Construction of a unified, high-resolution nitrous oxide data set for ER-2 flights during SOLVE



[1] Four nitrous oxide (N2O) instruments were part of the NASA ER-2 aircraft payload during the 2000 SAGE-III Ozone Loss and Validation Experiment (SOLVE). Coincident data from the three in situ instruments and a whole air sampler are compared. Agreement between these instruments was typically good; however, there are several types of important differences between the data sets. These differences prompted a collaborative effort to combine data from the three in situ instruments, using an objective method, to produce a self-consistent, high-resolution, unified N2O data set for each SOLVE flight. The construction method developed by the four N2O instrument teams is described in detail. An important step in this method is the evaluation and reduction of bias in each of the in situ data sets before they are combined. The quality of unified N2O data is examined through its agreement with high-accuracy and high-precision N2O data from whole air samples collected from the ER-2 during SOLVE flights. Typical agreement between these two data sets is 2.9 ppb (1.5%), better than the typical agreement between any pair of N2O instruments.

1. Introduction

[2] Nitrous oxide (N2O) is an extremely useful tracer of the dynamical motions of stratospheric air masses. This utility stems from three properties of atmospheric N2O: a lack of significant tropospheric sinks, well-characterized photochemical sinks in the middle stratosphere, and a ∼100-year atmospheric lifetime that is much longer than transport timescales. The equality of N2O mixing ratios in different stratospheric air masses is often taken to imply similarity in their photochemical histories. This implication leads to the frequent use of N2O mixing ratios as a basis to identify and quantify the roles of dynamics and chemistry in observed changes in stratospheric air mass composition, such as ozone loss, denitrification, and dehydration [e.g., Proffitt et al., 1990; Fahey et al., 1990a, 1990b; Hintsa et al., 1998].

[3] The ER-2 payload was modified for the 2000 SAGE-III Ozone Loss and Validation Experiment (SOLVE) by replacing the Airborne Tunable Laser Absorption Spectrometer (ATLAS) [Podolske and Loewenstein, 1993], a tunable diode laser spectrometer (TDL), with Argus, a compact, lightweight TDL [Loewenstein et al., 2002]. Three other N2O instruments remained as part of the longer-term ER-2 payload: the Aircraft Laser Infrared Absorption Spectrometer (ALIAS) [Webster et al., 1994, 2001], the Airborne Chromatograph for Atmospheric Trace Species (ACATS-IV) [Elkins et al., 1996; Romashkin et al., 2001], and the Whole Air Sampler (WAS) [Heidt et al., 1989; Schauffler et al., 1999]. The instrument type, mean N2O data interval, and mean measurement precision and accuracy for these instruments are given in Table 1.

Table 1. Instrument Types, N2O Data Intervals, and Mean Measurement Precision and Accuracy
InstrumentTypeN2O Data IntervalbData Integration Period,c sMeasurement Precision, %Measurement Accuracy, %
  • a

    Measurement precision and accuracy are estimated at the 95% confidence level (2σ). See section 4.2.3 for detailed information about estimating measurement uncertainties for each flight. Accuracy estimates are the sums, in quadrature, of estimated total measurement uncertainties (2σ) and the 1% uncertainties (2σ) of the NIST and NOAA/CMDL N2O calibration scales.

  • b

    Data intervals are the mean periods between reports during all SOLVE flights and may be larger that the data rate quoted in the text because of nonreporting periods.

  • c

    For ACATS and WAS this represents the span of time that air samples integrate within the instrument before measurement. For the TDLs this is the interval over which spectra are co-added to produce a reported datum.

ACATSgas chromatography with electron capture detection (GC/ECD)80 s3–50.351.4
ALIAStunable diode laser and quantum cascade (QC) laser spectrometer1.8 s1.41.33.5
Argustunable diode laser spectrometer3.7 s2.02.76.4
WASwhole air samples and GC/ECD32 samples per flight10–3000.151.2

[4] Until recently, little consideration was given to differences between N2O measurements made by multiple instruments on board the ER-2 aircraft. A recent examination of coincident data from three in situ N2O instruments during the 1995–1996 Stratospheric Tracers of Atmospheric Transport (STRAT) and 1997 Photochemistry of Ozone Loss in the Arctic Region In Summer (POLARIS) campaigns revealed that differences between instruments were often greater than expected from their combined measurement errors [Hurst et al., 2000]. Additionally, the four N2O instruments of the SOLVE ER-2 payload flew together with ATLAS in the 1999 Tracer Intercomparison Experiment for SOLVE (TIES). This study was a blind intercomparison of atmospheric N2O measurements on board the ER-2 aircraft, but the quantity of data was very limited compared to STRAT, POLARIS, or SOLVE.

[5] In the first part of this work we examine differences between coincident data from the four N2O instruments on board the ER-2 aircraft during SOLVE. The comparison reveals several instances of flight-long biases between instruments and occasional large measurement differences that significantly impaired the agreement between instruments. These biases and sporadic large differences between instruments motivated the N2O instrument teams to develop a method to reduce biases in the three in situ N2O data sets and combine them into a unified N2O (UN2O) data set for each SOLVE flight. The construction method and the resulting UN2O data sets are described in the second part of this paper.

2. ER-2 Flights and Instruments

[6] The 16 ER-2 SOLVE flights discussed in this work (Table 2) include a mission-ready test flight (flight 1) from NASA Dryden Flight Research Center, California (34°N), ferry flights (2–4) from Dryden to Kiruna, Sweden (68°N), via Westover Air Reserve Base, Massachusetts (42°N), flights conducted out of Kiruna (5–15), and the return ferry flight from Kiruna to Westover (16). The final return ferry flight from Westover to Dryden on 18 March 2000 was not fully instrumented and is omitted from this study. Tracks for flights 5–15 ranged from northern surveys deep into the core of the polar vortex to southern surveys outside the vortex. Overall, SOLVE ER-2 flights spanned the latitude range 21° to 90°N. Cruise altitudes for the ER-2 were typically 19–20 km which corresponded to pressures of 47–65 hPa and potential temperatures of 420–510 K. The SOLVE scientific objectives and flight details are discussed by Newman et al. [2002].

