Optimization of automated gas sample collection and IRMS analysis

13
Optimization of automated gas sample collection and IRMS 1 analysis of δ 13 C of CO 2 in air 2 Matthias J. Zeeman 1 , Roland A. Werner 1 , Werner Eugster 1 , Rolf T. W. Siegwolf 2 , Günther Wehrle 2 , 3 Joachim Mohn 3 , Nina Buchmann 1 4 1 Institute of Plant Sciences, ETH Zurich, Universitaetsstrasse 2, CH–8092 Zurich, Switzerland 5 2 Laboratory of Atmospheric Chemistry, Paul Scherrer Institute, Bachstrasse 1, CH–5232 Villingen, Switzerland 6 3 Laboratory for Air Pollution & Environmental Technology, Empa, Überlandstrasse 129, CH–8600 Dübendorf, 7 Switzerland 8 Abstract 9 The application of 13 C/ 12 C in ecosystem–scale tracer models for CO 2 in air requires accurate mea- 10 surements of mixing ratios and stable isotope ratios of CO 2 . To increase measurement reliability and 11 data intercomparability as well as to shorten analysis times, we have improved an existing field sampling 12 setup with portable air sampling units and developed a laboratory setup for analysis of δ 13 C of CO 2 in 13 air by isotope ratio mass spectrometry (IRMS). The changes consist of (a) optimization of sample and 14 standard gas flow paths, (b) additional software configuration and (c) automation of liquid nitrogen refill- 15 ing for the cryogenic trap. We achieved a precision better than 0.1 and an accuracy of 0.11±0.04 16 for δ 13 C of CO 2 in air and unattended operation of measurement sequences up to 12 hours. 17 18 The interest in the global atmospheric car- 19 bon cycle has intensified as a response to re- 20 ported trends in global climate change. These 21 trends are primarily related to atmospheric in- 22 creases in greenhouse gas concentrations. 1 On 23 the global average, carbon dioxide (CO 2 ) plays 24 the most important role and thus ecosystem ori- 25 ented research has particularly focused on CO 2 . 26 The potential use of the stable isotope ratios of 27 CO 2 (e.g. 13 C/ 12 C, 18 O/ 16 O) in ecosystem–scale 28 atmosphere–biosphere process studies has often 29 been highlighted and is believed to be a power- 30 ful tool for carbon cycle studies, in particular to 31 disentangle ecosystem flux components. e.g. 2–7 It is 32 commonly used to quantify mixing contributions 33 from sources with differing isotopic composi- 34 tions. 8,9 However, this requires accurate measure- 35 ments of both CO 2 mixing ratios and isotopic com- 36 position in order to be useable in ecosystem–scale 37 tracer model approaches. 4,10,11 On local (species 38 to ecosystem) scales this can be quite a challenge; 39 CO 2 mixing ratios and isotopic composition in the 40 air close to the vegetation are known to fluctu- 41 ate strongly, i.e. on short time scales (seconds to 42 hours), especially under less turbulent atmospheric 43 conditions due to accumulation of CO 2 . Moreover, 44 with conventional flask sampling the measurement 45 strategy is mostly limited to discrete sampling, and 46 typically these samples need to be transferred to a 47 distant laboratory for analysis by an Isotope Ratio 48 Mass Spectrometer, so the insight into ecosystem 49 processes is hampered by technical and logistical 50 constraints. 51 In this paper, we aim to optimize and ex- 52 tensively test air sampling and analysis of stable 53 carbon and oxygen isotope ratios in atmospheric 54 CO 2 for stable isotope studies at the ecosystem 55 level. The setup described here has been suc- 56 cessfully used for grassland ecosystem studies in 57 Switzerland and intercomparisons of stable iso- 58 * Correspondence to: MJ Zeeman, ETH Zurich, Institute of Plant Sciences, Universitaetsstrasse 2, CH–8092 Zurich, Switzer- land, Email: [email protected], Phone: +41 44 632 81 96, Fax: +41 44 632 11 53 1

Transcript of Optimization of automated gas sample collection and IRMS analysis

Page 1: Optimization of automated gas sample collection and IRMS analysis

Optimization of automated gas sample collection and IRMS1

analysis of δ 13C of CO2 in air2

Matthias J. Zeeman1, Roland A. Werner1, Werner Eugster1, Rolf T. W. Siegwolf2, Günther Wehrle2,3

Joachim Mohn3, Nina Buchmann14

1Institute of Plant Sciences, ETH Zurich, Universitaetsstrasse 2, CH–8092 Zurich, Switzerland5

2Laboratory of Atmospheric Chemistry, Paul Scherrer Institute, Bachstrasse 1, CH–5232 Villingen, Switzerland6

3Laboratory for Air Pollution & Environmental Technology, Empa, Überlandstrasse 129, CH–8600 Dübendorf,7

Switzerland8

Abstract9

The application of 13C/12C in ecosystem–scale tracer models for CO2 in air requires accurate mea-10

surements of mixing ratios and stable isotope ratios of CO2. To increase measurement reliability and11

data intercomparability as well as to shorten analysis times, we have improved an existing field sampling12

setup with portable air sampling units and developed a laboratory setup for analysis of δ 13C of CO2 in13

air by isotope ratio mass spectrometry (IRMS). The changes consist of (a) optimization of sample and14

standard gas flow paths, (b) additional software configuration and (c) automation of liquid nitrogen refill-15

ing for the cryogenic trap. We achieved a precision better than 0.1� and an accuracy of 0.11±0.04�16

for δ 13C of CO2 in air and unattended operation of measurement sequences up to 12 hours.17

