Anal. Chem. 1998, 70, 3656-3666
Measurement of Diffusive Flux of Ammonia from Water Zhang Genfa,† Tomoe Uehara,‡ Purnendu K. Dasgupta,*,† Antony D. Clarke,† and Wilfried Winiwarter§
Department of Chemistry and Biochemistry, Texas Tech University, Lubbock, Texas 79409-1061, Department of Oceanography, University of Hawaii at Manoa, Honolulu, Hawaii 96822, and Osterreichisches Forschungszentrum Seibersdorf Ges.m.b.H, A-2444 Seibersdorf, Austria
An instrument was developed for the measurement of gaseous ammonia concentration, NH3(sw,eq), in equilibrium with surface waters, notably ocean water. The instrument measures the ammonia flux from a flowing water surface under defined conditions and allows the calculation of NH3(sw,eq) from the principles of Fickian diffusion. The flux collector resembles a wetted parallel plate denuder previously developed for air sampling. The sample under study runs on one plate of the device; the ammonia released from the sample is collected by a slow flow of a receptor liquid on the other plate. The NH3 + NH4+ (hereinafter called NT) in the effluent receptor liquid is preconcentrated on a silica gel column and subsequently measured by a fluorometric flow injection analysis (FIA) system. With a 6-min cycle (4-min load, 2-min inject), the analytical system can measure down to 0.3 nM NT in the receptor liquid. Coupled with the flux collector, it is sufficiently sensitive to measure the ammonia flux from seawater. The instrument design is such that it is little affected by ambient ammonia. In both laboratory (NT 0.2-50 µM), and field investigations (NT 0.18-1.7 µM) good linearity between the ammonia flux and the NT concentration in seawater (spiked, synthetic, natural) was observed, although aged seawater, with depleted NT content, behaves in an unusual fashion upon NT addition, showing the existence of an “ammonia demand”. NH3(sw,eq) levels from ocean water measured in the Coconut Island Laboratory, HI, ranged from 6.6 to 33 nmol/m3 with an average of 17.4 ( 6.9 nmol/m3, in comparison to 2.821 nmol/m3 (average 10 ( 7 nmol/m3) NH3(sw,eq) values previously reported for the Central Pacific Ocean (Quinn, P. K.; et al. J. Geophys. Res. 1990, 95, 16405-16416). Ammonia is the dominant (and frequently the only detectable) gaseous base present in the atmosphere; its atmospheric presence has been known for nearly 200 years.1 The status of knowledge on ammonia as an environmental contaminant up to 1980 has been summarized in a monograph.2 Its basic nature and high water solubility leads to a very important role in atmospheric chemistry †
Texas Tech University. University of Hawaii at Manoa. § Osterreichisches Forschungszentrum Seibersdorf Ges.m.b.H. (1) de Saussure, Th. Recherches chimiques sur la ve´ ge´ tation; Nyon: Paris, 1804; Vol. V, (Facsimile edition, Gauthier-Villars: Paris, 1957). ‡
3656 Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
and physics and in effects on ecosystems at relatively low concentrations; in the intervening years, awareness about the importance of ammonia has only increased.3 NH3(g) can react with acidic aerosols or dissolve in atmospheric water associated with aerosols or in clouds. The consequent change in pH of aerosols or cloud droplets affects various atmospheric processes, notably the conversion of S(IV) to S(VI). Particles containing ammonium ion can act as condensation nuclei, affecting atmospheric visibility. The availability of NH3 affects the residence time, transport, and deposition of atmospheric sulfur and nitrogen. The deposition of atmospheric NT has a significant ecological effect. Nitrification of deposited NT results in acidification, while an excessive input results in eutrophication leading to a change in composition and distribution of vegetation types. Changes in prevalent plant types in many locations in Europe have been attributed to increased NT deposition. In Denmark, atmospheric N deposition is currently twice the N requirement based on the current growth rate of the extant plant species; forest ground flora in many parts of Europe has also shown shifts toward nitrogenfavoring species.3,4 On land, the major source of NH3 is animal waste decomposition, with lower contributions from fertilizer use. In 1987, Buijsman et al.5 estimated total European NH3 emission to be 6.4 Tg/yr, 81% from livestock and 17% from fertilizers. Other anthropogenic sources including coal combustion and auto exhaust contributed 5 gal) are protected from atmospheric NH3 intrusion by acid-form cation-exchange resin packed cartridges CF. All other symbols as in Figure 1.
