see Fig. 1). According to Merten et al. (2011), the fibre bundle in such a "classical" setup 5 (see upper row in Fig. 2 for a detailed schematic of the sections of the bundle) on the transmitting end typically consists of a mono fibre with a large diameter (typically 800 µm) to maximize light collection at the light source (at letter A in Fig. 1 and also column A, upper row in Fig. 2). This fibre is then coupled to a ring of smaller diameter (200 µm) fibres (at letter B/column B) leading to the combined end of the bundle.
The monofibre (800 µm) is required to guarantee an equal illumination of all small diameter fibres of the 10 ring. Then the end of the bundle (letter C/column C) is placed close to the focal point of the telescope mirror to create a parallelized light beam. For a fibre at the focal point of a parabolic mirror and omitting beam widening effects, the light emitted by the sending fibre bundle would be imaged on itself, so that no light would reach the receiving fibre. However, there are a number of effects that blur the reflected image of the light source and lead to a coupling of light into the central receiving fibre: (a) comatic aberration when the incident beam is parallel but not paraxial, (b) diffraction at the apertures of telescope and retroreflectors (c) surface irregularities of mirror and retro-reflectors, (d) defocussing of the fibre bundle, (e) atmospheric turbulence, and (f) for spherical main mirrors the spherical aberration in combination with the lateral offset of the beam at the retro-reflectors (Rityn, 1967;Eckhardt, 1971;Merten et al., 2011). Merten et al. (2011 have determined (a)-(c) to have a negligible influence for components typically used in LP-20 DOAS systems. Considering (d) and (e) (and (f) if a spherical main mirror is used), the light throughput of a fibre based system is optimised by setting the end of the fibre bundle to a slightly out of focus position in front of the main mirror.
To homogenize the illumination of the entrance slit of the spectrometer and hence the grating, different mode mixing techniques (see e.g. Stutz and Platt (1997)) can optionally be applied between telescope and 25 spectrometer (at D in Fig. 1/column D upper row in Fig. 2) before the light is coupled into the spectrometer passing an (optional) optical slit (letter E/column E) (see section 4).
For the analysis of the atmospheric absorption, a reference spectrum without absorption of the gases is required. This is obtained by temporarily inserting a diffuse reflector, e.g. a sandblasted white surface, close (i.e. around 1-4 mm distance) to the end of the fibre at (C) thus creating an optical "shortcut" (SC) for the light (for sketch of this mechanism see Fig. A1 in the appendix). In the following, spectra recorded this way will be referred to as reference. The reflector mechanism will be referred to as shortcut.
To account for scattered sun light from the atmosphere in both atmospheric and reference spectra as well as to correct for the CCD's dark current and offset signal, background spectra for both types of 5 measurement spectra are recorded on a regular basis by shutting off the light source at (A). All four spectrum types are recorded in an interleaved fashion with typically a couple of pairs of reference and atmospheric spectra followed by one atmospheric background and one reference background. Examples of such measurement routines are discussed in detail in section ?? below. :::: Sec. :: C ::: in ::: the :::::::::: appendix. :

LP-DOAS setups used in this study
We tested several improvements to the classic fibre based DOAS setup (as described in Sec. 2.1), an overview of the three different set-ups is given in Tab. 1 and will be described in the following sections.

5
The systematic comparison of improvements to the setup was done in the rooftop laboratory of the Institute of Environmental Physics at the University of Heidelberg (setup HD) with an Acton 500i spectrometer and a smaller laboratory telescope that allows quick changes of components but is not suited for outdoor deployment. For atmospheric measurements a 1.55 km light path (one way) passing over a residential area of Heidelberg to another institute was used. Tests with atmospheric measurements were performed during 10 6 weeks from March 11 until May 3, 2014. In each configuration, measurements were performed for at least 24 h to ensure sufficient statistics and the comparability of different setups.
As for all measurements not performed under fully controlled laboratory conditions, the influence of environmental parameters has to be considered in such a comparison. An important factor in LP-DOAS measurements are variations of the telescope-reflector alignment which can be influenced by changes to 15 the setup as well as environmental parameters such as air temperatures. To ensure an optimal alignment, as part of the measurement routine and in alternation with measurement periods, an optimisation of the received signal is performed on a regular basis by systematically varying the telescope alignment around the current position and selecting the alignment with the highest signal. LP-DOAS telescopes thus adaptively counter sudden changes to the system transmissivity e.g. through mechanical interaction with the 20 telescope structure as well as long-term drifts.
In addition to the alignment, atmospheric visibility between telescope and reflectors can vary. Potentially very low visibilities were removed from the comparison data set by excluding days with rainfall.
Other visibility conditions with a similar influence (e.g. fog or smog) did not occur during the comparison period. 25 We estimate the resulting variations of the absolute intensity of the measurement signal from both factors to be 20%. This value has to be considered when comparing absolute atmospheric intensities achieved with the different setups, which determine the temporal resolution of LP-DOAS measurements. For the accuracy ::::::::: accuracy :::: and ::::::::: precision, here assessed through the comparison of the RMS of fit residuals how-ever (Sec. 2.2), due to the use of differential absorption features in DOAS, variations of the recorded absolute intensity only influence photon statistics and hence photon shot noise . ::: (i.e. :::: the ::::::::::: precision). Therefore in the comparisons of residuals from atmospheric measurements, the square root of intensity variations has to be considered and an uncertainty of 10% has to be assumed. For the majority of setups tested here, this is much smaller than the systematic differences of residual RMS values between the different 5 configurations.
It should be noted that in contrast to passive DOAS instruments, changes of the global radiation do not affect LP-DOAS measurements since the atmospheric background signal is corrected with regularly recorded background spectra (see Sec. 2.2). Therefore measurements under e.g. overcast conditions can be compared to observations under clear skies.