18

The interest in the global atmospheric car-19

bon cycle has intensified as a response to re-20

ported trends in global climate change. These21

trends are primarily related to atmospheric in-22

creases in greenhouse gas concentrations.1 On23

the global average, carbon dioxide (CO2) plays24

the most important role and thus ecosystem ori-25

ented research has particularly focused on CO2.26

The potential use of the stable isotope ratios of27

CO2 (e.g. 13C/12C, 18O/16O) in ecosystem–scale28

atmosphere–biosphere process studies has often29

been highlighted and is believed to be a power-30

ful tool for carbon cycle studies, in particular to31

disentangle ecosystem flux components.e.g. 2–7 It is32

commonly used to quantify mixing contributions33

from sources with differing isotopic composi-34

tions.8,9 However, this requires accurate measure-35

ments of both CO2 mixing ratios and isotopic com-36

position in order to be useable in ecosystem–scale37

tracer model approaches.4,10,11 On local (species38

to ecosystem) scales this can be quite a challenge;39

CO2 mixing ratios and isotopic composition in the40

air close to the vegetation are known to fluctu-41

ate strongly, i.e. on short time scales (seconds to42

hours), especially under less turbulent atmospheric43

conditions due to accumulation of CO2. Moreover,44

with conventional flask sampling the measurement45

strategy is mostly limited to discrete sampling, and46

typically these samples need to be transferred to a47

distant laboratory for analysis by an Isotope Ratio48

Mass Spectrometer, so the insight into ecosystem49

processes is hampered by technical and logistical50

constraints.51

In this paper, we aim to optimize and ex-52

tensively test air sampling and analysis of stable53

carbon and oxygen isotope ratios in atmospheric54

CO2 for stable isotope studies at the ecosystem55

level. The setup described here has been suc-56

cessfully used for grassland ecosystem studies in57

Switzerland and intercomparisons of stable iso-58

∗Correspondence to: MJ Zeeman, ETH Zurich, Institute of Plant Sciences, Universitaetsstrasse 2, CH–8092 Zurich, Switzer-land, Email: [email protected], Phone: +41 44 632 81 96, Fax: +41 44 632 11 53