10 holes (1/4-28 threaded on one block) are provided on each block, and bolts B hold the two halves together in a sealed manner. As shown in Figure 1c, both the donor and receptor plates seat in deep recesses (8.75 and 5.75 mm deep for the donor and the receptor plates, respectively) that is wide enough to snugly hold the plates. In both cases, the recesses are, however, longer than the glass plates themselves. Both plates are held in place at both the top and the bottom by a 1 × 7.5 cm retaining clip RC (held by recessed Nylon screws NS) except for the bottom of the donor plate. In the latter case, the donor plate ends 3 mm before the end of the recessed chamber in the frame, and a 2.5-cm-tall baffle BF, shaped and placed much like the retaining clips, ensures a smooth exit flow of seawater that is aspirated out through a 14-gauge (1.60 mm i.d., 2.11-mm o.d.) stainless steel tube ET cemented into a hole drilled at an angle into the Plexiglas frame. At the top of the donor plate frame, holes are drilled into the bottom of the Plexiglas frame to match with those already present on the donor plate and three 18-gauge (0.84-mm i.d., 1.27-mm o.d.) stainless steel tubes IT are cemented in place, their terminal ends being ∼0.5 mm below the glass plate surface. Using hot-melt adhesive to create a miniature 1-mm-thick spacer S, a microscope slide cover glass (10 × 50 × 0.25 mm) G is put in place as a baffle for the inlet seawater, to smooth the inlet flow and to prevent spray formation. On the receptor side, three holes are made at the top and one at the bottom in the frame to accommodate 18gauge stainless steel inlet/outlet tubes (RIT/OT). No baffles are necessary on the receptor plate because the flow rate is very low. The glass bead-epoxy layer adds a thickness of 1-1.5 mm on the receptor plate. The distance between the donor and the receptor surfaces under dry conditions is thus 7-7.5 mm. With actual water flow (the flow rate is large on the donor surface and the film thickness is not negligible), the distance between the emitting surfaces will range between 6 and 7 mm, depending on the donor flow rate. The active emitter surface area is 72 cm2.
Operation of the Flux Sampler. The operating arrangement of the flux sampler is shown in Figure 2. A three-way valve V allows the selection of DI water or a sample (seawater, synthetic standards) to be pumped by a chemically resistant centrifugal pump (P1, Little Giant Inc., model 1-AA-MD-2P037), through a needle valve flow controller N1 and a flowmeter R to the donor inlet tube triad IT. Poly(tetrafluoroethylene) (PTFE) tubes (4.7mm i.d., 5 gauge, standard wall, Zeus Industrial Products, Raritan, NJ) with Flex-tube connections (Cole-Parmer 6424-03) are used for connecting the donor liquid. The donor flow rate can be varied from 25 to 250 mL/min), and a flow rate of 150 mL/min was used unless otherwise stated. Flow is started with the frames separate and unassembled. The water runs down the cover glass and the surface of the plate to the exit aperture where it is aspirated by pump P2 (same as P1). Several things need to be ensured here. The plate surface needs to be evenly wetted by the water sample. If necessary, a glass rod is used to roll over the surface to assist the entire surface being wetted. Second, the outlet flow must be carefully balanced with respect to the inlet, maintaining a flow just slightly greater than the inlet flow. This prevents the exit baffle from being flooded (that would otherwise contaminate the receptor side). However, the suction rate must not be much larger than the input rate, else a significant negative pressure will be formed in the cell when the two halves are assembled. A bypass flow controlled by needle valve N2 is used to fine tune the aspiration flow. The system should be run for 1 h in an open frame configuration to make sure that the input/output flows are balanced. Three channels of a four-channel peristaltic pump (PP1, Minipuls 2, Gilson Medical Electronics) using 0.025-in.-i.d. pump tubes are used to pump water through the receptor inlet tube triad RIT through PTFE connecting tubes (22SW). The total inlet flow can be varied from 0.4 to 1.00 mL/min, but was maintained at 0.62 mL/min unless otherwise stated. The water passes over Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
3659
Figure 3. Analytical system schematic. Flow rates (in µL/min) and reaction coil dimensions (in mm) are shown in the figure. Other details are given in ref 24.
the glass bead-coated receptor plate. Initially, water from a wash bottle is used to wet the entire surface evenly. The plate is maintained in a vertical position, and at the bottom, aspiration is provided at tube OT by one channel of another four-channel peristaltic pump (PP2) using a 0.046-in.-i.d. pump tube at approximately the same rotation rate, the water running down to the outlet and aspirated at the outlet tube OT with another peristaltic pump. This provides ∼1% greater aspiration flow, relative to the inlet flow. As for the donor plate, this system is also run in the open frame configuration for 1 h to ensure that balanced flow is maintained. On the aspiration side, evenly spaced bubbles will appear in the effluent; this is normal. Once it is ensured that both sides are operating with balanced flow, the two halves are bolted together with liquid running in both halves while they are held in a vertical configuration. It is necessary that, during every experiment, the liquid on both plates of the device be kept running at all times. If the flow on one side is interrupted for an extended period, it is generally necessary to disassemble the device and reestablish the