Table 2. Flight Numbers, Flight Dates, Numbers of N2O Measurements, and Ranges of Measurement Differences Between Each Instrument Pair
  • a

    Ranges of absolute values of measurement differences between instrument pairs are given as minimum-maximum in ppb N2O.

  • b

    NA, not available.


[7] The four-channel ACATS-IV instrument of the NOAA Climate Monitoring and Diagnostics Laboratory (CMDL) measures N2O every 70 s by in situ gas chromatography with electron capture detection (GC/ECD). Ten other trace gases, including methane (CH4), sulfur hexafluoride (SF6), and various halocarbons, are also measured at intervals of either 70 or 140 s. The GC is 0.8 ×0.5 ×0.3 m and weighs 48 kg. Air from the sample inlet is flushed through the instrument at 150–200 standard cm3 min−1 by a Teflon diaphragm pump. This flow rate is regulated by an absolute pressure relief valve between the pump and sample loops. Transit time of air from the inlet through the instrument is about 30 s, but air samples integrate in sample loops for only 3–5 s before injection, providing trace gas measurements for discrete air parcels.

[8] The 2.0 cm3 sample loop for the N2O (and SF6) channel is pressurized to 870 hPa and injected by a 12-port, two-position valve onto a 0.6-m long by 1.8-mm ID precolumn of 80/100 mesh Porapak Q. The main chromatography column is 1.8 m long but otherwise identical to the precolumn. To achieve the rapid separation of N2O and SF6 necessary for 70 s chromatograms, a 15-cm long by 1.8-mm ID column of 60/80 mesh molecular sieve 5A was plumbed directly to the main column outlet. The Porapak Q columns and molecular sieve column were maintained at 95° and 165°C, respectively. SF6 rapidly passes through the short molecular sieve 5A column while N2O is delayed by about 10 s. CO2 accumulates on this column and slowly bleeds into the detector, providing a near-constant level of ECD doping. The Valco ECD was operated in constant current mode at 350°C.

[9] ACATS-IV was calibrated during flight after every seventh sample analysis for N2O (every 560 s) by analyzing a whole air standard diluted by 20% with zero air. Calibration data were used to correct drifts in ECD responses during flight. Calibration curves were produced in the laboratory by analyzing five gas standards, including the in-flight standard, which were diluted from 0 to 80% with zero air. These gas standards were calibrated by NOAA/CMDL with certified N2O accuracy of ±1%. ACATS-IV and its one- and two-channel predecessors have flown on 137 ER-2 flights during six major NASA missions including SOLVE.

[10] The NASA Jet Propulsion Laboratory's four-channel ALIAS is a very high resolution scanning tunable laser spectrometer that for SOLVE made direct, simultaneous measurements of N2O, HCl, CO, and CH4 at sub-part-per billion level sensitivities. Infrared sources are tunable lead-salt diode lasers and quantum cascade lasers operating in the 3.4 to 8 μm wavelength region. The lasers scan over absorption lines and spectra are recorded using direct and second harmonic absorption spectroscopy. The 80-m optical path is defined by two spherical mirrors spaced 1 m apart in a multipass Herriott cell configuration. Design features include an in-flight wavelength reference cell rack, a mechanical fringe spoiler, and a liquid nitrogen Dewar with 48-hour hold time that houses the four lasers and four detectors. A sample flow rate of 13 standard L s−1 is maintained through the isokinetic sample inlet, and air samples transit from heated inlet to cell in about 0.03 s and flush the 16.75 L cell every 1.3 s. The Herriott cell is maintained at a constant 12°C. The instrument weighs 73 kg and fits easily in a 0.5 ×1.1 ×0.5 m mounting rack. ALIAS has flown over 250 times in six major NASA missions.

[11] During SOLVE, ALIAS measured N2O by scanning a tunable diode laser over two closely spaced N2O lines at 2232.518 cm−1, and 2233.273 cm−1. The reported N2O mixing ratios were based on averages of the mixing ratios calculated from these two lines. N2O data were reported at intervals of 1.4 s between 1.4 s reference scans performed every 14 s. The instrument response was calibrated before flight using gas standards prepared by NOAA/CMDL with certified N2O accuracy of ±1%. Additionally, a 13CO2 spectral line located halfway between the two N2O lines at 2232.946 cm−1 was used for quality control of in-flight N2O measurements to within ±1%.

[12] The NASA Ames Research Center's Argus instrument is a small (0.4 ×0.3 ×0.3 m), lightweight (21 kg plus 17 kg for calibration system), dual-channel tunable diode laser spectrometer that has successfully flown on balloon gondolas and most recently aboard the NASA ER-2 aircraft. Argus uses second harmonic detection at 3028.752 and 2206.659 cm−1 for CH4 and N2O, respectively. Lasers and detectors are contained in a liquid nitrogen-cooled Dewar. Both optical channels probe the same temperature-controlled, multipass, astigmatic Herriott cell with a 36 m optical path. Second harmonic spectra are processed by standard phase-sensitive amplifier techniques with demodulation occurring at twice the laser modulation frequency of 40 kHz [e.g., Podolske and Loewenstein, 1993]. A nonlinear, least squares Marquardt-Levenberg algorithm with five free fitting parameters is used to fit the measured second harmonic line shape. Spectra are coadded for 2.0 s for each reported N2O datum.

[13] Airflow into Argus is restricted to 0.3 to 0.5 standard L s−1, resulting in a sample transit time of 1–2 s and a cell flush time of 0.6 to 1.0 s. The instrument is calibrated in the laboratory by 0–100% dynamic dilution of a NOAA/CMDL gas standard with N2O-free air at six cell pressures between 40 and 300 hPa. During flight, calibrations are performed by alternately flowing NOAA/CMDL gas standards of 180 and 310 ppb N2O through the sample cell for approximately 1 min every 20 min of flight time. In-flight calibrations and the stable sample cell temperature minimize data corrections required by the strong temperature dependence of the N2O and CH4 spectral line parameters, resulting in high accuracy measurements.

