A continuous record of atmospheric N2O measured from a tower in northern California captures strong pulses of N2O released by coastal upwelling events. The atmospheric record offers a unique, observation-based method for quantifying the coastal N2O source. A coastal upwelling model is developed and compared to the constraints imposed by the atmospheric record in the Pacific Northwest coastal region. The upwelling model is based on Ekman theory and driven by high-resolution wind and SST data and by relationships between subsurface N2O and temperature. A simplified version of the upwelling model is extended to the world's major eastern boundary regions to estimate a total coastal upwelling source of ∼0.2 ± >70% Tg N2O-N/yr. This flux represents ∼5% of the total ocean source, estimated here at ∼4 Tg N2O-N/yr using traditional gas-transfer methods, and is probably largely neglected in current N2O budgets.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Coastal areas are hypothesized to be a large source of the important atmospheric greenhouse gas, nitrous oxide (N2O). Recent studies have estimated that N2O emissions from coastal areas may account for 15–45% of the global oceanic N2O source, which in turn contributes ∼1/4 to 1/3 of the total (natural + anthropogenic) N2O source [Capone, 1991; Bange et al., 1996a; Prather et al., 2001; Seitzinger and Kroeze, 1998]. Seitzinger et al.  have estimated that anthropogenic pollution is responsible for more than 30% of coastal N2O emissions. Naqvi et al.  have suggested that one extreme pollution event alone off the southwest coast of India may have emitted as much as 0.39 Tg N (∼10% of the estimated annual oceanic total). Coastal N2O sources have been estimated by a wide variety of methods, which include extrapolating surface anomalies and coupling assumed N2O emission coefficients to microbial nitrogen cycling rates and/or fixed N inputs to coastal regions. The biologically based methods generally have not addressed the issue of how dissolved N2O is physically transferred to the atmosphere. The N2O fluxes predicted by all of the methods involve large uncertainties and are difficult to validate with direct observations.
 Coastal areas, which have been defined in various ways according to depth or proximity to land, account for approximately 10% of the total ocean area [Sverdrup et al., 1942]. Coastal upwelling regions, which are commonly found along eastern ocean boundaries, comprise a relatively small (<10%) subset of the total coastal area [Bange et al., 1996a]. Several considerations suggest that coastal upwelling regions are particularly strong sources of N2O to the atmosphere. First, since oceanic N2O is produced during microbial respiration only in subsurface waters and sediments, coastal upwelling provides a pathway for ventilating N2O that otherwise might remain trapped below the surface [Horrigan et al., 1981; Ward, 1986]. Second, due to surface nutrient recharge, coastal upwelling regions are characterized by high rates of primary production. This leads to large fluxes of sinking particulate organic matter that create conditions favorable for enhanced subsurface microbial N2O production [Epply and Peterson, 1979; Capone, 1991; Chavez et al., 1991; Codispoti et al., 1992].
 Coastal upwelling events take place on a timescale of hours to days. The episodic nature of these events makes the associated air-sea gas exchange difficult to measure using traditional oceanographic methods [Chavez et al., 1997; Service et al., 1998; Pennington and Chavez, 2000; McClain et al., 2002]. Gas exchange associated with coastal upwelling is also difficult to model because conventional air-sea flux calculations rely upon seasonal or annual mean climatologies of surface N2O, which may not resolve upwelling signals [Nevison et al., 1995; Bange et al., 2001]. Recently, Lueker et al.  reported a continuous record of atmospheric N2O from the coast of northern California that captures rises in N2O associated with upwelling events. These events occur in close correlation with dips in atmospheric δO2/N2 ratio, consistent with subsurface O2 depletion by microbial respiration and concurrent N2O production. The upwelling events were independently confirmed by elevated satellite wind stress data and drops in sea surface temperatures (SST) recorded by moored buoys. The coastal upwelling fluxes necessary to produce the observed variations in atmospheric N2O and O2 were inferred from a simple atmospheric flux model.