10
The combined changes to the LP-DOAS setup, which were found to be the best combination were then tested with a campaign-grade telescope and a smaller Acton 300i spectrometer (setup "NR") during a six weeks campaign in a rural area in the "Nördlinger Ries" in southern Germany. Findings from both campaigns were incorporated in a new, low-maintenance automated LP-DOAS system (setup "NMIII"). It was operated on the German Antarctic station Neumayer III from January 2016 until May 2018 and allows 15 the assessment of the long-term performance of the different components. All telescopes investigated here, were equipped with spherical, aluminium coated mirrors. The light source of a LP-DOAS instrument is a key component because it has a major influence on the achievable signal-to-noise ratio and temporal resolution. The measurement quality depends on both, its temporal and, particularly for arc lamps, the spatio-temporal stability of the light emitting medium (i.e. the plasma) and its spectral characteristics, namely on its (spectral) radiance (see Platt and Stutz (2008) 5 for comparison of different light sources).

Comparison of light sources
In the past, for most LP-DOAS applications xenon arc lamps have been used that often suffered from poor stability of the light arc which affected the optical coupling into the fibre and hence the effective intensity and shape of the lamp's spectral structures. Furthermore, lifetimes of most models with high radiance 10 were relatively short (200 to 2000 h when in constant use; Kern et al. (2006)) and regular replacement during longer measurements required -in addition to the considerable expenses -a time-consuming realignment of the optics after each exchange. Depending on the lamp model used, power consumption was high (up to 500 W plus losses in the power supply), which limited the applicability of LP-DOAS instruments. Additionally, the high voltages necessary for ignition are a shock hazard and cause elec-15 tromagnetic interferences. Although LEDs are useful for compact applications due to their low power consumption and high spatial stability of the light emitting area, up to now light output is not high enough to achieve sufficient signal-to-noise ratios in the ultraviolet regime below 350 nm. Their application has been so far limited to very compact "single housing" systems and to ensure sufficient spectral stability, often considerable efforts for temperature stabilization are necessary (Kern et al., 2006Sihler et al., 20 2009).
In our new LP-DOAS setups presented here, a novel, commercially available Laser-Driven Light Source (Energetiq EQ-99 and the follow-up model Energetiq EQ-99X, in the following referred to as LDLS) was applied both, for laboratory tests and different field measurements. Supplying energy to the xenon plasma with an infra-red laser (rather than a high voltage), it combines the advantages of a high power xenon lamp 25 with long lifetime and high spatio-temporal stability of LEDs at a modest power consumption (140 W).
Similar to conventional xenon lamps, xenon emission lines, whose differential nature can limit sensitivity in DOAS applications (in particular around 450 nm, a spectral window in which e.g. IO or glyoxal can be  technical notes for details; Zhu and Blackborow, 2011a). In a test we performed, the radiance at 255 nm increased by about 30 % compared to no purging when a high flow of nitrogen (about 1 L/min) was used.
Due to logistical reasons, the LDLS most of the time was purged with filtered and dried air during the measurements reported in this study. In the system used for long-term observations in Antarctica (see Sec.