wetted surfaces. It is essential that all donor or receptor liquids (from standards to freshly deionized water) be protected from ammonia intrusion by incorporating H+-form cation-exchanger filters CF packed in a column in the vent line of the reservoirs. Such protection is also necessary for all reagents used in the analysis system. For DI water, small mixed-bed deionizing cartridges can also be incorporated in the line immediately before the flux collector. Analysis System. The ammonia analyzer and flux sampler is available commercially from AnalTech (Lubbock, TX). The analytical system is schematically shown in Figure 3. Briefly, the receptor effluent from the flux collector is aspirated by PP2 at 0.63 mL/min through a preconcentration column (1.93-mm-i.d. PTFE tube, 35 mm long), placed in the loop of a six-port electropneumatically operated injector (I, type 5020P, Rheodyne). Various types of cation-exchanger sorbents were packed in the preconcentration column; ∼30 mg of the sorbent was used in each column, using glass fiber filter frits (Whatman GF/B), 2 mm2) with nested PTFE tubes (22 gauge, 0.71-mm i.d., 1.07-mm o.d. inside a 17-gauge, 1.19-mm i.d., 1.78-mmm o.d. tube) to make terminal connections. The effective resin packed length is 2 cm; the column volume is ∼50 µL. Tested sorbents include the following: strong acid resins Dowex 50W×4 and Dowex 50W×1; anion suppressor resin (20-µm particle size, Dionex Corp., Sunnyvale, CA); weak acid resins: Amberlite CG 50, Amberlite 3660 Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
IRC-50 (a macroreticular resin), glass wool, glass fiber (from Whatman GF/B type filter), 0.3-mm-diameter glass beads, and silica gel (50-100 mesh, Sigma). Unless otherwise stated, the results pertain to the use of silica gel. When the valve is switched to the inject mode, 5 mM HCl flowing at 190 µL/min elutes the preconcentrated NH4+. Unless otherwise stated, the valve is continuously cycled between 4-min load and 2-min inject modes, using a CD-4S programmable timer (Chron-Trol, San Diego, CA). The rest of the analytical system is identical in the use of the reagents, hardware, and reactor dimensions to that previously described.28 Briefly, The eluite stream is mixed first with the OPA reagent and then with the sulfite reagent and allowed to react for ∼2 min at pH 11.0 and 85 °C. A fluorescent product is produced and measured by a fluorescence detector (λex, 365 nm; λem, 425 nm). All connections and reactors in the analysis system use 0.30-mm-i.d. PTFE tubing. RESULTS AND DISCUSSION Spatial Distribution of Ammonia within the Flux Collector and Overall Efficiency. The relative concentrations in the interior of the device were calculated by using a numerical method based on Fickian diffusion. From the known geometrical dimensions of the device and the total amount of ammonia collected by the receptor per unit time, such a model allows the computation of the NH3(sw,eq) at the donor surface. A three-dimensional grid was defined to model the concentration distribution. Within the grid, all cells corresponding to the emitter surface were set to a concentration of 1, and all cells of the receptor to 0; i.e., the receptor was assumed to be a perfect sink. This assumption is valid because at the concentrations involved in this study ammonia dissolved in the receptor will be essentially completely ionized (aided by any dissolved CO2 ). Indeed, for a given donor solution, using dilute acid solutions instead of DI water, there was no measurable increase in the collected NH4+. According to Fick’s law of diffusion
[
]
1 1 ∂c 1 ) D ∂2c 2 + 2 + 2 ∂t ∂x ∂y ∂z
(1)
where c is the concentration in gas phase, t is the time, D is the diffusion coefficient of NH3(g), and x, y, and z are Cartesian coordinates. At steady state, the concentration is invariant with time and eq 1 becomes
∂2c ∂2c ∂2c + + )0 ∂x2 ∂y2 ∂z2
(2)
or, expressed as concentration fluxes,
dc dc dc dy dz + dx dz + dx dy ) 0 dx dy dz
(3)
For the numerical method described here, eq 3 is approximated by differences instead of differentials. With appropriately small values of ∆x ) ∆y ) ∆z, the sum of the concentration differences in all directions of space is zero. For any given cell, the
f)
Figure 4. Calculated relative concentration contours in a horizontal cross section of the flux collector. Only half of the total length is shown. The relative concentration of NH3(g) is shown in increasing shades of gray in the following contours: 0-0.167, 0.168-0.333, 0.334-0.500, 0.501-0.667, 0.668, 0.933, and 0.934-1.000. Donor (bottom) and receptor (top) plates are displayed as black lines.
concentration may thus as well be expressed as the average of the concentrations of all neighboring cells in x, y, and z directions (total of six cells), and it is thus computed to be the average of the six neighboring cells. The cells belonging to the unwetted (inactive) areas of the receptor surface are considered totally inactive (not a sink). The numerical computations, carried out in Microsoft Excel, involve several hundred to several thousand iterations for each of the cells, depending on the size of the grid spacing chosen, until a steady state is observed. This computational approach closely resembles the actual attainment of diffusive equilibrium. Figure 4 shows the concentration profile obtained this way for a horizontal cross section of the flux collector. In the central part of the active surfaces, the concentration gradients are very homogeneous and reflect one-dimensional diffusion between donor and receptor surface. Near the edges, however, gradients become steeper, indicating enhanced mass transport in this area. Here mass transport is facilitated by the additionally available space adjacent to the active surfaces. This is not a computational artifact; it is well-known that diffusive flux is enhanced from a curved surface, for example, relative to a flat surface. The present computational procedure can accommodate any conceivable geometry, and relative concentrations are available for any grid point within the device. To relate the observed flux to the concentration next to the donor surface, we use ideal one-dimensional diffusion for scaling. Here it is possible to solve all equations analytically. The concentration changes between imaginary horizontal surfaces take place in a linear fashion; thus, at a given distance x from the donor, the concentration c1D is given by