 In this paper, we extend the atmospheric flux model analysis of Lueker et al.  to quantify the total N2O flux associated with coastal upwelling from the Pacific Northwest between 41°N and 50°N. We then develop a more general coastal upwelling model, which is driven by satellite wind data, SST from a network of coastal buoys, and observed relationships between subsurface N2O and temperature. Although the upwelling model is intended for general use, current restrictions on the availability of SST input data limit its application to the Pacific Northwest coastal region. The results of the upwelling model and atmospheric flux model are compared to each other and to conventional gas transfer estimates based on the annual composite surface N2O climatology. Finally, the importance of coastal upwelling to the global oceanic N2O budget is estimated by extending a simplified version of the upwelling model to the world's other major coastal upwelling regions in the Arabian Sea and along the eastern boundaries of the Pacific and Atlantic Oceans.
2.1. Atmospheric N2O
 Atmospheric N2O samples are collected every 40 min from a 19-m-high U.S. Coast Guard radio tower at the top of Trinidad Head, California (41.05°N, 124.15°W, ground elevation 120 m) as part of the Advanced Global Atmospheric Gases Experiment (AGAGE) monitoring program [Prinn et al., 2000; Lueker et al., 2003]. The relative precision of the N2O data is better than 0.05% (corresponding to a dry air mole fraction of 0.15 ppb). ΔpN2O anomalies (rises in the atmospheric N2O mixing ratio above the baseline level) during upwelling events are typically on the order of 1–2 ppb (Figure 1).
2.2. Satellite Wind Speeds and Stresses
 Ocean zonal and meridional wind components u and v at 10-m height above the surface are obtained from a blended satellite/global weather center analysis data set. The data set is derived from spatial blending of high-resolution satellite data from the NASA Quick Scatterometer (QSCAT) mission and global weather analyses by the National Centers for Environmental Prediction (NCEP) [Chin et al., 1998; Milliff et al., 1999; Milliff and Morzel, 2001]. The data set covers the years 2000–2002. The spatial resolution is 0.5° × 0.5°, and the temporal resolution is 6 hours. Zonal and meridional wind stress components τx and τy are calculated from the QSCAT/NCEP u and v inputs using the formulae given by Milliff and Morzel . The scalar wind speed at 10-m-height U10 is also calculated from u and v.
2.3. Buoy SST Data
 Hourly sea surface temperature data for 2000–2002 are obtained from 13 moored buoys ranging from 35.10°N, 121.01°W to 47.34°N, 124.75°W along the Pacific Northwest coastline. The buoy data are provided online by the National Buoy Data Center (www.ndbc.noaa.gov). The hourly sea surface temperatures are averaged into 6-hour intervals that correspond to the time frame of the QSCAT/NCEP wind data.
2.4. Coastal Coordinates, Angle, and Length
 The angle θ of the coastline relative to the equator is estimated from the latitude/longitude coordinates in the Matlab coastal database (www.mathworks.com). The angle is calculated as the centered difference between the coordinates on either side of each coastal latitude/longitude pair
where ϕ = latitude angle, λ = longitude angle, and R = 6371 km is the radius of the Earth. The coarse-resolution length of coastline dL at any site is estimated as one half the distance between the adjacent coordinates on either side of that site,
Since the coordinates in the Matlab coastal data set are unevenly spaced, this method of calculating dL minimizes the influence of smallscale features such as bays and peninsulas, whose coastal angles differ from the general slope of the coastline. Values of dL range from 5 to 70 km.
2.5. Surface ΔpN2O
 A data set of more than 80,000 measurements of the surface N2O anomaly, ΔpN2O in natm, was collected between 1977 and 1996 on expeditions that spanned all world oceans (Figure 2) [Weiss et al., 1992; Butler et al., 1988, 1989]. The data are extrapolated to derive a global annual composite gridded 0.5° × 0.5° map of ΔpN2O, as described by Nevison et al. . Approximately 20,000 new measurements were added to the data set since the publication of Nevison et al. , including new data from the Pacific and Indian Oceans (WOCE Indian and Atlantic cruises) and the Blast I expedition in the eastern Pacific [Lobert et al., 1996]. The data set includes several expeditions that took place near coastal upwelling areas (Figures 3–4), although we do not know how soon after upwelling events these observations were made.