Adaptation of the optical setup to the light source
In addition to a long life time and high spatio-temporal stability, a further advantage of the LDLS is the very small and stable plasma spot due to the very precise localisation of the plasma inside the bulb in 15 the focal point of the laser. Its dimension in the order of 100 µm (full width at half maximum) is about 3 times smaller than in conventional arc lamps (see Tab. 2). This can be exploited in several ways to further improve the design of LP-DOAS systems -first with respect to the configuration of the fibre bundle and overall system optical throughput. Table 3. Étendues for the coupling between the different components in the three setups used in this study (see Tab. 1). For setup HD, the spectrometer étendue for two fibre bundle configurations (classical and reversed) are indicated. See Fig. 2 and Sec. 3.3 for a description.
Coupling HD NR NMIII LDLS → fibre 19.34 · 10 −4 sr mm 2 21.10 · 10 −4 sr mm 2 9.80 · 10 −4 sr mm 2 Telescope 28.73 · 10 −4 sr mm 2 9.89 · 10 −4 sr mm 2 9.89 · 10 −4 sr mm 2 fibre → spectrometer 5.74 · 10 −4 sr mm 2 (classical config.) 76.66 · 10 −4 sr mm 2 76.66 · 10 −4 sr mm 2 29.25 · 10 −4 sr mm 2 (reversed config.) For optical systems in which light propagates unobstructed in a clear and transparent medium, an invariant, the étendue G can be defined as follows (e.g. Welford and Winston, 1978;Markvart, 2007): It is the product of the square of the refractive index n of the medium, the area A of the entrance pupil and the solid angle Ω subtended at this pupil by an object. Since the exact assignment of these quantities 5 depends on the components of an optical setup that are considered, its definition can vary. For the étendue of a light source for example, A could be the size of the emitting area and Ω the solid angle around the emitter that is covered by the light collecting optics.
The étendue allows to link the spectral radiant flux Φ(λ) (in W) through an optical system with transmittance τ (λ) to the spectral radiance R(λ) (spectral radiant flux per solid angle and surface area in W 10 sr −1 m −2 ) of the light source: For a system that consists of several components with different étendues, the overall spectral radiant flux Φ(λ) is limited by the component with the smallest G lim , which makes it a very useful quantity for optical design considerations. For an optimal overall throughput, the étendues of all components should match as 15 closely as possible.
In fibre-based LP-DOAS setups, typically either the spectrometer (where G is the illuminated area of the entrance slit times the solid angle of acceptance of the spectrometer) or the telescope (where G is the entrance area of the fibre core times the solid angle of the light cone hitting the main mirror) have the limiting étendue (see Tab. 3). 20 For a given light collection solid angle, the LDLS has a small étendue compared to other light sources owing to its small plasma spot. The manufacturer indicates for example a maximum attainable numerical aperture of N A = 0.447 (determined by the geometry of the lamp housing) corresponding to a solid angle Ω of 0.663. Assuming an emitting surface with 100 µm diameter, this yields an étendue G max = 52 · 10 −4 sr mm 2 . This is about four times smaller than for a conventional XBO-75 xenon arc lamp with 5 the same coupling optics (see Tab. 2). A small étendue is favourable for optimal utilization of the light source since, regardless of the coupling optics, the usable fraction of the emitted radiation cannot be increased beyond the radiant flux through the element of the system with the limiting étendue. This is illustrated by an investigation of the coupling between different light sources and fibres with different diameters. For LP-DOAS systems, a light source that efficiently can be coupled into a fibre with 10 a smaller diameter is advantageous because it allows to use the fibre bundle in a reversed configuration (see Sec. 3.3 and in Fig. 2 lower row).
Using setup HD (see Tab. 1) with a fixed exposure time and number of scans in reference mode, the intensity spectra of several light sources commonly used for LP-DOAS applications were recorded (shown in Fig. 3). Since angular information of the light is lost on the sand-blasted surface of the reference 15 plate, the spectrometer only samples the intensity of the scattered light. For the comparison between light sources and different fibres, therefore only the coupling between the light sources and the fibre has to be considered.
First, a classical fibre setup described in section 2.1 was used (Fig. 3 panel (a)). The light from the different sources was coupled into a 1 m fibre with 800 µm diameter that was coupled to a 6x200 µm ring 20 of a 3 m long y-shaped fibre bundle (see Fig. 2, upper row, 'classical setup'). The receiving fibre was the single 200 µm core fibre of this bundle which was coupled to a 10 m long 200 µm fibre that led to the spectrometer.
The recorded irradiance of the LDLS in this setup is about twice as high as the conventional xenon arc lamp Osram XBO 75W (in the following referred to as XBO-75), and differential xenon structures are 25 weaker. The commercially no longer available Hanovia PLI-500W (in the following referred to as PLI-500) xenon arc lamp delivers considerably higher irradiances. However, due to poor spatial and temporal arc stability and the resulting introduction of systematic spectral structures as well as a very complicated handling and short lifetime (around 200 h), this light source was excluded from further investigations.  spectral range a 3.5 W Cree XP-E Royal Blue) gives comparable (around 365 nm) or superior irradiances (around 450 nm) for the LEDs however only within the small spectral coverage inherent to the LED principle (Kern et al., 2006).
In a second measurement, a 1 m long 200 µm diameter single fibre with the same numerical aperture was added to the previous setup between light source and the 800 µm :::::::: diameter fibre ( Fig. 3 panel (b)). This In the setup with added 200 µm fibre (panel (b) in Fig. 3), the smaller étendue of the fibre bundle favours the LDLS with its small, high luminance plasma spot relative to the other light sources. The decrease of 10 the transmitted radiant flux compared to the previous setup (about a factor of 2-3 when correcting for a coupling loss of 25%) and relative to all other light sources therefore is the smallest for the LDLS. Both, the XBO-75 xenon lamp (reduction by a factor of 9) and the LEDs (reduction by a factor of 11-13) clearly have lower radiant fluxes than the LDLS and even that of the PLI-500 (reduction of a factor of 10) now is only a factor of 2-3 brighter than the LDLS.