[14] The National Center for Atmospheric Research (NCAR) whole air sampler consists of a four-stage metal bellows pump, stainless steel manifold, electronics control package, and 32 electropolished stainless steel canisters for each flight. All canisters were evacuated just prior to flight, and all but about six were filled with 4 hPa of water vapor to wet the interior surfaces for several halocarbons that degrade in dry canisters. Canisters were filled to about 3000 hPa at pre-determined intervals during each flight, then returned to the NCAR laboratory in Boulder, Colorado, for analysis of N2O and a variety of other trace gases. The transit time of air from inlet to canister ranged from 30 to 120 s depending on the atmospheric pressure. Filling periods for canisters ranged from about 10 s at 8 km altitude to 3.5–5 min at 20 km, hence each WAS datum represents an average mixing ratio during the fill period. Measurement timestamps assigned to WAS data denote the times when each canister was filled to half its final pressure. Samples were analyzed 3–21 days after their collection.

[15] WAS canisters were analyzed for N2O using a Hewlett-Packard 5890 series II+ GC with ECD. The GC precolumns and main columns were 3-m and 2-m long, 3.7-mm ID stainless steel filled with 80/100 mesh Porapak Q. N2O and CO2 are well separated by this column pair, as demonstrated by a CO2 analyzer plumbed to the ECD exhaust. Both columns and the 10 cm3 sample injection loop were maintained at 50°C, while the ECD was held at 350°C. Each sample was analyzed twice between analyses of a 314 ppb N2O secondary standard of whole air calibrated against a 300 ppb NIST-certified SRM (2608, ±1%). Volumetric dilutions of the NIST SRM with zero air were analyzed to create a calibration curve describing the nonlinear ECD response to N2O over a mixing ratio range of 60–300 ppb. A cylinder of whole air calibrated at 311 ppb N2O by NOAA/CMDL was analyzed on the WAS N2O GC and assigned a mixing ratio of 309.8 ppb. This 0.4% difference is well within the ±1% certified accuracy of the NIST and NOAA/CMDL N2O standards. Measurement uncertainties for these four instruments during SOLVE flights are discussed in detail in section 4.2.3.

3. Comparison of N2O Data

3.1. Data Matching

[16] Several methods of data matching were needed to deal with the dissimilar data rates of in situ instruments and the integrated sample data of WAS. These methods were designed to create sufficient populations of matched data for each flight to permit statistically meaningful evaluations of the agreement between instruments. In situ data were placed in 1 s bins based on their measurement timestamps recorded by independent computer clocks in each instrument. These clocks were synchronized just prior to each flight.

[17] In the most basic sense, a data match between two in situ instruments occurred when both instruments reported within a 1-s bin. Differences between matched in situ data (δ) were computed for each unique pair of in situ instruments: Argus-ACATS, ACATS-ALIAS, and ALIAS-Argus. The δ values for all SOLVE flights were placed in 2.5 ppb bins to produce a histogram for each instrument pair (Figures 1a, 1b, and 1c). For many flights, only <30% of ACATS data were coincident with TDL (ALIAS or Argus) data in 1 s bins. To increase the number of ACATS data used in these comparisons, data from both TDLs were interpolated to 1 s resolution over data gaps <10 s. Histograms of δ for ACATS and interpolated TDL data (Figures 1a and 1b) utilize 87% of the ACATS data but are not markedly different from the histograms of non-interpolated data matches. Further evidence that ACATS-TDL data comparisons are not skewed by the use of interpolated TDL data is found in the similar histograms of δ for ALIAS-Argus and interpolated ALIAS-Argus (Figure 1c).

Figure 1.

Differences between coincident N2O measurements of instrument pairs (δ) were placed in 2.5 ppb-wide bins to create histograms. Data matching techniques for the various instrument pairs are described in the text. Each histogram is centered on δ = 0 and approximates a normal distribution. Thin black bars represent δ between (a, b) ACATS and interpolated TDL data and (c) Argus and interpolated ALIAS data.

[18] WAS and TDL data were matched in time by integrating TDL data over the fill period of each WAS canister. Each TDL datum was weighted by the fraction of the final WAS canister pressure added during that second, as measured by a pressure gauge on the WAS manifold. The weighted TDL data were summed over the canister fill period and divided by the sum of weights. For most flights a data match required that TDL data were reported in ≥75% of the total number of 5 s intervals it took to fill the canister. This criterion was lowered to ≥66% for ALIAS data during flights on 000131, 000203, and 000311, and for Argus data during the flight on 000203 to match >50% of the WAS canisters with TDL data. This integration technique allowed an average of 75% of the WAS canisters filled during each flight to be matched with data from each TDL. Values of δ for matched WAS and TDL data during SOLVE were binned in the same way as δ for the in situ instrument pairs to create histograms (Figures 1d and 1e).

[19] ACATS data were integrated over WAS canister fill periods >70 s. A data match required >1 ACATS report during fill periods of 71–174 s, >2 reports during 175–244 s fill periods, >3 reports for 245–314 s fill periods, and >4 reports for 315–384 s fill periods. WAS data not matched by this method were directly matched with any ACATS datum reported within ±15 s of a WAS timestamp (i.e., the pressure midpoint of a canister). The stability of N2O mixing ratios during each canister fill period was carefully examined to determine whether ACATS data reported every 70 s could adequately describe the N2O content of the canister. Data matches were retained only if (1) ALIAS N2O data were reported in ≥50% of the total number of 5 s intervals it took to fill the canister, and (2) the coefficient of variation (CV, σ/mean) of ALIAS N2O data during the canister fill period was <6%. Using these criteria, ACATS data were matched and retained for an average of 16 (54%) of the WAS canisters filled each flight. Though more stringent criteria produced closer temporal matches between these two data sets and allowed less N2O variability during canister fill periods, they permitted too few data matches to carry out a meaningful comparison. The δ values for matched WAS and ACATS data were placed in 2.5 ppb bins to produce a histogram (Figure 1f).