2.6. Additional O2 and SST Climatologies and Coastal Data Sets
 Global annual mean 1° × 1° gridded maps of O2 and apparent oxygen utilization (AOU) were obtained from the World Ocean Atlas [Locarnini et al., 2002]. Monthly mean 1° × 1° global sea surface temperature climatologies were obtained from Reynolds et al. . The Land Ocean Interactions in the Coastal Zone (LOICZ) global gridded 0.5° × 0.5° database was used to identify coastal grid cells and to estimate their oceanic versus land area fraction [Maxwell and Buddemeier, 2002; Smith et al., 2003].
3. Description of Models
3.1. Atmospheric Flux Model
 The atmospheric flux model is a simple Lagrangian model that estimates the air-sea fluxes necessary to produce the observed increases in atmospheric N2O [Lueker et al., 2003]. The flux into a well-mixed moving column of air is represented by [Jacob, 1999]
where ΔCt is the observed atmospheric anomaly, h is a vertical mixing column height, td is the e-folding lifetime for dilution of the air in the column, and Lt is the time-varying wind fetch north, i.e., upwind, of Trinidad Head. Lt is estimated using 6-hourly QSCAT/NCEP wind fields and is allowed to extend from 41°N as far as 50°N. Ut is estimated as the average QSCAT/NCEP wind speed over the range of the fetch Lt. The parameters h = 400 m and td = 24 hours are estimated as described by Lueker et al. . The inferred N2O flux Ft (in mol m−2 d−1) is evaluated three times daily over all periods in the 2000–2002 atmospheric record in which the observed atmospheric N2O anomaly ΔCt rises significantly above the baseline level, accompanied by a simultaneous dip in ΔCt for atmospheric δO2/N2. An upwelling event is assumed to be over when δO2/N2 returns to its baseline value [Lueker et al., 2003]. To estimate the total coastal upwelling N2O flux for the 41°N–50°N stretch of coastline, the inferred flux Ft is multiplied by the variable coastal length scale Lt and a fixed width scale of Rb = 40 km, where Rb is the estimated offshore extent of upwelling, roughly represented by the Rossby radius of deformation [Chelton et al., 1998]. The product FtLtRb is summed from March 1 to November 1 (the months of peak upwelling events) to obtain integrated total air-sea fluxes for individual months and over the total 8-month upwelling season (Table 1). The fluxes Ft vary by ∼20% when the dilution time td, the least constrained parameter in equation (3), is reduced to 12 hours. In addition, the summed fluxes are directly proportional and thus highly sensitive to the choice of Rb.
Table 1. Coastal Upwelling Fluxes Estimated for the Northeast Pacific Estimated by 3 Different Modelsa
Atmospheric Flux Model 41°N–50°N
Wind-SST Upwelling Model 41°N–50°N (35°N–50°N)
Wind-Only Upwelling Model 41°N–50°N (35°N–50°N)
Values are total moles N2O × 107 emitted in given time period.
March 1 to Nov. 1 2000
March 1 to Nov. 1 2001
March 1 to Nov. 1 2002
3.2. Coastal Upwelling Model
3.2.1. Wind-SST Model
 Since high precision atmospheric N2O time series are not available in other coastal upwelling regions, a more general coastal upwelling model was developed that is driven primarily by wind and sea surface temperature. Although designed for general use, the application of the model is currently limited by the lack of SST data of sufficient precision and temporal resolution to detect coastal upwelling signals. Our analysis at present focuses on the Pacific Northwest coastal region from 35°N–50°N, encompassing and extending southward of the region studied by Lueker et al. . The region is described by 88 coastal coordinates (including Vancouver Island) in the Matlab database. Gridded QSCAT/NCEP wind stresses and SST data from the 13 NDBC buoys were matched to the 88 coastal coordinates. All model calculations described below were performed every 6 hours for the period 2000–2002.