Fibre bundle configurations
The favourable properties of the LDLS when coupled into smaller diameter fibres allow to reverse the yshaped fibre bundle (see Fig. 2 lower row). The ring of fibres previously used for sending is now used for receiving and the central fibre that previously received the returning radiation now sends it out. Reversing the fibre bundle setup has no influence on the instrument transmissivity since the light path is reversible (see C in Fig. 1 and Fig. 2), but offers several advantages. The larger 800 µm fibre is now coupled to the optical slit. If the limiting étendue of the system is that of the spectrometer (as e.g. in setup HD), a larger illuminated area of the slit increases it (e.g. by a factor of 3.8 when comparing a single 200 µm to 5 the 800 µm fibre with a 150 µm optical slit). Furthermore, a larger diameter fibre has better mode mixing properties as will be discussed in detail in Sec. 4 below.
However, the ring of fibres generally has a larger field of view. When in the reversed bundle setup the fibre ring is used for receiving, the atmospheric background light signal (see Sec. 3.4 below) can be six times larger -in particular when the reflector array size is much smaller than the field of view of the ring (often the case in field campaigns due to logistical reasons). This trade-off between the favourable aspects of the reversed fibre bundle configuration like the potential increase in the measurement signal and an 5 improvement of signal quality due to mode mixing (see Sec. 4) has to be weighed against the potentially increased atmospheric background light. If feasible, the latter should be reduced as much as possible by e.g. shielding the area of the ring's field of view around the reflectors with a low reflectance screen. In the following sections, the performance of LP-DOAS systems with both, classical and reversed fibre 15 configurations will be compared and discussed (see Fig. 2 upper and lower row respectively). For all classical configurations (Fig. 2 upper row), in the sending section a 800 µm diameter fibre is used to collect the light from the light source, which is then coupled to a ring of 6x200 µm fibres of y-shaped bundles. The receiving section consists of a single 200 µm that is extended by another 200 µm single fibre if necessary. These classical configurations will be denoted "800→200". In all reversed configurations 20 ( Fig. 2 lower row), the sending section consists of a single 200 µm fibre at the light source that is either the core of a y-shaped bundle or (if required) an extension fibre then coupled to a core fibre of a y-shaped bundle. The receiving section consists of the ring of 6x200 µm fibres then coupled to a 800 µm diameter single fibre. These reversed configurations will be denoted "200→800".
As the influence of lamp performance and fibre configuration in atmospheric measurements cannot be 25 assessed individually in our setup, the noise of the entire measurement system was investigated to quantify the improvement of measurement quality of this reversed setup and results are discussed in section 5.
When operating a LP-DOAS, different types of stray light occur that can cause residual structures and limit the measurement accuracy. Atmospheric background light is scattered into the instrument from outside sources (usually the Sun). It depends on the external light source's relative position and orientation as well as atmospheric properties e.g. the visibility. To correct for background light, background spectra for both 5 atmospheric and reference spectra are recorded on a regular basis by blocking the light source (see Sect. ?? for details).
Internal or spectrometer stray light is caused by unintended deflections of light inside the spectrometer.
In particular in the UV spectral range, where lamp intensities are low compared to the visible spectral range of the light source, this can lead to a systematic underestimation of optical densities since it repre-10 sents an additive quantity with respect to the total received radiance.
We investigated the amount and origin of spectrometer stray light for different spectral regions using setups HD and NR with the telescope shortcut in place and a set of band-pass filters that block increasing portions of the UV-VIS from 280 nm to 665 nm (see Fig. B1 for the set of filters and their effect on a spectrum). Considering the UV spectral region and the spectrometer of setup HD (f=500 mm) with a 15 600 grooves/mm grating (blaze 300 nm), stray light levels increase from less than 1% around 400 nm to about 15% at 240 nm. For smaller wavelengths it quickly reaches 80-90% due to the diminishing spectral radiance of the LDLS in this region. About 95 % of the stray light in this configuration originates from the visible spectral range between 400 nm and 650 nm. For a grating with 1200 grooves/mm in the same spectrometer, stray light levels are between 3 to 7 times smaller than in the previous configuration reaching 20 about 2% at 240 nm. For this grating about 50% of the stray light originates from the visible spectral range between 400 nm and 650 nm with the other half being from the IR. The smaller spectrometer of setup NR (f=300 mm) with a 1000 grooves/mm grating has stray light levels of less than 1% above 320 nm which increases to about 10% at 260 nm.
For evaluations in the spectral region around 330 nm and the 600 grooves/mm (a typical spectral win- 25 dow for the detection of SO2, BrO, formaldehyde, or ozone), an exemplary stray light distribution determined with setup HD with the shortcut in place is illustrated by the histogram in Fig. 4. The relative importance of stray light for UV measurements is further increased when atmospheric spectra are considered. Since Rayleigh scattering is proportional to λ −4 , loss of UV radiation is higher on the way through . the atmosphere compared to the visible parts of the spectrum. This decreases the ratio between UV and VIS and thus increases the relative importance of stray light from VIS spectral regions on UV measurements. Below 300 nm this is further augmented by strong ozone absorption bands. For the NR setup stray light in atmospheric measurements increases from 1.5% at 320 nm to 10% at 290 nm quickly reaching levels of more than 50% for 280 nm and below.