[20] All six difference histograms are centered about δ = 0 (Figure 1), illustrating that long-term biases and calibration differences between the four N2O instruments were <1.25 ppb during SOLVE. Each distribution of δ is near-normal, suggesting random measurement errors and perhaps shorter-term biases as the primary causes of nonzero δ values. Substantial wings in some of these histograms reflect occasional large δ values during some flights. Values of equation image for each instrument pair ranged from 0 to 42.8 ppb (Table 2), with large values often associated with rapid ascent or descent of the aircraft.

3.2. Mean Differences and Typical Agreement

[21] A mean difference equation image was computed for every flight from the δ for each instrument pair (Figure 2). Interpolated TDL data were used to calculate equation image for ACATS and each TDL. Each panel in Figure 2 displays the equation image values for one instrument paired with each of the other three. The sign and magnitude of equation image values for a specific instrument pair vary from flight to flight, further evidence that there were no significant long-term biases or calibration differences between any pair.

Figure 2.

Mean differences between N2O instruments (equation image) were calculated for each SOLVE flight. Each panel depicts the equation image between one instrument and the three others. The 68% confidence limits of equation image, ±3.0 ppb (dashed horizontal lines), were exceeded by 13 instrument pairs during six flights. Flight numbers are linked to flight dates in Table 2.

[22] The absolute values of mean pair differences (equation image) ranged from <0.1 to 6.2 ppb (<0.1% to 3.1%) and averaged 1.6 ppb (0.8%). Hereinafter, except where noted, relative statistical values presented in parentheses (%) immediately following statistical values in ppb N2O were calculated using the 200 ppb mean N2O mixing ratio measured during SOLVE. The 68% confidence limit (mean + σ) of equation image, 3.0 ppb (1.5%), was exceeded by a total of 13 instrument pairs during six flights. For several flights the equation image values for one instrument with the other three are of the same sign, indicating that data from the one instrument may be biased for that flight. A critical review of in-flight data by each instrument team (see below), including diagnostic data for each instrument, disclosed no correctable artifacts in the measurements of any instrument.

[23] Nonzero equation image values (Figure 2) illustrate that flight-long biases served to impair the agreement between instruments. Shorter-term biases and random measurement errors may also be important contributors to differences between instruments. The typical agreement between each instrument pair during SOLVE is calculated as the 68% confidence limit (CL) of the absolute value of their δ values during the entire mission. The use of gaussian statistics is warranted because each distribution of δ is near-normal (Figure 1). Typical agreement values range from 3.6 to 7.3 ppb (1.8 to 3.7%) for the different instrument pairs (Figure 3).

Figure 3.

Typical agreement between the six N2O instrument pairs for all SOLVE flights was calculated at the 68% level of confidence. Bars represent the mean plus 1σ of the absolute values of δ between instrument pairs for all SOLVE flights, in ppb N2O (bottom axis) and relative to the 200 ppb mean N2O mixing ratio measured during SOLVE (top axis).

[24] The idea of combining the N2O data of the three in situ instruments into a UN2O data set was motivated by indications that some instruments may be biased during certain flights or flight segments. A few large equation image and δ values for some instrument pairs resulted in values >6 ppb (3%) for their typical agreement. Intuitively, the UN2O construction method must utilize all three in situ N2O data sets to evaluate and reduce biases and large δ values between the instruments before their data are combined.

4. Construction of UN2O Data Sets

[25] The fundamental goal of combining in situ data sets was to report UN2O data at <10 s intervals with the highest accuracy and precision possible. The quality of UN2O data was to be assessed by its agreement with WAS data, and an agreement target of 3 ppb (1.5%) was prescribed. The construction method was to be developed in an iterative, cooperative effort of those persons with the greatest knowledge of the strengths and shortcomings of the four instruments, namely, the N2O instrument teams.

4.1. Initial Data Review

[26] Before the construction methodology was developed, each instrument team closely examined their own data for correctable problems. An area of concern was the synchronization of measurement timestamps from the three in situ instruments. The primary issues hindering timestamp synchronization between instruments include unsynchronized computer clocks and poorly characterized sample transit times from inlet to measurement. Without any knowledge of which instrument had the most accurate timestamps, it was agreed to synchronize all timestamps to those of the Harvard CO2 instrument [Boering et al., 1994]. The CO2 instrument reported very high precision data at 2-s resolution during every SOLVE flight, except during periods of calibration, and has well-characterized sample transit times from inlet to the instrument. Since N2O and CO2 are both long-lived in the stratosphere, their mixing ratios are tightly correlated. Hence a shift in N2O timestamps relative to CO2 timestamps for a given ER-2 flight would either improve or degrade the correlation.

[27] Sample transit times for ACATS were determined for each SOLVE flight by methodically reducing the N2O measurement timestamps in 1-s increments relative to CO2 timestamps to find the maximum in the N2O:CO2 correlation coefficient. This procedure was carried out on 3–5 different ranges of CO2 mixing ratios for each flight depending on the range of CO2 measured. Transit times calculated for the 3–5 CO2 ranges were averaged for each flight, and any value >3σ from the mean was discarded, and the mean was recalculated. ACATS timestamps for all SOLVE flights were reduced by an average of 31 ± 3 s (1σ). The adjusted timestamps are judged to be synchronized to better than ±3 s with CO2 timestamps for each flight.

[28] Sample transit and cell flushing times for Argus and ALIAS were small (<3 s) because of their high sample flow rates. Nevertheless, TDL measurement timestamps were checked against Harvard CO2 measurement timestamps by comparing, for each flight, correlation curves of CO2 with TDL N2O against reference N2O:CO2 curves. TDL N2O timestamps were shifted in 1-s increments relative to CO2 timestamps until the best agreement was found between the N2O:CO2 curves and the appropriate reference curve. Best agreement was indicated by the minimum standard deviation of differences between the TDL and reference curves. Reference correlation curves were generated by plotting timestamp-corrected ACATS N2O against CO2 for each flight. Correlations were fit with second-order polynomials over two different ranges of CO2 mixing ratios (low: 359.0–361.4 ppm; high: 361.4–367.0 ppm). The entire range of CO2 could not be fit well with a simple curve (i.e., second-order polynomial), making two ranges necessary. Correlations for two groups of flights, 1–5 (midlatitude) and 11–16 (late vortex), are very similar, so data from these flights were combined and fit over the low and high CO2 ranges. Correlations for flights 6–10 (early vortex) were similar to one another but significantly different from the other correlations. These data were combined and fit independently over the two CO2 ranges. In all, four curves were generated: midlatitude/late vortex and early vortex curves for both low and high CO2 ranges.