 The model assumes that the primary mechanism for coastal upwelling is offshore mass transport driven by surface winds. Winds blowing parallel to the coastline cause alongshore wind stress, which creates Ekman transport of water 90° to the right in the Northern Hemisphere (Figure 5). The water transported offshore is replaced by upwelling of deep waters that flow up against the physical barrier of the coastline [Bakun, 1973; Summerhayes et al., 1994; Schwing et al., 1996]. In the Southern Hemisphere, offshore transport is 90° to the left of the wind stress vector.
 Ekman offshore transport associated with alongshore wind stress is estimated by
where M is the mass transport in m2/s, τp is the wind stress parallel to the coastline in N m−2, ρw = 1025 kg m−3 is the density of sea water, and f is the Coriolis force in s−1. M is multiplied by the coastal length dimension dL to estimate the total volume of water displaced offshore,
where dL is in meters and V is in m3 s−1. An integer 0.5° × 0.5° land/ocean database is used to determine whether the mass transport is directed offshore (to the right in the Northern Hemisphere and to the left in the Southern Hemisphere) or onshore. In the former case, the model assumes that the volume of water transported offshore is replaced by, i.e., is equal to, the volume upwelled. In the latter case, upwelling is assumed to be zero. The coastal upwelling velocity w (in m/s) is given by
 The N2O mass flux brought to the surface by upwelling, and assumed to be entirely ventilated, is calculated as
where T is the sea surface temperature at the moored buoy closest to the coastal coordinate and N2O(T) (in nM) is a linear fit of subsurface N2O to temperature [Lueker et al., 2003, Figure 3].
The fit is based on a transect of profiles collected off Monterey in August 1990 [Codispoti et al., 1992], and two profiles collected off the Oregon coast in June 1998 [Vollmer and Weiss, 2002; Lueker et al., 2003]. N2Osat is assumed to be the observed baseline atmospheric value recorded at Trinidad (316–317 natm) for the time period studied. Partial pressure units are converted to nM using the N2O solubility coefficient of Weiss and Price .
 It is important to note that the fit between temperature and N2O production (equation (8)), does not reflect a true mechanistic relationship. Temperature more likely serves as a proxy for the gradient in N2O production with depth. N2O production occurs mainly as a byproduct of microbial nitrification in the North Pacific [Dore et al., 1998; Ostrom et al., 2000]. Nitrifying bacteria are inhibited by light in the euphotic zone, and yield a greater fraction of N2O relative to NO3−, the normal end product of nitrification, as O2 becomes increasingly depleted with depth. O2 rather than temperature is the primary factor that determines the microbial N2O yield [Goreau et al., 1980; Nevison et al., 2003]. Other oceanic regions have different oxygen profiles, and the additional microbial process of denitrification may produce and/or consume N2O in some regions [Cohen and Gordon, 1978; Law and Owens, 1990; Naqvi et al., 1998]. In addition, N2O production and consumption in continental shelf sediments may influence the amount of N2O released to overlying water [Seitzinger and Kroeze, 1998; Naqvi et al., 1998, 2000]. For these reasons, equation (8) cannot be applied universally. New, regionally specific relationships between dissolved N2O concentration and temperature must be developed and high-resolution SST data obtained before our wind-SST upwelling model can be extended beyond the Pacific Northwest.