1

Page 2: Optimization of automated gas sample collection and IRMS analysis

tope ratio instrumentation (e.g. a comparison of59

a quantum cascade laser based absorption spec-60

trometer, a field-deployable Fourier transform in-61

frared spectrometer and an Isotope Ratio Mass62

Spectrometer).e.g. 12–14 The basic considerations63

for the chosen measurement approach haves been64

(a) the collection of samples at multiple locations65

for (b) sample measurements by laboratory based66

high precision Isotope Ratio Mass Spectrometer.67

The most important implication of this approach68

is that the conditions (e.g. temperature, pres-69

sure) might be different between location of sam-70

ple collection and the laboratory. Thus, gas sam-71

ples might be contaminated during the storage pe-72

riod between sampling and analysis, which is es-73

pecially likely if samples are collected at higher74

altitudes under reduced ambient pressure.15–18 For75

time series analysis, e.g. to understand diur-76

nal cycles or effects of weather events, samples77

or series of samples are repetitively collected at78

equally spaced time intervals. If a Keeling plot ap-79

proach (inverse [CO2] related to isotope δ–value)80

is used, the accuracy of y–axis intercepts is di-81

rectly related to the precision and accuracy of the82

measurements.cf. 4 Thus, the analysis must be as83

accurate and precise as possible, deviations should84

be on the order of 0.1� for δ 13C at most.85

To achieve our aims, we have substantially86

improved existing gas sampling equipment previ-87

ously described by Theis et al. 19 and developed a88

new Isotope Ratio Mass Spectrometer setup, pro-89

gramming and measurement routines for δ 13C of90

CO2 in air. An overview of this improved setup91

is shown in Figure 1 for both field and laboratory92

setup. An important part of these improvements93

was to optimize the automation of the operations94

during sampling and isotope ratio analysis to allow95

for accurate timings and increased reproducibility.96

Thus, our objectives were to 1) apply digital com-97

munication protocols between the sampling unit98

and the control computer to store status informa-99

tion from the sampling unit in order to eliminated100

the potential error of sample misidenfication.cf. 20101

2) We wanted to increase precision and reliability102

of the IRMS measurements for CO2 in air sam-103

ples and optimize sample preparation steps. 3) We104

wanted to reduce the time required per IRMS anal-105

ysis of a CO2 in air sample to increase throughput106

in the laboratory and reduce storage times of the107

samples.108

Methodology109

Field setup110

Three devices are used in our field setup (Fig. 1),111

consisting of an home-built air inlet selection unit,112

followed by an InfraRed Gas Analyzer (IRGA)113

for CO2 mixing ratios (model LI-840, LI-COR,114

Lincoln, Nebraska, USA) and a sample mani-115

fold at the end. This sample manifold is a116

modified and improved version of the device117

termed Automated Sampler of Air (ASA) by Theis118

et al. 19 . It contains 33 glass flasks sample con-119

tainers connected to three multiport Valco-valves120

(EMTMA2ST12MWE, VICI, Schenkon, Switzer-121

land) allowing independent filling of each individ-122

ual sample container with sample air. We con-123

tinue to use the abbreviation “ASA” to refer to the124

portable air sampling unit described here, because125

its key components (the Valvo-valves) and its func-126

tion as sample manifold have not changed with127

respect to the Theis et al. 19 version, despite the128

modifications described here.129

During field deployment, a single inlet is se-130

lected from a series of continuously purged air in-131

lets (Synflex™Type 1300, formerly known as Dek-132

abon™, Gembloux SA/NV, Belgium; ID 4 mm,133

≈ 1 L min−1). After a particle filter (Gelman,134

LI-COR), a T-split diverts the airflow (a) to the135

IRGA and a subsequent small pump (DC12/8FK,136

Fürgut GmbH, Germany) inside the inlet selec-137

tion unit, and (b) to an ASA sample inlet. The138

flow through the IRGA is kept at a continuous139

rate of 0.9 L min−1 (Fig. 1), within the manufac-140

turer supplied specifications for the IRGA. Once141

inside the ASA (Fig. 2), the sample air is pushed142

by a pump, diverted on activation of a solenoid143

valve (EVT307-5D0-02F-Q, SMC, Weisslingen,144

Switzerland) through a drying column containing145

magnesium perchlorate (Fluka, Switzerland) and146

is filtered (SS-4FW-2, Swagelok, USA) before be-147

ing pushed further through 300 mL glass flasks148

(Ernst Keller & Co AG, Basel, Switzerland) or149

10 mL stainless steel loops (SL10KSTP, VICI,150

Schenkon, Switzerland) connected to the multiport151

Valco-valves with≈ 0.9 L min−1. At one of the 12152

positions of each Valco-valve a short stainless steel153

capillary is used as low volume bypass to allow154

aligning multiple Valco-valves in series, typically155

three or four per ASA. To create an over-pressure156

2

Page 3: Optimization of automated gas sample collection and IRMS analysis

of at least 50 kPa in the sample containers, we157

changed the original Theis et al. 19 design and the158

position of the Teflon membrane pump (N811K-159

DC, KNF, Germany) in combination with a poppet160

check valve (SS-6C-MM-1, Swagelok, USA) and161

an adjustable flow meter (V-100, Vögtlin, Switzer-162

land) at the exit. By having pressurized sam-163

ple containers, the chance of contamination dur-164

ing transport and laboratory analysis is minimized.165

In the laboratory (Zurich, 400 m above sea level,166

a.s.l.), the pressure excess (pressure above ambi-167

ent) directly after the pump inside the ASA (Fig. 2)168

was typically found to be ≈ 90 kPa and ≈ 50 kPa169

before the adjustable flow meter. This is suffi-170

cient for the collection of samples at alpine loca-171

tions (e.g. >2000 m a.s.l.), though higher pres-172

sures could be reached at the expense of flow rates.173

Although multiple ASAs can be used in series174

for sampling by using the bypass position of the175

solenoid valves, we have chosen a parallel setup176

utilizing a flow split with poppet check valves (SS-177

6CA-MM-3, Swagelok, USA) to prevent any back-178

flow from ASAs with inactive pumps or from the179

open connections after removal of one ASA. The180

position of the IRGA (Fig. 1) parallel to the ASA181

has shown no discernible different results. As an182

advantage over a sequential setup our parallel ver-183

sion allows to continue the concentration measure-184

ments of the sample gas even if the ASAs are dis-185

connected or inactive. The IRGA can alternatively186

be positioned directly before the ASA to directly187