[29] This procedure suggested 1–2 s timestamp adjustments for each TDL for most flights. Larger timestamps corrections of 7–10 s were required for a few flights. The corrected TDL timestamps are deemed to be synchronized with CO2 timestamps to within ±1 s for each flight.

4.2. Methodology

[30] A flow chart of the UN2O calculation methodology is presented as Figure 4.

Figure 4.

The flow chart depicts the different steps of the method employed to calculate unified N2O data for each flight.

4.2.1. Data Bins

[31] The disparate data rates of the three in situ N2O instruments aboard the ER-2 aircraft cause many 1-s bins to contain data from only one instrument. In formulating a method to combine the data of three instruments, the intent was to construct each output datum from the data of more than one instrument when possible. With this in mind, N2O data from each instrument were placed in 3-s bins to dramatically increase the overlap of data from different instruments. The bin width was limited to 3 s because the data within each bin must be measurements of one intrinsic value. Specifically, 3-s bins require that the N2O mixing ratio of the air mass probed by the ER-2 aircraft cruising at 200 m s−1 does not change significantly over a horizontal distance of 0.6 km. The long atmospheric lifetime of N2O typically results in little variation of its mixing ratio over this horizontal distance, except near barriers to quasi-horizontal transport (e.g., tropical pipe and polar vortex edge). Vertical movements of the ER-2 aircraft in the lower stratosphere also induce variability in N2O mixing ratios with time. During SOLVE, N2O abundance in the lower stratosphere decreased with altitude at an average rate of approximately 25 ppb km−1, so the maximum ER-2 ascent/descent rate of 22 m s−1 translates to a maximum change in N2O of 0.55 ppb s−1 due to aircraft vertical motion. This potential 1.6 ppb (0.8%) change in N2O within each 3-s bin during ascent and descent was deemed acceptable because it is <3 ppb (1.5%), the agreement target between the WAS and UN2O data sets. There are 3-s bins where CO2 variability and the reference N2O:CO2 correlations indicate that N2O should vary >3 ppb, but these bins represent only 0.6–2.7% (1.4% mean) of the UN2O data for each flight.

4.2.2. Calculation Method

[32] The agreed upon method of combining data from three instruments was to calculate a weighted mean mixing ratio for each 3-s bin containing data. The standard formula for determining a weighted average X from measurements with different uncertainties is

equation image

where χi and ϵi are the individual measurements and their uncertainties. Note that the appropriate weight for each measurement is the reciprocal square of its measurement error. The uncertainty γ in X based exclusively on estimated measurement errors ϵi is calculated from

equation image

[33] There were many multiple reports in 3 s bins from one or both of the TDLs. Of all the bins where ALIAS reported data, 26, 66, and 8% were single, double, and triple reports. For Argus these fractions were 69, 31, and 0%. Since each report within a bin is a repetitive measurement of the same quantity, multiple reports by a TDL were first condensed into a weighted bin mean mixing ratio X for that TDL using equation (1). An additional uncertainty σ for the mean of multiple reports, the root-mean-square (RMS) of differences between each repetitive measurement and the weighted bin mean for n reports by that TDL, was calculated as

equation image

The two uncertainty terms γ and σ were added in quadrature to compute the total uncertainty E of the weighted bin mean X for that TDL as

equation image

The condensation of multiple reports by a TDL into a weighted bin mean for that TDL allows σ to be included in E. This ensures that E accounts for variability in the data used to calculate X that may not be adequately represented by the values of γ which are typically 25–75% lower than the estimated measurement errors ϵi. The value of σ will significantly increase E (>3%) only if it is >25% of γ, otherwise it has little effect on E.

[34] Weighted bin means X and uncertainties E were then calculated from the data of all contributing instruments using equations (1)(4). Here each χi is either a weighted bin mean for a multiple-reporting TDL or a single report from an instrument, and the uncertainty σ is an assessment of the agreement between the weighted bin mean or single report of each contributing instrument. In this context, σ is an assessment of the agreement between contributing instruments, and its inclusion in E accounts for differences between the χi that may not be adequately represented by the ϵi.

4.2.3. Measurement Errors

[35] The calculation of weighted bin means depends strongly upon the estimated measurement uncertainties ϵi. A substantial effort was made by each instrument team to estimate measurement uncertainties for each flight using diagnostic information unique to their instrument. The 1% accuracy of both the NIST and NOAA/CMDL N2O calibration scales is ignored in error calculations because no long-term biases were observed between WAS and the three in situ instruments (Figure 1). The focus here is on measurement uncertainties specific to each instrument that can lead to differences between instruments. All uncertainties were estimated at the 95% CL (2σ). Flight-averaged uncertainties for each instrument are presented in Table 3.

Table 3. Average N2O Measurement Uncertainties
  • a

    Average uncertainties are given in ppb N2O at the 95% confidence level (2σ).


[36] The uncertainty of each ACATS N2O measurement was assessed from the precision of in-flight calibration data surrounding the measurement and the uncertainty of quadratic fits to laboratory calibration data that define the N2O calibration curve. Precision (1σ) is calculated as the rms of residuals of calibration data from the time series of running, three-point weighted means of calibration data [Romashkin et al., 2001]. Calibration curve fitting uncertainties include scatter in the ECD responses to each of five calibration tanks and random errors in the N2O mixing ratio assigned to each calibration tank. These three uncertainties were summed in quadrature to calculate the uncertainty of each in-flight measurement. ACATS measurement uncertainties (2σ) averaged over each flight ranged from 1.2 to 2.6 ppb N2O (0.6 to 1.3%) and averaged 2.0 ppb (1.0%) for all SOLVE flights.