3.2.2. Wind-Only Model
 For the present study, we extend the estimate of the N2O coastal upwelling flux beyond the Pacific Northwest using a simplified version of the upwelling model that is driven primarily by satellite wind data. The simplified model is designed to account for the location, frequency, and relative magnitude of coastal upwelling, as well as the concentration of excess N2O in upwelled subsurface waters in regions of varying oxygen content. The upwelling velocity w is calculated by equation (6) as described above. At any given time step, wmax is defined as the maximum w that has occurred within the last nback time steps, i.e., at time tmax. If wmax exceeds a critical value wcrit, surface ΔN2O is estimated as
where ΔN2O100m is the excess N2O (above saturation) at 100 m depth, which is assumed to be brought to the surface by upwelling. ΔN2O100m is estimated as a function of O2 and AOU [Locarnini et al., 2002] using the parameterization of Nevison et al. , in which ΔN2O increases nonlinearly with increasing O2 depletion. Qw and Qt are simple scaling functions that adjust the magnitude of the surface anomaly to the strength of upwelling and allow for exponential decay of excess N2O as a function of time since upwelling occurred,
We use default values of wpeak = 2 × 10−4 m/s (the highest rate of upwelling calculated for the Pacific Northwest), wcrit = 1 × 10−5 m/s, and nback = 12 (3 days), based on the persistence of elevated atmospheric N2O for several days after cold SST anomalies have returned to normal values in the Trinidad time series [Lueker et al., 2003].
 The N2O air-sea flux is calculated as
where k is the air-sea transfer velocity calculated as a quadratic function of wind speed and adjusted for N2O using Schmidt number scaling [Wanninkhof, 1992]. On all days in which wmax < wcrit, ΔN2O is assumed to be zero. The wind-only model is applied to the six major world coastal upwelling regions described in Table 2. All calculations are performed at the coastal coordinates of the 0.5° × 0.5° gridded LOICZ coastal database, using coastal slopes derived from the Matlab coastal database and 6-hourly QSCAT/NCEP winds. In low-latitude regions where the Rossby radius of deformation Rb exceeds the width of the 0.5° coastal grid, is assumed to extend up to 100 km offshore and is scaled up by as much as a factor of 2 [Huyer, 1983; Codispoti et al., 1989].
Table 2. Annual N2O Fluxes in 2000: Wind-Only Upwelling Model Compared to Gas Transfer Calculations Using Annual Composite ΔN2O Climatology (N95 Method)
 A second mechanism for coastal upwelling occurs when the wind stress curl causes divergent Ekman transport that is replaced by upwelling of deep water. Upwelling by this mechanism, known as Ekman pumping, can occur regardless of the orientation of the wind stress vector to the coastline. However, wind stress often decays toward land, creating an additive offshore divergence that enhances upwelling due to alongshore wind stress. Wind stress curl divergence due to decaying winds typically creates maximum upwelling within 200–300 km of the coast and commonly accounts for only a few percent of the upwelling driven by alongshore wind stress [Bakun and Nelson, 1991; Summerhayes et al., 1994; Smith, 1994; Milliff and Morzel, 2001]. Some exceptions occur, for example, wind stress curl upwelling may be comparable in magnitude to alongshore wind stress upwelling in the Arabian Sea during monsoon season [Brock and McClain, 1992; Smith, 1994; Bange et al., 1996b; Lendt et al., 1999]. In addition, recent work with high-resolution (9 km) modeled wind stress data suggests that finescale wind stress curl can account for much of the upwelling in the California Current that has been traditionally attributed to alongshore wind stress [Pickett and Paduan, 2003].
 We tested a variant of the wind-only upwelling model that also calculates the upwelling velocity due to wind stress curl,
Since the two mechanisms for coastal upwelling are additive, wwc is added to the alongshore wind stress w (equation (6)) to determine wmax. The model neglects other mechanisms, such as thermocline sloping and continental shelf uplift, which may also contribute significantly to bringing deep water close or up to the surface [Summerhayes et al., 1994; Roughan and Middleton, 2002].
3.3. Surface N2O Climatology-Gas Transfer Model
 Conventional gas transfer calculations based on the annual composite surface N2O climatology were performed using a methodology similar to that described by Nevison et al.  (hereinafter referred to as N95). The N2O flux is calculated as
where S is the N2O solubility coefficient [Weiss and Price, 1980], which converts the global gridded ΔpN2O map to ΔN2O, and k is the air-sea transfer velocity. S was calculated using the monthly mean sea surface temperature climatology of Reynolds et al. . The air-sea transfer velocity k was calculated with the formula of Wanninkhof , using 6-hourly QSCAT/NCEP wind speeds for the year 2000. The alternative air-sea transfer velocity formula of Nightingale et al.  was also tested. Calculations were performed for the entire ocean and for the specific coastal upwelling regions defined by the LOICZ 0.5° × 0.5° gridded database (Table 2).