5
There are different ways to suppress spectrometer stray light to reduce its influence on the measurement accuracy. One is to add band pass filters (usually colour glass) between light source and fibre to select only the part of the lamp spectrum needed for measurement. In Fig. 4  The LDLS with its small and stable arc spot allows a second stray light reduction before coupling the light into the fibre. By mounting the entrance of the fibre (A in Fig. 2) on a stepper motor that can translate around the focal point along the optical axis of the lens which projects the plasma spot onto the fibre end, the chromatic aberration of the lens can be exploited to selectively optimize the input of different spectral 15 Fibre Lens f UV f blue f red Due to the steep increase of the refractive index of quartz glass towards shorter wavelengths (Malitson, 1965), this chromatic filter is particularly selective for the UV spectral range. Figure 6 shows the spectral shapes of the radiation reaching the spectrometer when the position of the fibre is optimised for 5 (a) measurements around 260 nm (dashed line in Fig. 6) and (b) for 400 nm (solid black line in Fig. 6).
This can be compared against an artificial intensity distribution that represents the envelope of spectral distributions when the fibre position is tuned through several foci between 200 nm and 800 nm.
When this chromatic aberration filter is optimised for 260 nm, stray light originating around 400 nm, 540 nm, and 680 nm is reduced by about 60 %, 75 %, and 80 % respectively (details are given in appendix 10 B2).

Fibre modes
The modes of an optical fibre represent different distributions of the light travelling inside the core of the fibre. The solutions of the Helmholtz wave equation for the fibre core using Maxwell's equations and considering core geometry and boundary conditions yield the possible modes (Kaminow et al., 2013). 15 For a given wavelength λ 0 the number n of modes is proportional to both, numeric aperture N A and fibre radius a: When the light travels only in a few modes, the resulting light spot leaving the fibre can be inhomogeneous because of the intensity patterns of the individual modes. The intensity distribution between different modes can change along the fibre when energy is transferred from one mode to another which is referred to as mode coupling. This can be caused e.g. by impurities, temperature changes or mechanical stress on the fibre (Stutz and Platt, 1997;Kaminow et al., 2013). In fibres with a smaller diameter in which fewer modes are possible, this inherent or 'natural' mode coupling has a less homogenizing effect than for an otherwise identical fibre with a larger diameter. Therefore, placing a fibre with a larger diameter between 5 telescope and spectrometer as it is the case in the 'reversed fibre configuration' (see Fig. 2 lower row) improves the mode mixing capacity of LP-DOAS setups.

Comparison of mode mixing techniques
In applications with grating spectrometers, an irregular and temporally unstable illumination of the grating resulting from non-uniform illumination of the spectrometer field of view, e.g. due to fibre modes can 10 create systematic, temporally unstable residual structures in the DOAS analysis (see Stutz and Platt (1997) for a detailed study) and thus degrade the measurement accuracy. To reduce these structures, 'artificial' i.e. intentional mode coupling can be induced by different methods. This is referred to as mode mixing (some publications also use the term mode scrambling).
Suitable fibres containing more impurities inherently lead to more mode coupling. However, since impu-  fibre purities leading to a reduced inherent mode coupling in modern fibres (Kaminow et al., 2013). Since larger diameter fibres allow more modes (Eq. 5), adding such a fibre to the y-shaped bundle in front of the spectrometer can have a homogenising effect. The same is achieved by adding diffusing discs to the optical setup (at the expense of transmissivity). Furthermore, mode coupling by mechanical stress can be induced artificially by squeezing or bending the fibre (micro-bending, e.g. Blake et al., 1986;Stutz and 5 Platt, 1997). However, this requires bare fibres to ensure the transmission of the pressure, which makes the handling quite delicate. The mode mixing is also difficult to reproduce and easily influenced by environmental factors such as temperature. Stutz and Platt (1997) solved this by mechanically vibrating a coiled section of the bare fibre in front of the mode mixer to temporally average over different mechanical conditions. When the intensity of the fibre vibration is increased, micro-bending and temporal averaging 10 are effectively combined and can be applied to fibres with a protective coating. This was done in this investigation by attaching the fibre to a vibrating filter pump.
A new method we tested for this study is the intentional degrading of fibre ends to create a quasi builtin diffusing disc. To achieve this, fibre ends were treated with polishing sheets -first with 5 µm and then 12 µm granulation. A homogeneous treatment of the surface was insured by visual inspection using a fibre 15 microscope. Figure 7 shows microphotographs of a fibre end surface before and after the treatment.
This new mode mixing approach was compared to the previously used techniques in a series of atmospheric measurements over a 1.55 km light path (one way) using setup HD (see Tab. 1) and the LDLS as light source. All methods were applied to the fibre bundle between light source (A in Fig. 1) and telescope (C) as well as between telescope (C) and spectrometer (E). However, mode mixing between light source 20 and telescope had almost no effect which probably is due to the fact that homogenized light sent out into the atmosphere still can selectively induce modes in the fibre(s) leading to the spectrometer. One reason for this can be the inhomogeneous illumination of the telescope's field of view because the retro reflector elements do not entirely cover the surface of the array or the light beam partly misses the array. Therefore, in the following only mode mixing between telescope and spectrometer is considered. Light losses caused by the different methods were also quantified and both, the classical and the reversed fibre configurations were tested (Sec. 3.3). In the former setup, a 200 µm fibre was added between the single 200 µm core fibre 5 and the spectrometer. For the latter, a 800 µm fibre was used to couple the 6 x 200 µm fibre ring to the spectrometer. The residual was determined from a fit of 500 added scans between 313 nm and 325 nm considering the cross-sections of O 3 (Bogumil et al., 2003), NO 2 (Bogumil et al., 2003), HCHO (Meller and Moortgat, 2000), and SO 2 (Bogumil et al., 2003). Results for both setups are shown in Tab. 4.
Taking the atmospheric intensity of the uninfluenced fibre as a fixed reference, vibrating the fibre (in the 10 following indicated by "V") causes the smallest light losses for both configurations followed by the novel "Roughened fibre end mode mixing method" (in the following indicated by "R", 12 µm grit). The diffuser (denoted by "D") leads to light losses of 90% (classical setup) and 75% (reversed setup) respectively.
The roughened fibre end yields the lowest residual RMS values for both fibre configurations. Vibrating the fibre in the classical configuration yields comparable residuals at almost twice the intensity of the uals and only a modest loss of intensity (38% compared to the uninfluenced fibre against 23% vibrating against the uninfluenced fibre).