[37] Argus measurement uncertainties were determined from calibrations performed during each flight. Standard gases of 180 and 310 ppb N2O in air were alternately flowed through the Herriott cell for approximately 1 min every 20 min. Short-term measurement precision was evaluated for each calibration period from the variability of calibration data. The flight-long uncertainty resulting from instrumental drift was estimated from the variability of the mean N2O mixing ratio calculated for each calibration performed during the flight. Short-term precision and flight-long drift estimates were added in quadrature to determine the uncertainty of each N2O measurement. Flight-long uncertainty estimates were not determined for several flights because of calibration contamination problems, so drift estimates from adjacent flights were used. Flight-averaged measurement uncertainties for Argus ranged from 6.8 to 26.9 ppb (3.4 to 13.4%) and averaged 12.6 ppb (6.3%) for all flights.

[38] The average uncertainty of ALIAS N2O measurements during each SOLVE flight was estimated as the sum in quadrature of the average random error and bias. Uncertainty estimates were based on in-flight spectral noise and laboratory calibration data because ALIAS does not perform calibrations during flight. Random error was assessed from the RMS amplitude of noise in the spectra recorded on the N2O channel during flight. Each amplitude was converted from percent to equivalent N2O mixing ratio and averaged over the entire flight. Instrumental bias was evaluated from the stability of calibrations performed in the laboratory. For the periods 6 January to 3 February 2000 (flights 1–10), and 26 February to 16 March 2000 (flights 11–16), the 2σ bias was estimated to be 2.3 and 2.6%. These biases were converted to ppb N2O for each flight using the mean N2O mixing ratio measured by ALIAS during that flight.

[39] The method of error estimation for ALIAS is different from the ACATS and Argus methods because ALIAS does not perform in-flight calibrations with N2O gas standards. However, the ALIAS N2O channel is continuously calibrated using a 13CO2 spectral line as described earlier. It is critical for calculating weighted means from these three instruments that their error estimates are based on similar criteria. The similarity in ACATS and Argus error estimation methods leads to the assertion that their error estimates are on equivalent foundations. To check the equivalence of ALIAS error estimates, the 95% CL (mean plus 2σ) of the absolute value of differences between WAS and each TDL was calculated for every flight. Average Argus error estimates for every flight were greater than or equal to the WAS-Argus 95% level of agreement. ALIAS error estimates were less than the WAS-ALIAS 95% level of agreement for eight flights. It was agreed that the ALIAS error estimates for these eight flights should be increased to the value of WAS-ALIAS agreement to put ALIAS error estimates on a similar basis as Argus error estimates. The increases averaged 3.1 ppb, but significantly reduced the weight of ALIAS data in UN2O calculations for only the 000127 and 000311 flights. Average ALIAS uncertainties for flights ranged from 4.0 to 13.4 ppb (2.0 to 6.7%) and averaged 6.7 ppb (3.4%) during SOLVE.

[40] Uncertainties in WAS N2O are based on the precision of measurements and uncertainties introduced by sample canister filling and storage. Uncertainties were calculated as the root sum of squares of the CV for repetitive standard and sample analyses and the CV for analyses of multiple canisters. The average uncertainty (2σ) for WAS N2O data from all SOLVE sample canisters was 0.7 ± 0.1%. This value was converted to ppb N2O for each flight using the mean N2O mixing ratio measured in the WAS canisters for each flight.

4.2.4. Evaluation and Reduction of Instrumental Bias

[41] As illustrated above (Figure 2), there are significant biases between instruments for entire flights. Shorter-term biases may also degrade the agreement between instruments. It was therefore necessary to develop a method to identify and reduce biases between instruments before combining their data. If estimated errors truly account for measurement artifacts, it can be argued that biased data will be lightly weighted in computing bin means. However, if only the biased instrument reports within a bin, the bias will be transferred directly to the UN2O product. For this reason it was decided to evaluate and reduce the bias in each instrument's data prior to the calculation of UN2O.

[42] First, a basis was formed for the evaluation and reduction of bias in each instrument's data. Reference N2O mixing ratios (χREF) were computed as the weighted mean (equation (1)) of every bin where ACATS and at least one TDL reported. Multiple reports within bins by a TDL were first condensed into weighted bin means for that TDL, and the σ term was added to the uncertainty E, as discussed above (Figure 4). There was an average of 249 χREF computed for each flight.

[43] Next, data from each instrument were divided into discrete 600 s segments, and the data in each segment were compared to coincident χREF (Figure 5a). The evaluation of bias in the data of an instrument required that there were >2 matches with χREF within a segment, as well as one match <300 s before and after the segment. The number of matches required per segment guaranteed that ≥5 data points were involved in each linear fit (see below). Data segments failing these criteria were not evaluated for bias and were discarded. The segment length (600 s) and presegment and postsegment match windows (300 s) were chosen such that each TDL met the match criteria for 90% of the data segments.

Figure 5.

The bias identification and reduction method was based on three steps. (a) Time series of raw N2O data from an instrument (grey) and the calculated reference N2O data (black) are shown. The 600-s segment under evaluation lies between the two black vertical lines. Grey vertical lines depict differences between coincident raw data and reference N2O. Note the requisite data match both before and after the 600-s segment. (b) The time series of differences is fit by linear regression. The standard deviation of residuals from the fit (σr) is compared to a threshold value for that flight. (c) If σr is less than the threshold value, the linear equation is used to shift the raw data within the segment into better agreement with reference N2O.