c1D ) c0 x/d
(4)
where c0 is the concentration at the emitter surface and d is the distance between the surfaces. For the arbitrary scale used here, c0 was set to unity. For the three-dimensional grid, we may assume that, for a plane very close to the receptor surface, edge effects become negligible, and one-dimensional diffusion may be considered the only transport mechanism. The flux will then be proportional to the amount of material (mass) present in all grid cells neighboring the receptor surface. A factor can be derived by relating this mass to the mass that would be expected from one-dimensional diffusion. This factor describes the efficiency of diffusion at the given conditions with respect to one-dimensional diffusion. The overall efficiency factor f may now be calculated at a layer very close to the receptor as
∑c(∆x) /c 3
1DAE∆x
(5)
where AE is the area of the emitter surface and ∑c(∆x)3 is the sum of the mass in the individual cells near the receptor as described above, expressed as concentration times volume. To calculate a c1D from eq 4 for a layer very close to the receptor, an x value of (d - ∆x) is used. To achieve actual convergence of f by continually reducing ∆x was difficult in PC-based computation because of unacceptably large computational time and memory requirements. However, it was possible to compute the upper and lower limits of f. The lower limit was obtained neglecting any edge effects, while for the upper limit the cells within the same layer situated at the edges of the receptor area were also included. The lowest value for ∆x studied was 250 µm; under these conditions and the distance between the active surfaces being 6 mm, the lower and upper bounds of f could be computed to be 0.95 and 1.00 for the device described here. Supplementary information is given in an MS Excel spreadsheet that contains the complete calculation scheme. Each cell of the modeled grid is represented by a cell in the spreadsheet. Note that even for the relatively large value of ∆x ) 1000 µm, the file is ∼3 MB in size. Finally, the concentration in gas phase at the emitter surface can be calculated in a modified version of 1-D diffusion from the flux measured at the receptor:
csg ) Fd/Df
(6)
with F being the flux per unit area of the emitting surface, determined by dividing the amount of NH3 collected by the analytical system divided by the collection period and the donor area. The concentration csg is identical to the NH3(s,eq) insofar as the ammonia flux is concerned. Optimization of the Analysis Technique. Choice of the preconcentration sorbent. NT in the flux collector effluent is typically in low-nanomolar to subnanomolar concentrations. Preconcentration is necessary for reliable measurement. It is important to note that the optimum preconcentration sorbent not only should take up the NT in the sample but the collected NT must also be efficiently released during the elution step. For the efficient capture of NT as NH4+, it is necessary that, prior to loading the sample, the sorbent should be present in the H+ form. It is therefore convenient to use an acid as the eluent. However, all reagents are contaminated to some degree with NH4+; to keep the detector background low, it is necessary to limit the maximum concentrations of eluents/reagents used. Our experience indicated that it is desirable to keep the eluent concentration below 10 mM H+. If NH4+ is retained too strongly, it will not be efficiently eluted by this acid concentration. The results are shown in Table 1 in comparison with the empty column (the column volume being the effective loop volume in this case) in terms of relative peak height obtained with 5 mM HCl as eluent. There is no reason to believe that strong acid resins do not capture the NH4+ effectively, it appears therefore that the bound NH4+ is not efficiently eluted from these resins under this condition. Weak acid resins have a much greater affinity for H+. However, we Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
3661
Table 1. Performance of Different Sorbents as a Preconcentrator Phase preconcentrator phase
rel preconcn factor
empty PTFE column (control) strong acid resins Dowex 50W×4 Dowex 50W×1 Dionex anion suppressor resin (20 µm) weak acid resins Amberlite CG 50 Amberlite IRC-50 (macroreticular) siliceous sorbents glass wool glass fiber filter glass beads (300 µm) silica gel (50-100 mesh)
1.00 ( 0.07 0.42 ( 0.12 1.56 ( 0.02 1.48 ( 0.10 4.54 ( 0.26 3.04 ( 0.12 3.76 ( 0.18 4.24 ( 0.08 4.18 ( 0.20 5.84 ( 0.21
Figure 6. Liquid-phase analyzer system response. Triplicate peaks of 500 pM NT concentrated for 18 min each; note excellent reproducibility. House deionized water (DI water 1) shows measurable levels of NT, while effluent from a Nanopure water system sampled directly has NT levels lower than that in the carrier HCl.
Figure 5. Acid concentration in eluent needed to fully elute the collected NT.