analyze the gas that subsequently flows through the188

sample containers of the ASA.189

The inlet selection unit and ASAs are config-190

ured and operated by a field computer via RS-191

232 serial communication lines connected to dig-192

ital controllers (C-Control I STATION 2.0, Con-193

rad Electronics GmbH, Germany), programmed to194

operate the rotation valves, solenoid valve, pump,195

and digital flow meter, and to provide status infor-196

mation for later use in post-processing of CO2 con-197

centration and stable isotope data. The keypad and198

LCD display of the digital controllers are used to199

confirm correct operation of the devices before and200

during field deployment. The serial communica-201

tion, data storage and post processing are handled202

by scripts written in Perl language.203

After having completed an in situ field sam-204

pling sequence, the ASAs with up to 33 or 44 gas205

sample containers per ASA are transported back to206

the laboratory for subsequent (same day) isotope207

ratio analysis.208

Laboratory setup209

The precise determination of δ 13C and δ 18O val-210

ues in CO2 of large numbers of air samples im-211

plies a precise and reproducible sampling tech-212

nique as well as an automated and easy-to-use213

coupling of the sample containers (glass flasks or214

steel loops) to the Isotope Ratio Mass Spectrom-215

eter (DeltaplusXP, Finnigan MAT, Bremen, Ger-216

many).217

A series of multiposition valves are used for218

the flow path of the sample preparation (Fig.219

3). A 6-position dead-end path Valco-valve220

(ASD6MWE,VICI, Schenkon, Switzerland) and a221

4-port 2-position Valco-valve (AC4UWE, VICI,222

Schenkon, Switzerland) allow the alignment of up223

to four independent reference air gas bottles (lab-224

oratory air gas cylinder with different CO2 mix-225

ing ratios and δ -values) or helium, using the same226

sample preparation path as the gas sampled with227

the ASA allowing referencing according to the228

Identical Treatment (IT) principle.21 A feed cap-229

illary delivers pure He to the ASA (Fig. 3, valves230

1 and 2), allowing a pressure build-up in the glass231

flasks that flushes the sample gas at a rate of about232

5 mL min−1 through a water trap (Nafion dryer) to233

the cryogenic focus trap where condensable gases234

(mainly CO2 and N2O) are cryogenically trapped.235

After diverting the non-condensable gases to a vent236

(Fig. 3, valve 4), the cryogenically trapped sample237

is thawed and subsequently flushed by He into the238

Gas Chromatograph column (Poraplot Q 25 m ×239

320 nm i.d., Varian, Walnut Creek, USA, held at240

24 ◦C) to allow separation of CO2 from N2O and is241

subsequently led to the Isotope Ratio Mass Spec-242

trometer for analysis (Fig. 3, valve 5). The trap-243

ping efficiency was checked beforehand with an244

IRGA (LI-840, LI-COR, Lincoln, Nebraska, USA)245

behind the frozen cryogenic trap. In contrast to246

Theis et al. 19 , who used a Precon (Finnigan MAT)247

hooked up to the Isotope Ratio Mass Spectrome-248

ter, we modified the Gasbench II system (Finnigan249

MAT) to directly interface with individual ASA250

units. This modification of the Gasbench (Fig. 3,251

bottom panel) comprises the replacement of the252

“Gas Chromatograph”-type split22 by a ConFloIII-253

like split23 and the replacement of the stainless254

3

Page 4: Optimization of automated gas sample collection and IRMS analysis

steel sample loop with a home-built cryogenic fo-255

cus trap (1/16" stainless steel capillary filled with256

Ni-wire) at the 8-port valve inside the Gasbench,257

which is configured to operate as 6-port valve (Fig.258

3, valve 4). A second 4-port 2-position Valco-259

valve (AC4UWE, VICI, Schenkon, Switzerland)260

inside the sample preparation path (Fig. 3, valve261

3) operates as a vent to release the pressure ex-262

cess inside the sample containers and allows for263

high flow purging of the sample preparation path264

(10 mL min−1) and the Isotope Ratio Mass Spec-265

trometer flow path (17 mL min−1) with pure He.266

In our system without Precon this would otherwise267

not have been possible. Without the posibility to268

flush with high flow (50 mL min−1) as shown by269

Theis et al. 19 , our measurement time would have270

been 835 s. With the help of the pressure vent and271

related high He flows we are able to reduce the272

analysis time per sample to 610 s, a period com-273

parable to Theis et al. 19 .274

The cryogenic trap and all valves in the Gas-275

bench, the external referencing unit, the rotary276

valve systems of the ASA and an automated liquid277

nitrogen refill procedure are computer controlled278

by modified Isodat script language (ISL) scripts,279

available in the vendor supplied ISODAT NT soft-280

ware package (Ver. 2.0 SP2.63, Finnigan MAT).281

To avoid overloading of the cryogenic trap282

with sample gas of high CO2 concentration (>283

1000 µmol mol−1) and to circumvent a possible284

non-linearity of the Gasbench and Isotope Ra-285

tio Mass Spectrometer combination with signal286

strength a, the signal strength of each sample is ad-287

justed to be close to that of the Isotope Ratio Mass288

Spectrometer reference by changing the cryogenic289

trapping period depending on the sample concen-290

tration. We first tested the relation between the291

sample CO2 concentration and the cryogenic trap-292

ping period empirically for each of the different293

sample container volumes and tube lengths, deter-294

mined the best fit (Fig. 4) and tested the results295

with known dilutions of a CO2 in air mixture of296

a known stable isotope composition. Furthermore,297

to improve stable conditions for the cryogenic trap,298

the liquid nitrogen (LN2) level in the Dewar was299

kept within the 95–100% range, either by manual300

refill or automated refill. To facilitate partly unat-301

tended operation (12 hours) and thus a certain level302

of autonomy of this setup, we installed a LN2 re-303

fill system that consists of a balance measuring the304

weight loss of the evaporating LN2 from the Dewar305

in combination with timing signals received from306

the Isotope Ratio Mass Spectrometer (Fig. 5) and307

that is supplied by a 30 L LN2 tank (up to 48 hours308

operation). Empirically, de-icing of the Dewar is309

required every 12 hours.310

As a last step, the ISODAT NT software config-311

uration had to be adjusted to our modified setup, in312

particular the measurement timing schedule (Fig.313

6). First, the sample preparation steps of flush-314

ing the capillaries with He and sample gas through315

the input lines before the gasbench, which is con-316

ventionally done between two measurements us-317

ing a “pre-script”, is scheduled during and in par-318