[44] Data segments meeting the match criteria were evaluated for bias by fitting the time series of their differences (δ) from χREF (δ = χ − χREF) using linear regression (Figure 5b). If the 1σ standard deviation of residuals from the linear fit (σr) was less than the threshold value chosen for each flight (Table 4), data within the segment were shifted using the slope and intercept of the fit line (Figure 5c). The threshold value (Table 4) for each flight was chosen so that data from at least one TDL were shifted and retained for each 600 s segment. The linear equation for the time series of differences was y(t) = αt + β, and each datum in the segment was shifted by subtracting (αt + β). Shifted data were were written to new data files used to calculate UN2O and were never used to evaluate bias in the other data sets. The value of σr was added in quadrature to the estimated error of each measurement (ϵi) within the segment to account for the additional uncertainty introduced by the bias reduction procedure. This procedure was carried out on the data from each in situ instrument for every SOLVE flight. The mean and maximum shifts applied to retained data from each of the in situ instruments for each flight are reported in Table 4.

Table 4. Threshold Values for σr and the Mean and Maximum Data Shifts Resulting From the Bias Reduction Method
FlightThreshold Value (ppb N2O)Argus MeanArgus MaximumALIAS MeanALIAS MaximumACATS MeanACATS Maximum

[45] The bias evaluation and reduction method required one data match with reference N2O before and after each 600 s segment to improve the continuity of UN2O data from one segment to the next. Hence data shifts for adjacent segments were not entirely independent because two data matches were common to both segments.

[46] After bias reductions were completed for each instrument, the shifted data were combined into UN2O and uncertainties using equations (1)(4). Raw and UN2O data from a 2000 s segment of the 000226 flight (Figure 6) illustrate that the bias reduction technique resulted in self-consistent data. The data interval for UN2O averaged 3.2 s during all flights, and ranged from 3.0 to 4.2 s for individual flights. The longest data gaps in UN2O for each flight ranged from 6 to 69 s, with a median gap of 14 s for all flights.

Figure 6.

(a) 2000 s (33 min) of N2O data from the three in situ instruments are shown along with the calculated reference N2O data for the 000226 flight. Timestamps are given as elapsed seconds since 0000 that day at the Greenwich meridian. (b) UN2O data (black squares) for the same flight segment are shown along with reference N2O (grey diamonds). Note the good agreement between unified and reference N2O that results from the bias reduction method.

[47] There is good correspondence between the UN2O and CO2 for a 950 s segment of the 000127 flight (Figure 7). These data were taken as the ER-2 aircraft turned from a southeast heading to a northwest course about 500 km south of Moscow, Russia, and descended from 20 to 18.6 km altitude. A large, rapid CO2 increase of 1.1 ppm in 36 s (42,242 to 42,278 s UT) and subsequent rapid CO2 decrease were measured as the aircraft passed through 19.8 km altitude at 51.7°N latitude and 36.5°E longitude. UN2O nicely tracks this and other CO2 changes during the flight segment (Figure 7).

Figure 7.

Time series of UN2O (upper symbols) and Harvard CO2 (lower symbols) data for a 950-s segment during the flight of 000127. Note the good general correspondence between the two data sets while the ER-2 flies through a region of considerable tracer variability at 19.8 km altitude about 500 km south of Moscow. Rapid, large changes in CO2 between 42,200 and 42,320 s UT are well tracked by the UN2O data. CO2 data are not available between 42,593 and 42,704 s UT because of an in-flight calibration during this period.

[48] Data match criteria and threshold values for σr were chosen specifically to retain at least one set of TDL data per 600 s segment. ACATS data met these criteria without exception, so UN2O data are reported for every segment. It was not possible to retain data from both TDLs for every segment without greatly relaxing the threshold values. Consequently, data from only one TDL and ACATS were retained for 18% of the 600-s segments. This average excludes the 000123 flight where ALIAS and ACATS data were retained for all data segments. ACATS reports once every 70 s (every 23 bins), so 90–95% of the unified data within these 18% of data segments are based on solo reports by one TDL. Solo reports are an important issue when considering the impacts of instrumental noise on UN2O (see below).

4.2.5. Instrument Codes and Uncertainties

[49] An instrument code was generated for each UN2O datum to depict which instrument(s) contributed to it. Instrument codes were based on a number assigned to each instrument: Argus, 1; ALIAS, 10; ACATS, 100. Instrument codes representing combinations of instruments were calculated as the sums of numbers assigned to the contributing instruments (e.g., Argus and ALIAS, 11).

[50] Table 5 lists the percentages of UN2O data for each flight that are associated with each of the seven instrument codes. Except for flight 10, instrument codes 10 (ALIAS solo) and 11 (ALIAS and Argus) together represent 74–96% of the UN2O data for each flight. Code 1 (Argus solo) represents <15% of UN2O except for flights 10 and 14. The maximum contribution of any other code (100, 101, 110, or 111) during a flight is 3.6%. The fractions of UN2O data based partly or wholly on data from ALIAS, Argus, and ACATS were 92, 53, and 4%, respectively.

Table 5. Percentages of Unified N2O Data Contributed by Each Instrument Code
FlightCode 1Code 10Code 11Code 100Code 101Code 110Code 111
  • a

    The instrument code for each UN2O datum indicates which instrument(s) contributed data to the datum. Codes are the sum of numbers assigned to the contributing instruments: Argus, 1; ALIAS, 10; ACATS, 100. For example, a UN2O datum represented by code 11 indicates that Argus and ALIAS data were used.


[51] Uncertainties in UN2O were calculated as described above (equations (2)(4)). For each flight the average uncertainty of UN2O data associated with each instrument code is presented in Figure 8. For a given flight the wide range of mean uncertainties for different codes stems predominantly from the wide range of measurement uncertainties quoted for the instruments. Mean uncertainties for a specific code vary from flight to flight because of changes in instrument performance during SOLVE.

Figure 8.

Plotted instrument codes indicate the mean uncertainties of UN2O data associated with each of the seven instrument codes for each flight. Codes for each flight appear between the vertical grid lines that separate different flights. For flight 1 the mean uncertainty for code 1 data lies off the graph at 24.0 ppb. There were no UN2O data with code 101 for flight 1, code 100 for flight 3, and codes 1, 11, 101, and 111 for flight 6.