found that both types of synthetic weak acid resins tested slowly lose effectiveness. It is well-known that glass and other siliceous materials also behave as very weak acid-type cation exchangers. The glass material worked effectively but the capacity was limited, limiting the upper dynamic range. Silica gel worked by far the best not only in giving a high effective preconcentration factor but also in exhibiting a high loading capacity, good linearity, and long-term stability (continuously used for more than two months). Silica gel was henceforth chosen as the preconcentration sorbent. Parameters Affecting Preconcentration. Eluent acid concentration for use with the chosen preconcentrator column was further optimized. A 400 nM solution of NH4Cl was preconcentrated for 4 min; the peak height response resulting from different concentrations of HCl as eluent is shown in Figure 5. The peak height increases sharply with the acid concentration used but levels over ∼5 mM; this concentration was henceforth used. During these experiments it also became evident that the purity of the acid eluent has a profound importance on the detector blank, especially at several millimolar acid concentrations. Of the high-purity acids we tested, HCl showed the least contamination. The column size was optimized using a 4-min preconcentration of 500 nM NH4Cl as sample; this is above the maximum 3662 Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
anticipated loading of the column in use. The height response increased by 50% from 18 to 25 mg of sorbent in the column, but a further increase in amount (up to 140 mg was studied) actually reduced the height response due to increased dispersion caused by increased bed dimensions. Preconcentration columns were henceforth made with ∼30 mg of sorbent. The volume of the sample preconcentrated is dependent on both the sample flow rate and the preconcentration period. The flow rate, which needs to be carefully balanced (vide supra), is obviously not an easily adjustable factor. In contrast, the sample volume is conveniently varied with the preconcentration period. With 100 nM NH4Cl as sample, the response was strictly linear over a preconcentration period of 2-20 min (linear r2 ) 0.9992). Linearity and Detection Limits. The system is capable of determining subnanomolar to low-nanomolar levels of NH4+ bearing samples, however, it is not feasible to handle such samples off-line without contaminating them. Such samples were therefore prepared on-line, diluting a higher concentration standard pumped at a low flow rate with a much higher flow rate of DI water (with an in-line deionizer bed) pumped by a second pump and aspirating the desired flow rate of the sample thus prepared through the preconcentration column while the rest is discarded to waste. Over a range of 0.5-750 nM, the response is linear (linear r2 ) 0.9976). For a sample concentration in the 2.5-700 nM range, the withinday relative standard deviation (rsd) ranged from 0.95 to 5.3%. However, over four disparate days, the overall rsd ranged from 3.8 to 15.4%. A variety of factors, including the stability of reagents and standards and the reproducibility of reaction temperature and pumping rate, can be involved in controlling reproducibility over different days. Considering the levels involved and the ease of contamination, this performance is respectable. Based on a S/N ) 3 criterion, the limit of detection (LOD) is better than 0.3 nM. As previously stated, the performance can be improved further by using longer preconcentration times. Figure 6 shows triplicate
Figure 7. Conductivity detector output showing (a) top, the occurrence of spray from the donor side, leading to spikes in the detector signal, and (b) bottom, elimination of spray by the inlet glass baffle.
injections of 0.5 nM samples, deionized water from the house deionizer system, and Barnstead Nanopure water directly preconcentrated, in each case for a preconcentration period of 18 min. The limits of detection under these conditions are better than 0.1 nM and, to our knowledge, represent the most sensitive measurements of liquid-phase NT or NH4+ demonstrated to date. Note that the best water sample actually produces a negative peak; this reflects the level of ammonium still present in the dilute, highpurity acid solution, prepared as carefully as possible, relative to that in the best water. Flux Receptor Surface. The flux collector used in this work is physically similar to parallel plate wet denuders previously described by Simon and Dasgupta.29 The wetted surfaces consist of high-silica porous glass, prepared in situ. Investigations with such wetted surfaces had been confined in the past to anionogenic acid gases such as SO2. With a cationogenic basic gas such as NH3, there can be a potentially integrating effect because of the cation-exchanger behavior of these surfaces and the poor ability of water to elute the collected NH4+. Surfaces composed of the above type of high-silica porous glass, silica, glass fiber mat, and glass beads were prepared in the form of wet denuders with water as the denuder liquid, a standard denuder test arrangement was set up with zero air flowing through the denuders, and the liquid water effluent from the plates was monitored with a conductivity detector. A pulse of SO2 or NH3 was injected into the sampled air and the rise/fall times of the detector signal was monitored. The (10 f 90%) rise time for both gases on all surfaces were rapid (∼1 min). With SO2, the (90 f 10%) fall times were also equally (or more) rapid on all the test surfaces. However, the fall times were g3 times slower for NH3 on all the surfaces, approaching 4 min for porous glass and glass fiber surfaces. Glass beads in comparison exhibited a fall time of 2.6 min and were chosen for further use. (29) Simon, P. K.; Dasgupta, P. K. Anal. Chem. 1993, 65, 1134.