allel with the IRMS analysis for the next measure-319

ment. A second important modification that helped320

to save time was achieved by switching to high He321

flow rates while purging the input line capillaries322

before and in the Gasbench as described above.323

Third, changes to the ISODAT NT software con-324

figuration for the Gasbench allowed for variably325

timed operations parallel to the measurement of326

the reference standards (multitasking) within the327

“chromatography” part of the measurement time328

schedule. Thus, the cryogenic trapping period329

could be varied within the chromatogram sequence330

without disturbing other tasks (e.g. the measure-331

ment of the reference standards) and was no longer332

required to be executed before the chromatogram333

(e.g. in a pre-script). The combination of the de-334

scribed modifications allowed to shorten the IRMS335

measurement time per sample and increased the336

number of samples analyzed per day.337

Isotope ratio analysis338

The carbon and oxygen isotopic composition ofthe CO2 is expressed as the relative difference ofits isotope abundance ratio relative to that of an in-ternational standard. This difference, usually ex-pressed in per mill, is defined as

δ 13C[�]V–PDB =[(13C/12C)Sample

(13C/12C)V–PDB−1

]·103 (1)

δ 18O[�]V–PDB–CO2 =[

(18O/16O)Sample

(18O/16O)V–PDB–CO2

−1]·103

(2)aThe effect should be < 0.06�V−1 for δ 13C for the reference standards, according to the vendor supplied instruction manual.

4

Page 5: Optimization of automated gas sample collection and IRMS analysis

Post-run off-line calculation and drift correc-339

tion for assigning the final δ 13C and δ 18O val-340

ues on the V-PDB (Vienna PeeDee Belemnite) and341

V-PDB-CO2 scaleb were done following the IT342

principle as described by Werner and Brand 21 .343

The δ 13C and δ 18O values of the laboratory air344

standards (Zurich CO2-in-air standards) were de-345

termined at the Max-Planck-Institut für Biogeo-346

chemie (MPI-BGC, Jena, Germany) according to347

Werner et al. 25 . The accurate assignment of the348

corresponding δ -values on the V-PDB and the V-349

PDB-CO2 scale was performed in Jena by mea-350

suring the Zurich CO2-in-air standards versus the351

“Jena-Reference AirSet” (J-RAS) as standard ref-352

erence material (SRM).e.g. 24 Any isotope ratio353

data presented in this article are reported in [�]354

deviation from V–PDB and V–PDB–CO2 for 13C355

and 18O, respectively. Typically several measure-356

ments of a laboratory reference standard are placed357

at the beginning and at the end of each measure-358

ment series, for post-calculation of corrections.359

Quality control (QC) standards are used to eval-360

uate this correction procedure.361

Results and discussion362

After ensuring linearity, we tested the effects of the363

variable cryogenic trapping period of our ASA–364

Gasbench–Isotope Ratio Mass Spectrometer setup365

on δ 13C measurements for a range of CO2 concen-366

trations and for different sample containers used367

(glass or metal). Furthermore, we tested the per-368

formance (precision, accuracy) of the δ 13C mea-369

surements for typical use of the described ASA–370

Gasbench–Isotope Ratio Mass Spectrometer setup.371

Linearity tests with gases of different CO2 in372

air or He mixing ratios have shown a strong re-373

lationship between the IRMS peak amplitude and374

the offset between the δ 13C (or δ 18O) of CO2 in375

a sample and its δ 13C (or δ 18O) reference value376

(Fig. 7). For δ 13C, this offset increases strongly377

with lower relative peak amplitude. We suppose378

the origin of this effect is the signal to noise ratio379

of the analysis. Thus, to ensure measurement inter-380

comparability, it is required to correct for this ef-381

fect, e.g. by optimizing the peak amplitudes of the382

samples to a limited range close to the amplitudes383

of the Isotope Ratio Mass Spectrometer reference384

gas. This optimization in effect means that the385

cryogenic trap should freeze the same amount of386

CO2 for each sample, independent of sample con-387

tainer and capillary volumes. Due to differences388

in pressure build-up as function of container vol-389

ume, the relationships between CO2 concentration390

of the sample and the trapping period required for391

a peak amplitude close to the reference have to be392

determined empirically (cf. Fig. 4, Table 1) and393

were thus tested by analyzing a broad range of di-394

lutions of a CO2 in air mixture with CO2 free air395

(Fig. 8, top panel). The resulting relative ampli-396

tudes for this range of diluted samples (Fig. 8, bot-397

tom panel) are between 85 and 125%, well within398

the typical variability of peak amplitudes (cf. Fig.399

7), reflecting the quality of the chosen fit function400

and the inaccuracy caused by the low time res-401

olution of the variable trapping period (Fig. 4).402

This last aspect is mostly defined by the 1 s time403

resolution of the Isotope Ratio Mass Spectrome-404

ter chromatogram procedure, for which the inaccu-405

racy increases at higher CO2 concentrations. For406

example, CO2 delivered by steel capillary for the407

concentration range [355,390] µmol mol−1 and408

[1365,1502] µmol mol−1 are represented by a 34 s409

and 20 s trapping period, respectively. If very410

high concentrations (> 5000 µmol mol−1) are ex-411

pected, the relative peak amplitude of the samples412

could be allowed to be > 100%. However, a new413

empirical relation would need to be determined414

for lower flow rates of sample through the cryo-415

genic trap or the Isotope Ratio Mass Spectrometer416

software for the variable trapping period (see Ap-417

pendix) would need to be changed to use a time418

resolution shorter than 1 s. In any case, applica-419

tion of the empirical relations of trapping period420

and sample concentration requires not only that the421

concentration needs to be known prior to IRMS422

analysis, but also that the CO2 concentration data423

from the IRGA (Fig. 1) needs to be collected and424

processed prior to the laboratory analysis.425

Since we modified the Isotope Ratio Mass426

Spectrometer setup substantially, we tested the per-427

formance (i.e. precision and accuracy) by mea-428

bThe virtual non-existing standard V-PDB is defined by adopting a δ 13C value of +1.95 � and a δ 18O value of –2.2 � forNBS 19 exactly. Via assigning these δ -values the hypothetical mineral V-PDB or rather the CO2 produced from it would be thestandard for δ 13C and δ 18O values. The term V-PDB-CO2 refers to the oxygen isotopic composition of the CO2 evolved fromthe mineral by reaction with water-free H3PO4 at 298 K. For details, see e.g. Ghosh et al. 24