4. Results and Discussion
4.1. Observed Surface N2O Anomalies
 The observed surface N2O anomalies from the expedition database provide some useful checks on the underlying theory of the coastal upwelling model. The available observations near the Pacific Northwest coastline show that ΔpN2O increases sharply with decreasing distance from the coast. The highest anomalies, of up to 11 nM, are found at a radius of <50 km from shore, consistent with our assumed Rossby radius of deformation of 40 km (Figures 3a–3c). ΔN2O tends to increase with decreasing sea surface temperature (Figures 3d and 3f), suggesting that excess N2O has been brought to the surface by cold upwelling water. In one transect off Oregon at 44°N, elevated ΔN2O is found in relatively warm water and does not appear well correlated to SST (Figure 3e). This observation suggests that the excess N2O brought to the surface during an upwelling event may persist after surface waters have returned to warmer temperatures.
 Plots of observed ΔN2O versus distance from shore in other upwelling regions also tend to increase with decreasing distance from shore (Figure 4). In the lower latitude eastern boundary sites (Mexico, 18°N, Senegal, 14°N, Panama, 7°N), maximum ΔN2O anomalies range from 4 to 9 nM and tend to occur farther from shore (up to 50–100 km) than in the Pacific Northwest, consistent with the theoretical increase in Rb with decreasing latitude. At the coastal sites near Oman on the Arabian peninsula (19° and 22°N) and near southwest India (8°N), ΔN2O does not increase as distinctly toward the coast as in other transects, and maximum values are found as far offshore as 150 km. This suggests that ocean-surface divergence driven by wind stress curl, which tends to be maximum farther from shore than alongshore wind stress-driven upwelling, may be a significant mechanism for bringing excess N2O to the surface in the Arabian Sea.
4.2. Comparison of Atmospheric Flux Model to Upwelling Models in the Pacific Northwest
 The Pacific Northwest is a well-known region of coastal upwelling [Bakun, 1973, 1975; Huyer, 1983; Schwing et al., 1996; Huyer et al., 1998]. Winds blowing from the north to northwest create strong equatorward wind stress that runs largely parallel to the coastline (Figure 6a). This creates an westward offshore Ekman transport that is replaced by upwelling waters. A particularly strong upwelling event occurred in mid-June 2000. This event led to the longest sustained rise in atmospheric N2O observed in the Trinidad Head record (Figure 1) and drove upwelling velocities as high as 2 × 10−4 m/s (Figure 6b). The region of strongest upwelling was centered around 41°N, coincident with the latitude of the Trinidad Head station. In comparison, the upwelling event 2 weeks earlier at the end of May 2000 (Figure 1) had a smaller maximum upwelling velocity of 1.2 × 10−4 m/s, according to the upwelling model, but the area of strong upwelling extended farther down the coast to ∼35°N.
 The wind-SST upwelling model, which is driven primarily by satellite winds and buoy SST records, and the atmospheric flux model, which is driven primarily by observed atmospheric N2O, predict N2O upwelling fluxes of similar timing and magnitude for the mid-June 2000 event (Table 1, Figure 7). Both models predict three main peaks in the N2O flux, with the atmosphere flux model peaks lagging the upwelling model peaks by 1–2 days. The small lag most likely results from the upwelling model's assumption of instant transfer of subsurface N2O to the atmosphere, while the atmospheric flux model more realistically captures the delay between the onset of strong equatorward wind stress and the subsequent upwelling and ventilation of N2O. Inclusive in this delay is the thermal N2O flux associated with the decrease in N2Osat as cold upwelled water warms at the surface. N2Osat decreases by 1.1 nM as SST warms from 9° to 12°C, for example. Since excess N2O may still exist in surface waters after warming, the upwelling model (equation (7)) tends to underestimate the ventilation flux that may occur several days after strong wind stress.