Temporal stability of mode mixing methods
To reduce measurement errors and to lower detection limits, spectra in the LP-DOAS analysis can be added up before the DOAS fitting process. At the expense of temporal resolution, this reduces photon 5 shot noise. It relies on a high temporal stability of the measurement system. A good mode mixing method therefore should also be temporally stable and residuals should decrease when spectra are summed (with photon shot noise as fundamental limit). The potential for residual RMS reductions by adding spectra for both, the classical and the reversed fibre configuration was investigated summing spectra over up to 10 hours (see Fig. 8 upper and lower panel). ::: For ::::: this, ::::::: groups ::: of ::::::: single, :::::::::::::: consecutively ::::::::: recorded :::::::: spectra 10 ::::: from ::: the ::::: tests ::::: with ::: the ::::::::: different ::::::::::::::: configurations ::::: were ::::::::: summed ::: up ::::: after ::: the ::::::::::::::: measurements :: to ::::::::::: correspond ::: to ::::::: periods ::: of :: 1 :: to ::::: 600 :::::::: minutes. : Note that in Fig. 8 results for given measurement times ar ::: are : plotted, thus the effects of the higher photon shot noise due to signal reduction by the various mode-mixing techniques are included. Averaging measurement data is not necessary for typical applications, but may be required if very weak signals need to be identified. 15 In the classical fibre configuration, the diffusing disc only attains residual RMS values comparable to vibrated and roughened fibres when 10 hours of observations are added up. Vibrating and roughening for all time spans yields lower results than without additional mode mixing, with lowest residuals for the roughening (Fig. 8, upper panel). In the reversed setup, the differences between the methods are generally smaller which could be due to the favourable effect on mode mixing of the 800 µm fibre coupling into 20 the spectrometer. Residuals for measurements with the diffuser after 60 minutes are even smaller than for a vibrating fibre. The consistently overall best results for both fibre configurations and all tested mode mixing methods are attained by the reversed (200→800) configuration with roughened fibre end.

The optimal mode mixing setup
Considering light losses (looking at counts per milli second, see Tab. 4), vibrating the fibre leads to 25 smaller losses compared to roughening the fibre end. In the reversed (200→800) fibre configuration, this disadvantage of the roughening is more than compensated by the smaller residuals and also yields the best results when considering the summation of spectra over longer time periods. Compared with vibrating and especially bending the fibre (Stutz and Platt, 1997), it has the additional advantage of being very reproducible and limiting mechanical stress on the fibre.
For the laboratory comparison with setup HD and a 3 m 800 µm fibre coupled to a 3 m y-shaped bundle with 200 µm fibres, a 12 µm grit gave the best results. For a long-term LP-DOAS instrument for operation in Antarctica with a 8.55 m fibre bundle in reversed configuration that includes a 1 m 800 µm fibre (setup 5 NMIII), 5 and 12 µm roughening gave comparable results with a lower light loss for the 5 µm grit size.
We conclude that the fibre bundle needs to be optimized for the particular measurement setup it is used with considering the trade-off between homogenization of the field of view illumination and light loss for the intended application.
5 Overall improvement of LP-DOAS measurement performance 10

Intercomparison measurements in Heidelberg
In order to quantify the combined effect of the changes to the fibre-based LP-DOAS setup discussed above, atmospheric measurements with different configurations were performed with setup HD over a residential area of Heidelberg for a period of six weeks from March 11 until May 3, 2014. During this time, the different configurations were each tested for at least one day. The influence of atmospheric 15 conditions as well as comparability and representativeness of these observations were discussed in Sec.