4.2.6. Noise Reduction

[52] It was not possible to develop a method to reduce noise in the measurements of each instrument. Reference N2O data available every 70 s reveal nothing about the variability of N2O over distances <14 km. Significant variability of N2O is present in the TDL data at or below this spatial scale during numerous flight segments (e.g., Figure 6). Potential causes of this small-scale variability range from atmospheric variability, such as that encountered when the ER-2 flew near or through the vortex edge region, to pure measurement noise. Differentiation between these two potential causes of small-scale variability in N2O data is extremely difficult because the 3-s bins often contained data from only one instrument (i.e., there were no other N2O data to compare with). Though there is a good general correlation between N2O and other rapidly measured, long-lived tracers such as CO2 (Figure 7), peculiarities in these correlations observed near the vortex edge region during SOLVE [e.g., Richard et al., 2001] impair their ability to provide a reliable basis for evaluating noise in N2O data.

[53] The primary strategy to reduce any instrumental noise in the N2O data was to rely on the data averaging method (1), both for multiple reports by a solo instrument and for reports by all contributing instruments. It was expected that the averaging of several data in each 3 s bin would reduce any noise present, yet at the same time preserve any real atmospheric structure that was observed by the instruments over timescales >3 s. In general, the averaging method reduced noise in the raw data without eliminating what are believed to be atmospheric variations in the data (Figure 6).

[54] A concern stemming from the inability to reduce instrumental noise in the individual data sets was that UN2O data based on a single datum from one instrument would contain any noise in that single, solo report. The fractions of UN2O data based on single, solo reports by ALIAS and Argus ranged from 7–21% and 1–24% for individual flights. Overall, 16% of all UN2O data are based on single, solo reports. Given the inability to discriminate between atmospheric variability and instrument noise, and the requirement that real atmospheric structure be preserved by the UN2O calculation method, it was not possible to keep noise in single, solo reports from being directly transferred to the UN2O data. However, close examination of conspicuously noisy UN2O data reveals that their differences from seemingly less noisy data are almost always within the quoted 2σ uncertainties of UN2O. We recommend that users of the UN2O data sets take into account the uncertainty of each datum before performing analyses.

4.3. Agreement With WAS

[55] Agreement between UN2O and WAS was evaluated by integrating UN2O over each WAS canister fill period. Two flights (000114 and 000120) lacking WAS data could not be appraised. A data match required that UN2O data were present in 90% of the total number of 5-s intervals it took to fill a canister. This criterion matched only 7 canisters for the 000203 flight and was reduced to 80% for this flight to yield 23 matches. The number of matches for each flight ranged from 19 to 31 (25 mean), representing 59–94% (82% mean) of the WAS canisters filled during each flight.

[56] Differences between matched WAS N2O and UN2O data during 14 SOLVE flights averaged 0.6 ± 1.9 ppb and are near-normally distributed (not shown), indicating no long-term bias between these two data sets. The agreement for each flight was calculated as the 68% CL of the absolute values of δ for WAS and UN2O data (Figure 9). Agreement for individual flights ranged from 1.2 to 4.4 ppb (0.6 to 2.2%). Typical agreement during SOLVE, the 68% CL of the absolute values of differences (δ) for all flights, was 2.9 ppb (1.5%). This agreement is slightly better than the target of 3.0 ppb (1.5%) and is superior to the agreement between any instrument pair (Figure 3). For reasons not yet understood, WAS-UN2O agreement during flights 6, 7, 10, and 16 was worse than the 3.0 ppb target.

Figure 9.

Agreement between WAS and UN2O data for each flight was calculated as the 68% confidence limit of the absolute values of differences between matched data. See text for the data matching technique and criteria. Typical agreement for combined flights was 2.9 ppb (1.5%), slightly better than the target of 3 ppb (dotted horizontal line). Agreement was worse than 3 ppb for flights 6, 7, 10, and 16.

5. Summary

[57] An objective method was developed to construct unified, high-resolution N2O data sets from three in situ instruments on board the NASA ER-2 aircraft during SOLVE. The impetus for creating UN2O data sets was the realization that several types of important differences existed between the measurements of individual instruments. Targets of temporal resolution higher than 10 s and agreement with WAS N2O better than 3 ppb (1.5%) were prescribed for UN2O. An important step in the construction method was the identification and reduction of biases in the individual data sets before they were combined. The product UN2O for all SOLVE flights was reported at an average interval of 3.2 s, and typical agreement with WAS N2O was 2.9 ppb (1.5%). Each UN2O datum is reported with a 2σ uncertainty calculated from the estimated uncertainties of individual measurements and the variability of data combined into the datum. An instrument code depicting the instruments that contributed data to each UN2O datum is also reported. The UN2O data files from SOLVE are available at the anonymous ftp site, in /archive/solve/data/er2/UN*.ER2, where the asterisk is the flight date (YYYYMMDD).


[58] We are greatly indebted to ER-2 pilots Jan Nystrom, Dee Porter, Jim Barrilleaux, and Ken Broda for their heroic efforts in flying these sorties, and the ER-2 ground crew for their diligence in getting the aircraft safely airborne in less-than-balmy weather. The logistical organization of Kathy Wolfe, Mike Craig, Sue Tolley, Steve Hipskind, Quincy Allison. Mike Kaptizke, and Wendy Dolci made our lives considerably easier during SOLVE. Thanks to Steve Wofsy, Arlyn Andrews, Bruce Daube, and Christoph Gerbig for providing Harvard CO2 data for the ER-2. N2O standards for ACATS, ALIAS, and Argus were prepared and calibrated by Richard Myers and Brad Hall of NOAA/CMDL. Geoff Dutton and Fred Moore of CMDL provided invaluable assistance and advice for ACATS. We thank Jeff Grose, NASA Ames, for his expertise in all aspects of Argus instrument operation. J.B.G. gratefully acknowledges the National Research Council for funding his Research Associateship at NASA Ames. Special thanks to B. S. Ka-Ching for the spark that initiated this collaborative effort. Part of the research described in this paper was carried out by the Jet Propulsion Laboratory, California Institute of Technology, under a contract to the National Aeronautics and Space Administration. This work was funded in part by the NASA Upper Atmosphere Research Program (UARP).