Spray Generation in Flux Collector and Sample Flow Rate. On the donor side, seawater is run at a relatively high flow rate so that there is no appreciable depletion of NT content during its passage. It is essential that the water flows as a smooth film such that the actual surface area (due to turbulence) is not different from the geometric surface area used for the calculations. Even more importantly, it is vital that no sprays are generated from the donor side and carried over to the receptor surface. The flux of NH3(g) is so small relative to dissolved NT, a single microdroplet of seawater carried over to the slowly flowing receptor solution can render all measurements completely meaningless. In our experience, the seawater inlet baffle is of paramount importance in preventing such spray formation. Because the conductivities of seawater and DI water used for the receptor are vastly different, the occurrence of any spray formation is easily monitored by installing a flow-through low-volume conductivity detector in the line between the liquid output from the flux collector and the preconcentration column. With an appropriately designed cell, no contamination or additional dispersion is introduced in this process and some assurance is available on the quality of the data obtained. Figure 7 shows the conductivity detector output both with (a) the flux collector normally operating and (b) spray occurring (inlet baffle deliberately removed). The sampler can operate with donor flows between 25 and 800 mL/min; the occurrence of spray becomes more frequent above this range, and above 1500 mL/min, spraying cannot be prevented. The system can operate very safely without spray problems with donor flows in the range of 50-500 mL/ min. The effect of sample flow rate on the amount of NH4+ measured by the analytical system for tap water fed into the system was studied. Over a flow rate range of 25-200 mL/min, the measured flux increased linearly with flow rate: Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
3663
measured [NH4+] ) 62.05 + 0.086 × donor flow rate, mL/min (arbitrary units, r2 ) 0.9883) (7)
The effect is small and is believed to be due to an increase in the film thickness. This causes both a decrease in the diffusion distance and an increase in the total emitter surface area because the edges also act as a source. In addition, at higher flow rates, there may be greater undulation in the surface film; although we believe this effect to be minimal. The observed increase is not due to less relative depletion of NT at higher flow rates; as would be seen later from the data presented, even at 25 mL/min, the decrease in NT due to NH3 loss from the surface is minuscule. We operated the sampler with a donor flow of 150 mL/min, corrected the observed results (-20.8%) to extrapolated zero donor flow conditions, and used this with the planar geometric surface area of the emitter surface for calculating flux. Performance of Flux Collector. Response Times and Initial Calibration. For convenience, initial experiments were conducted with Lubbock municipal tap water as the donor. During our experiments, this sample typically contained 15 µM NT, with a pH of 7.9; the resulting flux was much higher than what would be expected with seawater. From the onset of the experiment, the signal rises; it was found that 30 min is required for the flux collector to fully equilibrate and produce stable signals. After reaching a stable plateau value, when the donor sample is switched to DI water, the signal starts decreasing immediately but requires ∼75 min to come back to original “zero” levels. These periods are dependent on the exact donor concentrations; for a pH 8 sample containing e5 µM NT, the rise and fall times are less than 30 and 60 min, respectively, when the donor solutions are alternated between deionized water and the above sample. One set of data from these experiments is used in the following as an illustrative example for ab initio system calibration. The sample contained 15.5 µM NT, as measured by direct fluorometric flow injection analysis using the present arrangement. The sample pH was measured to be 7.874, and the donor sample was used at the ambient laboratory temperature of 295 K. On the basis of the detailed analytical data provided by the city of Lubbock water treatment plant for this sample, we estimate ionic strength, I, for the sample of 0.025 M. The following computation is based on the work of Dasgupta and Dong.30 The Davies modification of the Debye-Hu¨ckel limiting law31,32 is used to compute the activity coefficient of the ammonium ion γNH4+:
-log γNH4+ ) [1.82 × 106(DT)-3/2xI/ {1 + (50.3RNH4+xI)/x(DT)}] - 0.1I (8)
where D is the dielectric constant of water at temperature T and RNH4+ is the ion size parameter of ammonium ion in angstroms, given by Kielland33 to be 2.5. Then the concentration of un-ionized (30) Dasgupta P. K.; Dong S. Atmos. Environ. 1986, 20, 565. (31) Laitinen, H. A.; Harris, W. E. Chemical Analysis, 2nd ed.; McGraw-Hill: New York, 1975; pp 9-16. (32) Christian, G. D. Analytical Chemistry, 5th ed.; Wiley: New York, 1994; p 139. (33) Kielland, J. J. Am. Chem. Soc. 1937, 59, 1675.
3664 Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
NH3 in the aqueous phase, NH3(aq), can be computed:
[NH3(aq)] ) γNH4+[N]TK/(γNH4+K + 10-pH)
(9)
where pH is the measured pH and K is the dissociation constant for NH4+ given in terms of activities to be34
K ) 10-(2729/T + 0.09018)
(10)
NH3(sw,eq) can now be calculated from the Henry’s law constant for ammonia, KH:
NH3(sw,eq) ) [NH3(aq)]/KH
(11)
Where NH3(sw,eq) is given in atmospheres and KH is given in units of molar per reciprocal atmosphere as
KH ) exp(4092/T - 9.70)
(12)
Before the NH3(sw,eq) value computed above can be used as c in eq 6, it should be transformed to units of g/cm3. From eqs 8-12, NH3(sw,eq) is computed to be 259 nmol/m3. From the experimentally observed flux per unit area, NH3(sw,eq) can be computed from eq 6. The diffusion coefficient of NH3(g) is 0.228 cm2/s at 298 K35 and is thus estimated to be 0.226 cm2/s at our operating temperature of 295 K. We therefore calculate the experimentally measured NH3(sw,eq) value to be 220-257 nmol/m3, depending on a diffusion distance of 6 or 7 mm and for a f value of 0.95. Considering the many uncertainties, this result is in excellent agreement with the ab initio calculations above and provides an absolute de facto calibration. For more accurate calibration, one of the NT-containing buffer compositions, for which NH3(sw,eq) has been measured directly by Dasgupta and Dong,30 should be used, but was not pursued in the present study. Negligible Loss of Ammonia from Donor. For the present approach to work meaningfully, the composition of the donor solution must not change significantly during passage through the donor. We have alluded to this previously, but it is useful to show this numerically as well. The example immediately above involving 15.5 µM NT analyte flowing through the donor side at 150 mL/min amounts to ∼40 nmol of NT/s. The measured total loss of ammonia to the receptor plate was ∼5 pmol/s, ∼0.01% of the total NT. The relative loss is independent of the exact NT content. The loss would be slightly greater for seawater, which has marginally higher pH, and when the experiment is conducted at a higher temperature. Nevertheless it will still be a negligible fraction of NT and would not result in any perceptible change in its composition as a result of passage through this device. Flux from Synthetic Ocean Water. Because of the considerable distance of Lubbock, TX, from the ocean, synthetic ocean water was used in the next set of experiments. A commercially available product, Instant Ocean, that claims to contain all necessary nutrients for marine life and is used to prepare synthetic ocean water for saltwater aquaria, was utilized. The ammonia flux (34) Emerson, K.; Russo, R. C.; Land, R. E.; Thurston, R. V. J. Fish Res. Board Can. 1975, 32, 2379. (35) Spiller, L. L. Anal. Lett. 1989, 22, 2561.