5

Page 6: Optimization of automated gas sample collection and IRMS analysis

suring δ 13C (and δ 18O) of a laboratory reference429

standard and QC standards that passed through the430

sample flow path (Fig. 9) or were sampled by431

the ASA beforehand (Fig. 10). The overall pre-432

cision of δ 13C measurements was determined to433

be <0.08� (σ ) for samples with standards stored434

in glass flasks inside an ASA (N=33), <0.11�435

(σ ) for samples with standards stored in stainless436

steel loops inside an ASA (N=44) and <0.06�437

(σ ) for directly supplied standards (N=5), over the438

course of several measurement campaigns between439

February 2006 and March 2008. The decrease in440

precision with increasing sample numbers (N=5,33441

or 44) suggests that reference standards must be in-442

cluded in the sample sequence more frequently to443

correct for possible drift effects. The slight differ-444

ence between the QC standard I and II (Fig. 9) can445

be explained by methods to determine the respec-446

tive reference value. For the QC standard I the ref-447

erence value was determined by the ETH IsoLab448

based on an average difference (N=5) to the labo-449

ratory reference standard. For the second QC stan-450

dard on the other hand, the reference value was de-451

termined by an internationally acknowledged lab-452

oratory against several international standards with453

high precision. In general, based on periodic mea-454

surements of standards (mostly QC Standard I) us-455

ing the ASAs, the overall accuracy was determined456

to be 0.11±0.04� (σ ), reflecting measurements457

with the ASA–Isotope Ratio Mass Spectrometer458

setup during one year, i.e. March 2007 to March459

2008. For δ 13C, we did not find an effect of the460

surface properties (e.g. volume, surface:volume,461

surface material) of the stainless steel loop ver-462

sus the glass flask sample containers on the pre-463

cision and accuracy of stored samples. However,464

for δ 18O, the used stainless steel loop contain-465

ers appeared inadequate and would require exten-466

sive pre-treatment, such as the removal of resid-467

ual water from the surfaces that can otherwise ex-468

change oxygen atoms in an equilibrium reaction469

with CO2 and thus potentially change the stable470

isotope composition of the sample. A treatment471

with long periods of dry air flushing in combina-472

tion with heating, as suggested by Gemery et al. 16 ,473

makes steel loops far less practical for δ 18O mea-474

surements than their counterpart, i.e. glass flask475

as sample containers. We evaluated the reliability476

of the filling procedure of the ASAs and the influ-477

ence of transport of samples from the field to the478

laboratory by filling the same ASA in the field and479

sebsequently in the laboratory with the same stan-480

dard gas. The resulting δ 13C and δ 18O measure-481

ments showed no significant difference (∆δ 13C=-482

0.04�, ∆δ 18O=0.05�, N=5) between the respec-483

tive filling locations (Sophia Etzold, Institute of484

Plant Sciences, ETH Zurich, Switzerland, unpub-485

lished data).486

For studies using the Keeling plot approach,487

e.g. to determine the signature of the respiration488

source via a statistical regression approach, the489

optimization of peak amplitudes provides a clear490

and essential improvement for intercomparability491

of δ 13C measurements.8 Small (systematic) errors492

in δ 13C values would lead to much larger uncer-493

tainty in the determination of the respiration sig-494

nature (the intercept of the Keeling plot regres-495

sion line).4,11 Our results show that the system de-496

scribed here not only can provide precision of at497

least 0.1� with an accuracy of 0.11±0.04� (σ )498

for δ 13C, but also allows unattended operation in499

the field and retain a measurement time per sam-500

ple of 610 s. This clearly fullfills the quality crite-501

ria necessary to perform gradient measurements of502

stable isotope ratios of CO2 in air for the study of503

atmosphere–biosphere interactions.504

Acknowledgments505

Peter Plüss (ETH) and Patrick Flütsch (ETH) are506

kindly acknowledged for their extensive techni-507

cal support. We would like to thank Matthias508

Saurer (PSI), Willi A. Brand (MPI-BGC), Michael509

Rothe (MPI-BGC) and Sophia Etzold (ETH) for510

their advice and helpful discussions. Our work has511

also benefited from discussions with Peter Weigel512

and Andreas Hilkert (Thermo Fischer). This work513

has been supported by the Swiss National Science514

Foundation (SNF), grant 200021-105949.515

References516

1. IPCC. Climate Change 2007: The Physical517

Science Basis. Contribution of Working Group518

I Contribution to the Fourth Assessment Re-519

port of the Intergovernmental Panel on Cli-520

mate Change. Cambridge University Press521

2007. ISBN 0521705967.522

6

Page 7: Optimization of automated gas sample collection and IRMS analysis

2. Yakir D, Wang XF. Nature 1996; 380: 515–523

517.524

3. Bowling DR, Tans PP, Monson RK. Global525

Change Biology 2001; 7: 127–145.526

4. Pataki DE, Ehleringer JR, Flanagan LB,527

Yakir D, Bowling DR, Still CJ, Buchmann528

N, Kaplan JO, Berry JA. Global Bio-529

geochem. Cycles 2003; 17: 1022. DOI:530

10.1029/2001GB001850.531

5. Ogee J, Peylin P, Ciais P, Bariac T, Brunet532

Y, Berbigier P, Roche C, Richard P, Bar-533

doux G, Bonnefond JM. Global Bio-534

geochem. Cycles 2003; 17: 1070. DOI:535

10.1029/2002GB001995.536

6. Knohl A, Buchmann N. Global Biogeochem.537

Cycles 2005; 19: GB4008.538

7. Bowling DR, Pataki DE, Randerson JT. New539

Phytologist 2008; 178: 24–40. DOI:540

10.1111/j.1469-8137.2007.02342.x.541

8. Keeling CD. Geochim. Cosmochim. Acta542

1958; 13: 322–334.543

9. Keeling CD. Geochim. Cosmochim. Acta544

1961; 24: 277–298.545

10. Ogee J, Peylin P, Cuntz M, Bariac T, Brunet546

Y, Berbigier P, Richard P, Ciais P. Global Bio-547

geochem. Cycles 2004; 18: GB2019. DOI:548

10.1029/2003GB002166.549

11. Zobitz JM, Keener JP, Schnyder H, Bowling550

DR. Agric. For. Meteorol. 2006; 136: 56–75.551

12. Tuzson B, Zeeman MJ, Zahniser MS,552

Emmenegger L. Infrared Physics &553

Technology 2008; 51 : 198–206. DOI:554

10.1016/j.infrared.2007.05.006.555

13. Mohn J, Zeeman MJ, Werner RA, Eugster W,556

Emmenegger L. Isot. Environ. Health Stud.557

2008; (in press).558

14. Tuzson B, Mohn J, Zeeman MJ, Werner RA,559

Eugster W, Zahniser MS, Nelson DD, Mc-560

Manus JB, Emmenegger L. Applied Physics B.561

Lasers and Optics 2008; 92 : 451–458. DOI:562

10.1007/s00340-008-3085-4.563

15. Rothe M, Jordan A, Brand WA. In: World564

Meteorological Organization Global Watch565

World Meteorological Organization, 2005;566

161: 64–70.567

16. Gemery PA, Trolier M, White JWC. J. Geo-568

phys. Res.-Atmospheres 1996; 101: 14415–569

14420.570

17. Sturm P, Leuenberger M, Sirignano C, Neu-571

bert REM, Meijer HAJ, Langenfelds R, Brand572

WA, Tohjima Y. Journal Of Geophysical573

Research-Atmospheres 2004; 109: D04309.574

18. Knohl A, Werner RA, Geilmann H, Brand575

WA. Rapid Communications In Mass Spec-576

trometrys 2004; 18: 1663–1665. DOI:577

10.1002/rcm.1528.578

19. Theis DE, Saurer M, Blum H, Frossard579

E, Siegwolf RTW. Rapid Commun. Mass580

Spectrom. 2004; 18: 2106–2112. DOI:581

10.1002/rcm.1596.582

20. Clark MJR, Whitfield PH. Water Resources583

Bulletin 1994; 30: 1063–1079.584

21. Werner RA, Brand WA. Rapid Commun.585

Mass Spectrom. 2001; 15: 501–519. DOI:586

10.1002/rcm.258.587

22. Merritt DA, Brand WA, Hayes JM. Org.588

Geochem. 1994; 21: 573–583. DOI:589

doi:10.1016/0146-6380(94)90003-5.590

23. Werner RA, Bruch BA, Brand WA. Rapid591

Commun. Mass Spectrom. 1999; 13: 1237–592

1241.593

24. Ghosh P, Patecki M, Rothe M, Brand WA.594

Rapid Commun. Mass Spectrom. 2005; 19:595

1097–1119.596

25. Werner RA, Rothe M, Brand WA. Rapid Com-597

mun. Mass Spectrom. 2001; 15: 2152–2167.598

7

Page 8: Optimization of automated gas sample collection and IRMS analysis

Appendix599

The ISL script code used for multitasking the con-600

centration dependend activation of the cryogenic601

trap is described in the following example.602

1 s c r i p t TrapTimer603

2604

3 i n c l u d e " l i b \ s t d i s l . i s l " ;605

4 i n c l u d e " l i b \ I n s t r u m e n t . i s l " ;606

5 i n c l u d e " l i b \ GasBench_ l ib . i s l " ;607

6608

7 f u n c t i o n TrapTimerFun ( number A)609

8 {610

9 number B = (−10.393* l o g (A) ) −96;611

10 r e t u r n B ;612

11 }613

12 main ( )614

13 {615

14 number nA = _GetSequenceNumber ( "616

C o n c e n t r a t i o n " , 6 0 ) ;617

15 number nB = c a l l TrapTimerFun ( nA) ;618

16 number nC = 150 − nB ;619

17 number nD = _RefGe t P ro f i l eNumber ( "ASA620

" , " S t a r t " , 0 ) ;621

18 number nE = abs ( _GetTickCount ( ) − nD)622

;623

19624

20 i f ( nE > ( nC * 1000) )625

21 {626

22 _S e t ( " Gas Bench / Valco " ,LOAD) ;627

23 _ U s e r I n f o ( " CryoTrap a c t i v e ! " , 0 , 0 ) ;628

24 }629

25 }630