 The atmospheric flux model predicts a considerably weaker flux than the wind-SST upwelling model for the upwelling event of May 31, 2000 (Figure 7), especially when the upwelling model domain is expanded southward to 35°N (Table 1). A closer inspection of the May 31 event reveals that it is centered farther to the south of the Trinidad Head station than the mid-June event (Figure 8). Since the prevailing summer winds in the Pacific Northwest are north to northwesterly, Trinidad Head “sees” only the cumulative N2O flux that is emitted at or north of the station and thus will not tend to register upwelling events that occur south of 41°N. However, the high N2O supersaturations observed at 34°N–35°N in June 1991 (Figures 3a and 3d) provide general evidence that coastal upwelling can bring a large mass flux of N2O to the surface at these more southerly latitudes.
 Variations in the latitudinal distribution of the N2O flux may also explain some of the differences in the interannual variability of the total upwelling season (March 1 to November 1) source predicted by the two models. The source predicted by the atmospheric flux model is relatively similar in 2000 and 2001, but triples in 2002 (Table 1). The wind-SST upwelling model also predicts the largest total N2O flux in 2002 from the 41°N–50°N latitude band. However, when the range is expanded to 35°N–50°N, the differences between 2000, 2001, and 2002 are reduced by ∼50% (Table 1). A breakdown of the upwelling model flux by latitude suggests that in 2000 and 2001 more than half the seasonal flux occurred south of 41°N, while in 2002 a larger fraction of the flux was concentrated to the north of the Trinidad Head tower.
 A major uncertainty in the atmospheric flux model summation calculation is that the total flux estimates are directly proportional to the assumed width scale of coastal upwelling, i.e., the Rossby radius of deformation Rb. In contrast, the upwelling model results are independent of Rb, since Rb is only implicitly included in the calculation of the volume flux (equations (4) and (5)). The generally good agreement between the magnitude of the total N2O flux calculated by the atmospheric flux and upwelling models for the Pacific Northwest (Table 1) suggests that our choice of Rb = 40 km is reasonable. This choice is also supported by physical oceanography theory [Chelton et al., 1998; Milliff and Morzel, 2001] and by the observed sharp increase in surface N2O within ∼40 km of the coast (Figure 3).
 Another uncertainty in the atmospheric flux model analysis lies in the definition of the period of an upwelling event. Observations at Trinidad Head show that elevated atmospheric N2O can persist for days after atmospheric O2, wind, and SST data indicate that upwelling has ceased (Figure 1), suggesting that excess N2O is still being ventilated from the surface ocean [Lueker et al., 2003]. Since the atmospheric flux model analysis only considers flux during active upwelling events, it may neglect significant post-upwelling N2O emissions. Furthermore, since N2O is a very long-lived (atmospheric lifetime ∼120 years [Prather et al., 2001]) and well-mixed trace gas, strong pulses are required to drive the 1–2 ppb atmospheric anomalies observed in the Trinidad Head record. Even after atmospheric N2O returns to its baseline level of ∼317 ppb, coastal areas may continue to contribute a “background” flux, which will not be included in the atmospheric flux model analysis.
 The wind-SST upwelling model suffers from uncertainties associated with its assumption of instantaneous and complete transfer of upwelled N2O to the atmosphere. This assumption ignores the possibility that upwelled water may be subducted, become thermally stratified, or be recirculated below the mixed layer before N2O can fully exchange with the atmosphere. Air-sea transfer depends largely on wind speed and full equilibration can take up to several weeks [Wanninkhof, 1992]. For this reason, the N2O mass flux predicted by the upwelling model may tend to represent an upper limit on the actual amount of N2O emitted to the atmosphere. In addition, like the atmospheric flux model, the wind-SST model may tend to improperly concentrate the N2O flux into a brief upwelling period, since an excess of upwelled N2O may remain in the surface mixed layer for days or even weeks after the cessation of the upwelling event and may be ventilated only slowly to the atmosphere.