was measured with the sample as such and spiked with NH4Cl in an amount sufficient to increase [NT] by 1, 3, 5, and 6.5 µM. No change in pH was discernible as a result of NH4Cl addition. This standard addition experiment showed an increase in the ammonia flux linearly related to the spike concentration but also indicated that as such the sample contained 11.9 ( 3.9 µM NT. It does fall in the broad normal range of NT in seawater, 1-50 µM.36 Sapozhnikov37 reported NT in the surface layers of the Black Sea to be ∼0.2 µM increasing to ∼6 µM in deep layers; Quinn et al.10 also reported quite low levels, 0.4 ( 0.3 µM, for the surface waters of the remote oceans. However, in coastal regions, the levels can be much higher; Fukushi and Hiro38 reported concentrations of 50-210 µM for coastal Osaka Bay. Nevertheless, it is clear that tests relevant to background ocean water cannot be performed with synthetic ocean water prepared for aquaria. Experiments with High-Purity NaCl. Experiments showed that synthetic ocean water prepared from standard analytical grade NaCl also contains more NT relative to the remote oceans. Therefore ultrapure grade NaCl (99.999+%) was used to prepare water of salinity comparable to the oceans, 0.4 M NaCl, and used as the donor solution with and without spiking with NH4Cl. Over the range of 0-13.5 µM (five different spike concentrations) the observed flux was strictly linear with the spike amount (linear r2 ) 0.9970). Further experiments were conducted within a spiking range of 0-1.5 µM NT at five different NT spiking levels. These results also showed excellent linearity (linear r2 ) 0.9985). The observed NH3 flux from the high-purity 0.4 M NaCl sample was measurably different from the control blank observed from deionized water. Based on the standard addition data, the residual NT value computed for the NaCl solution is 50 nM, well within the purity specifications of the salt. As a matter of reference, the following relative signals were respectively observed with the blank (DI water), high-purity NaCl solution (0.4 M) and a batch of the synthetic ocean water: 0.0382 ( 0.0002, 0.0605 ( 0.0012; 1.24 ( 0.03. Experiments with Transported Seawater and at Field Sites. In December 1995, Seawater samples were collected 40 miles off the coast of Corpus Christi, TX, and shipped in scrupulously precleaned polypropylene carboys to our laboratories in Lubbock. The observed NH3(sw,eq) values were close to blank values and did not increase in the expected manner upon spiking with NT at least up to a level of 2 µM added NT. The experiment was repeated three months later, this time using coastal seawater from Corpus Christi, with the expectation that the NT levels and thence the observed NH3 potential flux will be higher. However, the experimental results were the same and in fact required even more NT addition before there was an observable increase in the flux. Note that this behavior is completely different from spiking experiments conducted with DI water or synthetic ocean waters prepared with any type of salt. The results are shown in Figure 8. In both cases, a minimum of four days elapsed between the collection of the sample and the analysis. Apparently this is sufficient for organisms to consume the NT, at least to a level where the observed potential flux is vastly reduced. We hypothesize that there are NH4+ binding sites present in the samples,
whether in the biological detritus or suspended siliceous matter. During deliberate addition of NT, the binding capacity of these sites must be first satisfied before a significant increase in the NH3 flux is observable. (Strictly, the onset of observable increase upon spiking and the slope of that increase are expected to be dependent on the individual binding capacities and the individual binding constants of different types of binding sites present.) If synthetic colloidal cation exchangers are deliberately put in a sample, the same type of behavior would be expected. Note that most types of cation exchange sites have much higher affinity for binding NH4+ than Na+; also, when an ion is present in very low concentrations, the apparent affinity of the exchanger for the trace ion increases even more.39 Note that if our hypothesis is correct, there can be serious errors in predicting NH3 flux based on measurements of NT; most wet chemical methods will measure both the bound and free NT. Clearly, samples should be analyzed immediately or preferably on-site in an in-line manner. To facilitate this, the instrument was set up in the oceanography laboratory at the University of Hawaii. NH3(sw,eq) values of 11 and 113 nmol/m3 were respectively measured for samples collected from Waikiki beach near Diamond Head and near the Waikiki Aquarium (on the southeast shore of Oahu). The system was then moved to the Hawaii Institute of Marine Biology on Coconut Island, Oahu, HI. The seawater characteristics, including the NT level at this site has been well studied in the past.40 The seawater intake is 150 ft from the edge of the island, 20 ft below the surface. The sample was directly pumped into the flux collector from the sampling pipeline connected to laboratory. In the five months immediately prior to the experiment, the NT concentration varied within the range of 0.18 ( 0.15 µM. The collector needed 20 min or longer
(36) Millero F. J.; Sohn M. Chemical Oceanography: Composition and stoichiometry of average seawater. CRC Press: Boca Raton, FL, 1992; pp 65-78. (37) Sapozhnikov, V. V. Oceanology 1990, 30, 39. (38) Fukushi K.; Hiro K. Talanta 1988, 35, 799.