The sample concentration is read from the se-631

quence table and converted to a trapping period via632

an empirically fitted function (code line 14–15) as633

shown in Figure 4, from which a target start time634

in seconds is calculated (16). The elapsed time is635

calculated (17–18) and compared to the target start636

time in order to decide if the trap should be set to637

an activated state or not (20–24). For correct cal-638

culation of the elapsed time, the start time of the639

chromatogram process was stored from the inter-640

nal millisecond counter with the command code641

1 numbers nF = ( _GetTickCount ( ) ) ;642

2 _Re gSe tP ro f i l eNumb er ( "ASA" , " S t a r t " ,643

nF ) ;644

3 c a l l S t a r t c h r o m a t o g r a m ( ) ;645

within the Acquisition ISL script used for this646

method. Please note that the included lines (1–2)647

are added just before the chromatogram is started648

(3). For effective use of the script in the ISODAT649

NT acquisition software, it was incorporated into650

the Gasbench configuration as an ActionScript de-651

vice. The execution of the script code (or Action-652

script device) within the acquisition method was653

scheduled each second during a specific time win-654

dow as shown in Figure 6. Contrary to the ISL655

delay command for delayed operations, the pro-656

posed solution does not interfere with the timing657

of other tasks during the acquisition and proved658

to be a reliable and effective way of performing659

time variable tasks parallel to tasks with fixed tim-660

ing during the IRMS chromatogram part of a mea-661

surement. Figure 4 shows different regression fits.662

Based on this, we have used different fitting func-663

tions for different concentration ranges, or applied664

a look-up-table approach using if statements. We665

were not able to program the power function fits666

(y = a · xb) with decimal values of “b” in the ISL667

script, and have relied on Taylor or logarithmic668

functions instead.669

FIGURE CAPTIONS670

Figure 1: Overview of improved setup, from sam-671

pling air in the field (top) to measurement by Iso-672

tope Ratio Mass Spectrometer (bottom). Commu-673

nication connections and sample gas flow paths674

are indicated by broken lines and thick lines, re-675

spectively. After field operation, the ASA is trans-676

ported to the lab and interfaced with the Isotope677

Ratio Mass Spectrometer.678

Figure 2: Flow diagram of the ASA showing the679

difference between field (top) and laboratory (bot-680

tom) operation. The ASA is shown in field sam-681

pling mode: the flow is diverted by the solenoid682

valve, the air sample is dried, filtered and pushed683

through the Valvo-valve (shown here in bypass684

loop position). During sampling, an adjustable685

flow regulator (set to ≈ 0.9 L min−1) and a pop-686

pet check valve (one-way, opening at >7 kPa) help687

create a pressure excess in the glass flask or stain-688

less steel loop sample containers. Per ASA, three689

or four Valvo-valves with sample containers are690

connected in series, adding up to a maximum total691

of 33 or 44 samples per unit. Two manual three-692

way valves are used to switch between field and693

laboratory setup.694

Figure 3: Flow diagram of the laboratory setup.695

Shown here is the situation for flow of He (valve 1)696

with sample air from the ASA (2) to the Gasbench697

8

Page 9: Optimization of automated gas sample collection and IRMS analysis

(3) and subsequent cryogenic focus trap (4), while698

at the same time the Gas Chromatograph and the699

inlet of the Isotope Ratio Mass Spectrometer are700

flushed with He (4 and 5). See text for details.701

Figure 4: Empirical relations between sample CO2702

concentration and trapping periods required for703

peak amplitudes that are equal to the Isotope Ra-704

tio Mass Spectrometer reference gas, for samples705

delivered to the Isotope Ratio Mass Spectrome-706

ter from three different types of containers (glass707

flasks, steel loops, steel capillary). For glass flask708

sample containers, two fit functions are shown. For709

fit parameters and further details, see Table 1 and710

Appendix.711

Figure 5: Decision diagram for the automated liq-712

uid nitrogen (LN2) refilling system used for the713

cryogenic trap. Activation of the solenoid valve for714

LN2 flow into the Dewar depends on fulfillment of715

the conditions for the weight of the cryogenic trap716

Dewar (A), the temperature of the balance (B) and717

a signal from the Isotope Ratio Mass Spectrometer718

during a specific period of the IRMS measurement719

protocol (C) and a waiting period after filling.720

Figure 6: The timeline of the Isotope Ratio Mass721

Spectrometer measurement protocol implemented722

in the ISODAT NT software. The flow path of the723

input line and Gasbench is flushed with sample air724

(S) and the cryogenic trap is lowered into liquid725

nitrogen before sample air is let into the trap to be726

frozen (FS) for a period that is variable and derived727

from the sample CO2 mixing ratio. After the trap728

is raised to thaw, the cryogenically focused content729

is carried by He through the Gas Chromatograph to730

the Isotope Ratio Mass Spectrometer for analysis.731

Meanwhile, the sample inlet tubes are flushed with732

pure He (He) or the next sample (Snext) with higher733

flow rate (+).734

Figure 7: Linearity performance of δ 13C analysis735

using the modified Gasbench, expressed as the de-736

viation from the δ 13C reference value against the737

relative peak amplitude (A) in the chromatogram.738

See text for details.739

Figure 8: Application of concentration-dependent740

variable cryogenic trapping periods in the Gas-741

bench for measurements of δ 13C. In the top panel,742

the deviation from the δ 13C reference value is743

shown for different dilutions of a CO2 in air mix-744

ture with a constant δ 13C value. In the bottom745

panel, the deviation from the δ 13C reference value746

is shown for the corresponding relative peak am-747

plitude in the chromatogram. The SD for the δ 13C748

measurements is 0.04� (N=13).749

Figure 9: Deviation between measured δ 13C750

(δSample) and reference δ 13C (δRef) for a laboratory751

reference standard and two quality control (QC)752

standards that were directly suplied through steel753

capillaries, at different dates. The deviations of QC754

standard I (top) and QC standard II (bottom) are on755

average –0.02±0.04� and 0.06±0.05�, respec-756

tively. The maximum observed SD for each five-757

sample series of reference standard and QC stan-758

dards (I and II) is 0.05.759

Figure 10: The deviation between measured δ 13C760

(δSample) and reference δ 13C (δRef) for a laboratory761

reference standard and a quality control (QC) stan-762

dard that was sampled beforehand in glass flasks763

(3 min of ≈ 1 L min−1 flushing into an ASA).764

The maximum observed SD for each five-sample765

series of reference standard is 0.05 and 0.07 for766

δ 13C and δ 18O, respectively. For QC standard I,767

the average deviations from their reference value768

are 0.08±0.04� and –0.19±0.05� for δ 13C and769

δ 18O, respectively. The QC standard I is the same770

as shown in the top panel of Figure 9.771

Table 1: Fit parameters for the relationships be-772

tween sample CO2 concentrations and trapping pe-773

riods for different types of sample containers. See774

text and Figure 4 for details.775

9

Page 10: Optimization of automated gas sample collection and IRMS analysis

FIGURES AND TABLES776

777

Figure 1: Overview of improved setup, from sam-pling air in the field (top) to measurement by Iso-tope Ratio Mass Spectrometer (bottom). Commu-nication connections and sample gas flow pathsare indicated by broken lines and thick lines, re-spectively. After field operation, the ASA is trans-ported to the lab and interfaced with the IsotopeRatio Mass Spectrometer.

778

Figure 2: Flow diagram of the ASA showing thedifference between field (top) and laboratory (bot-tom) operation. The ASA is shown in field sam-pling mode: the flow is diverted by the solenoidvalve, the air sample is dried, filtered and pushedthrough the Valvo-valve (shown here in bypassloop position). During sampling, an adjustableflow regulator (set to ≈ 0.9 L min−1) and a pop-pet check valve (one-way, opening at >7 kPa) help

create a pressure excess in the glass flask or stain-less steel loop sample containers. Per ASA, threeor four Valvo-valves with sample containers areconnected in series, adding up to a maximum totalof 33 or 44 samples per unit. Two manual three-way valves are used to switch between field andlaboratory setup.

779

Figure 3: Flow diagram of the laboratory setup.Shown here is the situation for flow of He (valve 1)with sample air from the ASA (2) to the Gasbench(3) and subsequent cryogenic focus trap (4), whileat the same time the Gas Chromatograph and theinlet of the Isotope Ratio Mass Spectrometer areflushed with He (4 and 5). See text for details.