2.3.
The measurements were analysed in a UV spectral window between 300 and 350 nm (where trace gases such as SO 2 , BrO, formaldehyde, and ozone absorb). As a benchmark, the different improvements were compared against a setup with a XBO-75 xenon lamp in a classical 800→200 fibre configuration. Comparing the two light sources with a classical fibre configuration (setups LDLS-800→200 and XBO_75-800→200 in Tab. 5), the higher intensity and better temporal stability of the LDLS discussed in Sec. 3 are apparent. The average residual RMS for the LDLS is about 30% smaller while the received radiance is about 10% larger. The mode mixing by roughening of the fibre end ("R") reduces the residuals by approximately a factor of 2 for XBO-75 and LDLS (setups LDLS-800→200-R and XBO_75-800→200-R). In the reversed fibre configuration with roughened fibre end mode mixing (setups LDLS-200→800-R and XBO_75-200→800-R), average residual RMS values are comparable. However, the received radiance 5 (and hence temporal resolution for a given signal to noise level) of the XBO-75 is one order of magnitude smaller than for the LDLS, again illustrating the advantage of the smaller arc spot of the latter. Considering the overall improvements from the benchmark setup XBO_75-800→200 to setup LDLS-200→800-R, Table 5. Comparison of atmospheric measurements in Heidelberg utilising the LDLS with different fibre configurations with and without roughening (marked with "R") against the benchmark light source XBO 75 for the HD setup. The number combination in the setup indicates the fibre diameter on the light source end (first number) and spectrometer end (second number). The 800 µm fibre is always coupled to a ring of six 200 µm fibres. The 500 scans cover different averaging times due to different intensities for the configurations but thus contain a similar amount of recorded photons. Errors of the atmospheric intensities reflect variations through alignment and atmospheric conditions. Fibre config.

5
Especially when atmospheric conditions change quickly, short temporal offsets between measurement (atmospheric and reference) and background spectra are important. In each spectral window, several (3-5) of these sets was recorded before the grating was turned to the next spectral position.
To avoid or minimise the influence of detector non-linearity, exposure times for atmospheric and reference spectra in LP-DOAS measurements were adjusted to yield comparable saturations of the CCD. The Given a sufficient temporal stability of the instrumental setup, spectra within sets, across sets, or even 15 across subsequent recordings of a spectral window can be summed to improve the signal to noise ratio and to lower detection limits.  (Nasse et al., 2019, , in prep.). It was set up with two light paths 3.1 km (1.55 km one way) and 5.9 km (2.95 km one way) with nearly the same geographical orientation between which the instrument could switch autonomously depending on atmospheric conditions. This 10 was achieved by moving the end of the fibre (C in Figs. 1 and 2) by a motorised x-y translation stage in the focal plane of the telescope main mirror in order to point the light beam to the respective reflector array.
Since visibility on the ice shelf at Neumayer III station is often reduced by blowing snow or atmospheric refractions due to strong vertical temperature inversions (optical scintillation and El Mirage effects), most of the time measurements were performed on the shorter light path. The measurement routine was similar to the one applied during the Nördlinger Ries campaign with five different spectral windows (see Tab.

D1). Average residual RMS values and detection limits for selected absorbers measured with this setup
can be found in Tab. 6.
During this long period of continuous operation, maintenance requirements were mostly limited to a 5 monthly cleaning of optical components of the setup, in particular the outside of the quartz front window (added to the Neumayer telescope to prevent snow from entering the telescope and allow the interior heating of the telescope), and regular wavelength calibrations. This routine maintenance and the repair of smaller mechanical malfunctions could be performed by the station's wintering crew. Regular major maintenance was only conducted on a yearly basis.