(39) Helfferich, F. Ion Exchange; McGraw-Hill: New York, 1962; pp 151-160. (40) Uehara, T. Measurement of Potential Ammonia Flux with a Wetted-wall Gas Diffusion System. M.S. Thesis, University of Hawaii, Honolulu, HI, Dec, 1997.
Figure 8. Results of standard addition experiments conducted with several days old seawater samples from (a) Gulf of Mexico, 40 miles from the South Texas Coast, and (b) Corpus Christi, TX, coast.
Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
3665
Figure 9. Measured NH3(s,eq) vs NT values at Coconut Island, HI, over May-September, 1996. Samples were thermostated at 27 °C.
preconcentration times to see clear responses with this sample. A larger collector, similar in design but much larger in size (emitter active surface area 180 cm2, a fine mesh polymer screen cemented on the surface for wetting, instead of glass beads) was built and used with the same donor flow velocity (a donor flow of 600 mL/min was used); the receptor flow was maintained at 0.6 mL/min. In these experiments, the seawater was passed through a heat exchanger (heater/cooler) that thermostated the temperature of the seawater to 27 °C. The seawater NT value was measured by a Technicon Autoanalyzer II indophenol blue procedure41 in samples filtered through a glass fiber filter. The NT and the corresponding NH3 flux from the seawater was measured at this site from May to September 1996. The results are shown in Figure 9. The NT value and the observed NH3(sw,eq) values are well correlated (linear r ) 0.8734, n ) 21, p < 0.01) As the sample concentration changes from 0.3 to 1.7 µM, the NH3(sw,eq) value increases from 6.6 to 33 nmol/m3 with an average of 17.4 ( 6.9 nmol/m3. In comparison, Quinn et al.10 reported NH3(sw,eq) values of 2.8-21 nmol/m3, average 10 ( 7 nmol/m3, for the Central Pacific. The best fit slope for the data shown in Figure 9 is 16.7 nmol/ m3 NH3(sw,eq) per µM NT. On the other hand, if we proceed exactly as in the case for tap water with eqs 8-12 using the following parameters t ) 27 °C, I ) 0.714 (based on a measured salinity of 34.6 ppt; see ref 42), the calculated slope is 47.8 nmol/m3 NH3(s,eq) per µM NT, nearly a factor of 3 higher. Salting out of un-ionized (41) Walsh, T. H. Mar. Chem. 1989, 26, 295. (42) Lyman, J.; Fleming, R. H. Mar. Resour. 1940, 3, 134.
3666 Analytical Chemistry, Vol. 70, No. 17, September 1, 1998
NH3 has not been considered here; at this ionic strength, such considerations will further increase the predicted slope by another ∼6%. It is possible that some part of this disagreement between expectations and observations can be ascribed to various computational and experimental uncertainties. However, in view of the facts that (a) very good agreements can be obtained for tap water and (b) identically carried out activity calculations well predict the experimentally observed value of the dissociation constant of NH4+ in seawater (such data are presented in detail in ref 40 and were obtained by deliberately changing the pH of the donor seawater and measuring the change in the observed diffusive flux), we feel that such uncertainties are not large enough to account for the observed difference. As we have stated previously, part of the measured value of NT may be bound in a manner that is not available for direct participation in an equilibrium with gas-phase NH3. Whether or not that is true, it is quite clear that NH3(sw,eq) values computed on the basis of measured values of NT, pH, T, and I may not be reliable. We believe that we have presented here an instrument that is attractive in solving these problems. With the development of a reliable and affordable approach to measuring background levels of NH3, a major improvement toward experimental studies of oceanic ammonia fluxes will result. ACKNOWLEDGMENT Research at Texas Tech University and the University of Hawaii was supported by the Office of Naval Research, respectively through ONR Grant N00014-94-1-0295 and ONR Grant N0014-92-J-1388. However, this paper has not been subject to review by ONR and no endorsements should be inferred. We thank G. Geernaert for his enthusiasm and encouragement, F. Grima, Chemist, City of Lubbock Water Supply Laboratory, for the analytical data on the City of Lubbock municipal water samples, and C. B. Hilton, Hoechst-Celanese Corp., for the Gulf of Mexico samples. P. K. Quinn, of NOAA/PMEL is acknowledged for her detailed constructive criticism of the manuscript. SUPPORTING INFORMATION AVAILABLE A Microsoft Excel File (∼3 MB, can be supplied in zipped versions in 1.44 MB diskettes). This file serves to calculate the relative concentrations inside the flux collector in seven imaginary layers, each composed of 65 × 229 cells of 1 mm × 1 mm × 1 mm volume each. For ordering information, see any current masthead page. Received for review February 10, 1998. Accepted May 19, 1998. AC980144K