●●

●●

● ● ● ● ●

●● ●

●● ●

0

20

40

60

80

100

0

20

40

60

80

100

300 500 700 900 1100 1300 1500

CO2 concentration [µµmol mol−−1]

Ent

rapm

ent p

erio

d [s

]

Glass flasks (ASA); y = a1·x1+a2·x

2+a3·x3+b

Glass flasks (ASA), alt. fit; y = a·xb

Steel loops (ASA); y = a·ln(x)+bSteel capillary (Ref. inlet); y = a·ln(x)+b

780

Figure 4: Empirical relations between sample CO2concentration and trapping periods required forpeak amplitudes that are equal to the Isotope Ra-tio Mass Spectrometer reference gas, for samplesdelivered to the Isotope Ratio Mass Spectrome-ter from three different types of containers (glassflasks, steel loops, steel capillary). For glass flasksample containers, two fit functions are shown. Forfit parameters and further details, see Table 1 andAppendix.

10

Page 11: Optimization of automated gas sample collection and IRMS analysis

781

Figure 5: Decision diagram for the automated liq-uid nitrogen (LN2) refilling system used for thecryogenic trap. Activation of the solenoid valve forLN2 flow into the Dewar depends on fulfillment ofthe conditions for the weight of the cryogenic trapDewar (A), the temperature of the balance (B) anda signal from the Isotope Ratio Mass Spectrometerduring a specific period of the IRMS measurementprotocol (C) and a waiting period after filling.

782

Figure 6: The timeline of the Isotope Ratio MassSpectrometer measurement protocol implementedin the ISODAT NT software. The flow path of theinput line and Gasbench is flushed with sample air(S) and the cryogenic trap is lowered into liquidnitrogen before sample air is let into the trap to befrozen (FS) for a period that is variable and derivedfrom the sample CO2 mixing ratio. After the trapis raised to thaw, the cryogenically focused contentis carried by He through the Gas Chromatograph tothe Isotope Ratio Mass Spectrometer for analysis.Meanwhile, the sample inlet tubes are flushed withpure He (He) or the next sample (Snext) with higherflow rate (+).

11

Page 12: Optimization of automated gas sample collection and IRMS analysis

●●●

●●

● ●●●●●

●●

●●

●●●●

●●

●●

●●●●

−0.4

−0.2

0.0

0.2

0.4

−0.4

−0.2

0.0

0.2

0.4

0 50 100 200 400

(ASample / ARef) × 100 [%]

18O

● CO2 in air (400 µµmol mol−−1)CO2 in He (4000 µµmol mol−−1)

●●●

●●

●●●

●●●●

●●●●

●●

●●●●●●

●●●

●●

●●●

●●

●●

−0.2

0.0

0.2

0.4

−0.2

0.0

0.2

0.4

0 50 100 200 400

δδ Sam

ple

− δδ

Ref

[‰

]

(ASample / ARef) × 100 [%]

13C ● CO2 in air (400 µµmol mol−−1)CO2 in He (4000 µµmol mol−−1)

783

Figure 7: Linearity performance of δ 13C analysisusing the modified Gasbench, expressed as the de-viation from the δ 13C reference value against therelative peak amplitude (A) in the chromatogram.See text for details.

−0.2

0.0

0.2

−0.2

0.0

0.2

0 300 600 900 1200 1500 1800

CO2 concentration [µµmol mol−−1]

13C

2007−07−18

−0.2

0.0

0.2

−0.2

0.0

0.2

0 50 100 150 200

13C

2007−07−18

δδ S

ampl

e −

δδR

ef

[‰]

(ASample / ARef) *100 [%]784

Figure 8: Application of concentration-dependentvariable cryogenic trapping periods in the Gas-bench for measurements of δ 13C. In the top panel,the deviation from the δ 13C reference value isshown for different dilutions of a CO2 in air mix-ture with a constant δ 13C value. In the bottompanel, the deviation from the δ 13C reference valueis shown for the corresponding relative peak am-plitude in the chromatogram. The SD for the δ 13Cmeasurements is 0.04� (N=13).

●●●●●

●●●●

●●●

●●●

●●●●●

−0.2

0.0

0.2

−0.2

0.0

0.2

02:00 06:00 10:00 14:00 18:00

13C

2008−02−26 ● QC standard I Reference standard

●●●

●●●●●●●

●●

●●● ●

●●●●

●●●●●

●●●●● ●●●●

● ●●●●●

●●●

●●

−0.2

0.0

0.2

−0.2

0.0

0.2

16:00 20:00 00:00 04:00 08:00

13C

2008−03−13

δδS

ampl

e −

δδR

ef

[‰]

Time of day [h]

● QC standard II Reference standard

785

Figure 9: Deviation between measured δ 13C(δSample) and reference δ 13C (δRef) for a laboratoryreference standard and two quality control (QC)standards that were directly suplied through steelcapillaries, at different dates. The deviations of QCstandard I (top) and QC standard II (bottom) are onaverage –0.02±0.04� and 0.06±0.05�, respec-tively. The maximum observed SD for each five-sample series of reference standard and QC stan-dards (I and II) is 0.05.

●●●●●●

●●●●●●●●●●●●●●

●●

●●●●●●●●●●

−0.4

−0.2

0.0

0.2

0.4

−0.4

−0.2

0.0

0.2

0.4

16:00 18:00 20:00 22:00 00:00

13C

2008−03−12 ● QC standard I Reference standard

●●●

●●●●

●●●

●●●●●●

●●●●●●●●

●●●●●●●

●●

−0.4

−0.2

0.0

0.2

0.4

−0.4

−0.2

0.0

0.2

0.4

16:00 18:00 20:00 22:00 00:00

18O

2008−03−12

● QC standard I Reference standard

δδS

ampl

e −

δδR

ef

[‰]

Time of day [h]786

Figure 10: The deviation between measured δ 13C(δSample) and reference δ 13C (δRef) for a laboratoryreference standard and a quality control (QC) stan-dard that was sampled beforehand in glass flasks(3 min of ≈ 1 L min−1 flushing into an ASA).The maximum observed SD for each five-sampleseries of reference standard is 0.05 and 0.07 forδ 13C and δ 18O, respectively. For QC standard I,the average deviations from their reference valueare 0.08±0.04� and –0.19±0.05� for δ 13C andδ 18O, respectively. The QC standard I is the sameas shown in the top panel of Figure 9.

787

12

Page 13: Optimization of automated gas sample collection and IRMS analysis

Table 1: Fit parameters for the relationships between sample CO2 concentrations and trapping periods fordifferent types of sample containers. See text and Figure 4 for details.

Fit function Fit parametersGlass flasks (in ASA) y = a1 · x1 +a2 · x2 +a3 · x3 +b a1 =−3.71 ·10−1, a2 = 3.536 ·10−4,

a3 =−1.215 ·10−7, b = 174Steel loops (in ASA) y = a · ln(x)+b a =−15.37, b = 144

Steel capillary (Ref. & QC standards) y = a · ln(x)+b a =−10.39, b = 96

13