10
The long operation time allows to draw conclusions about the long-term performance of the light source and potential ageing effects of the entire setup. Due to logistical reasons and also in view of the low abundances of organics in the air, the LDLS in this setup was only purged daily for 30 min with filtered air (rather than nitrogen gas). Ambient air was passed through a three stage filtering system of silica gel to remove water vapour, active charcoal to remove gaseous pollutants and a particle filter. The light source 15 reached a total operation time of about 22500 hours before a permanent failure occurred. During this time, no realignment of the light source-fibre coupling optics was required which indicates an exceptional spatio-temporal stability of the plasma spot inside the bulb.
Evolution of the net radiance of reference spectra in the NMIII setup over time for the five spectral ranges UV I through VIS II defined in Tab. D1. The average intensity value (in counts) of the reference 20 spectra in the different measurement windows was corrected for background light by subtracting corresponding reference background spectra (light source blocked and shortcut in front of the fibre end in the telescope).
Then weekly medians were calculated. Since in the VIS I window, the grating was changed in February 2017, the recorded radiances (which due to the lower dispersion of the new grating were about three times higher) were adjusted to that of the first grating (dotted line). Maintenance was performed on a monthly 25 basis (as far as meteorological conditions allowed) and optical components (filters in the light source) were exchanged after roughly one year of operation. This resulted in temporarily enhanced intensities.
erage intensities for all spectral windows (see Fig. 11). The largest decreases of 17% and 12% until the first general maintenance of the instrument after the first year of operation are observed for the UV I and UV II spectral windows. A thorough cleaning of all optical components which was not always possible in Antarctic winter and an exchange of the band pass filters in the light source could restore intensities to 93% and 95.5% respectively. Throughout the 31 months measurement period, intensities were never 5 lower than 80% of the initial values (except for the days directly before the final lamp failure) even when components probably were dirty. The seemingly irreversible intensity decreases of 4.5-7% for the two UV spectral windows over two years might be explained by a permanent reduction of the transmissivity of optical components e.g. by solarisation of fibres and lenses or indeed a decreased output of the LDLS.
Here we present a series of improvements to fibre based long-path-differential optical absorption spectroscopy (LP-DOAS) systems and discuss their respective contributions to the overall improvement of the measurement accuracy ::: and :::::::::: precision. The basis for this study was a mono-static LP-DOAS setup using optical fibre bundles for light coupling between the different instrumental components.
Compared to the : a XBO 75W arc lamp, which was used as the benchmark in our analysis, employing the LDLS leads to 35% smaller residuals while the received signal and hence the temporal resolution of a 15 setup is about 70% higher.
The application of optical fibres in LP-DOAS greatly reduces the complexity of instrument alignment 15 and increases the overall light throughput compared to light coupling with mirrors in the Newton type telescopes used before (Merten et al., 2011). Even though fibres produce a nearly homogeneous illumination, remaining fibre modes can still play a significant role, e.g. by a slightly ::: To ::::::: reduce :: an : inhomogeneous illumination of the spectrometer grating . This causes differential spectral structures limiting the measurement accuracy of DOAS instruments. To homogenise the illumination of the grating, additional ::::::: caused ::: by ::::: fibre 20 ::::::: modes, mode mixing can be introduced. Here, we compared previously applied techniques like vibrating or micro-bending of the fibre, adding diffusers to the optical setup and a new fibre roughening method (see Sec. 4). In this approach, the highly polished end faces of the fibre bundle are artificially degraded with polishing sheets (5 to 12 µm grit size). Thus a quasi built-in diffuser is added to the fibre.

5
By combining the changes to the LP-DOAS setup, namely the use of an LDLS with the improved stray light suppression, the reversed configuration of the fibre bundle, and the new mode mixing method, in intercomparison field measurements in Heidelberg, the residuals could be reduced by a factor of 3-4 compared to the benchmark setup and residual RMS values in the order of 6 · 10 −5 in units of optical density could be achieved (See Fig. 10). Example residual spectra illustrating the improvement are can be 10 found in Fig. 9. As the systematic comparison of the different configurations with benchmark light source and the LDLS for several summation periods in Fig. 10 indicates, this advantage of the new LP-DOAS configuration even increases to a factor of 5 in residual RMS when spectra are summed for 10 ( Fig. 8 and Sec. 5).
When the improvements described above were applied to two campaign grade LP-DOAS setups using 15 smaller spectrometers (see Tab. 1), residuals of the order of (0.9−1.0)·10 −4 were achieved under optimal conditions and on average 1.1 − 2.0 · 10 −4 in a long-term measurement campaign in Antarctica. Measurements in the UV spectral region particularly benefit from these improvements. During the measurements in Antarctica, for instance average detection limits of ClO were between 6 to 7.5 pptv at temporal resolution between 4 and 40 minutes. BrO could be detected at detection limits of 0.6 to 0.8 pptv at temporal 20 resolutions of 2 to 30 minutes (see Tab. 6).
In conclusion, the application of the LDLS with its greatly reduced operational complexity and maintenance requirements, its high spatial and temporal stability as well as its long life time has enabled a number of technical improvements to the fibre based LP-DOAS setup. These increase measurement accuracy : , :::::::::: precision, : and reliability of LP-DOAS systems and make this versatile remote sensing technique 25 much easier to deploy even in longer field campaigns or permanently operated applications.
B2 Locations of foci for different spectral ranges Appendix D: Measurement routine in field campaigns in the Nördlinger Ries/Germany and Antarctica not included in the DOAS fit, the residual RMS value in the spectral region where the absorber would be fitted can be used. A upper limit of a detectable concentration c lim can be inferred by calculating the concentration of the absorber along the light path L for which the optical depth is as large as the root 5 mean square (RMS) of the residual optical density: This estimate tends to be an upper limit for actually achievable detection limits.
wintering crews of the station with respective air chemists Thomas Schaefer, Zsófia Jurányi, and Helene Hoffmann for taking good care of the instrument. JMN is ::: was : supported by scholarships of the Evangelisches Studienwerk Villigst and the Studienstiftung des deutschen Volkes.