 The wind-only upwelling model predicts monthly and total upwelling-season N2O fluxes that generally agree in magnitude with the atmospheric flux and wind-SST upwelling models (Table 1). The wind-only model also does a reasonable job of reproducing the timing of the main upwelling events in May–June 2000 (Figure 7). It captures the magnitude of the May 31 peak and the first of the three mid-June peaks, although it tends to underestimate the two subsequent mid-June peaks, which are driven in large part by the observed dips in SST at coastal buoys.
 The wind-only upwelling model predicts a significant upwelling flux from the latitudes between 46°N and 50°N, which is lacking in the wind-SST model. Upwelling at these latitudes is known to occur in summer [Huyer, 1983; Schwing et al., 1996] and is suggested in the current study both by parallel alongshore wind stress (Figure 6a) and high observed N2O anomalies in the surface waters (Figures 3c and 3f). The failure of the wind-SST model to reproduce the flux at these latitudes may be due to the sensitivity of the model to a single buoy at 47.34°N, which is the nearest buoy to all coastal grids north of 47°N. This buoy is located somewhat farther from shore (40 km) than most of the other 12 buoys, at a distance corresponding to the outer edge of the Rossby radius of deformation. Consequently, the buoy may not observe the sharp drops in SST below 13°C that are necessary to drive a significant N2O flux (equations (7) and (8)). The comparison of the two variants of the upwelling model demonstrates that incorporation of observed SST can lead to significant improvement over the wind-only approach, in terms of agreement with the observational constraint provided by the Trinidad Head record (as interpreted by the atmospheric flux model). On the other hand, the wind-only approach may yield more accurate results for coastal areas where SST input data are unreliable or poorly matched.
 The wind-only upwelling model offers a significant improvement over the N95 method, which involves coupling gas transfer coefficients to the gridded and extrapolated surface ΔN2O climatology. Since N95 assumes a constant annual value of ΔN2O, seasonality is driven only by variation in the air-sea transfer velocity. The latter is a scalar function of wind speed and does not consider the orientation of the wind to the coastline. As a result, the N95 method predicts similar fluxes on June 11, 2000 (day 162), when winds were strong but poleward, such that Ekman transport was onshore, as on June 14 (day 166), when winds were both strong and equatorward, such that Ekman transport was offshore (Figure 7). As discussed below, the N95 method may also poorly represent the spatial distribution of the coastal N2O flux.
4.3. Extension of Wind-Only Upwelling Model to Other Coastal Upwelling Regions
 The wind-only upwelling model is useful in identifying regions of strong and/or frequent coastal upwelling that are likely to sustain large surface N2O anomalies. The model predicts the highest and most continuous upwelling rates in regions that generally correspond well to the major world coastal upwelling zones documented in the literature (Figures 9 and 10). The Pacific Northwest coast from Oregon down to Baja California emerges as a particularly important upwelling region [Bakun, 1973; Huyer, 1983; Schwing et al., 1996]. Upwelling strength and frequency decline significantly to the south along the Pacific coast of Mexico and Central America, which is not traditionally considered part of the California Current upwelling system. Upwelling is predicted along much of the western coast of South America associated with the Peru/Humboldt current. Highest rates are found off Peru centered around ∼15°S and along the Chilean coast [Peterson and Bellantoni, 1987; Codispoti et al., 1989, 1992]. Upwelling associated with the Canary Current extends as far north as Portugal, and is strongest around 25°N along the northwest coast of Africa [Wooster et al., 1977]. Strong and frequent upwelling occurs along the southwest African coast between 17°S and 35°S in association with the Benguela Current, with particularly high rates around 25°S–27°S [Nelson and Hutchings, 1983; Hutchings, 1992]. In the Arabian Sea, the southwesterly winds of the May–September monsoon season lead to strong upwelling along the coasts of Somalia and Oman. As these winds bend to the south on the eastern side of the Arabian Sea, weaker upwelling occurs along the southwest coast of India [Brock and McClain, 1992; De Wilde and Helder, 1997].
 The average surface ΔN2O anomalies predicted in coastal areas do not correspond directly to the frequency and strength of upwelling, since the anomalies are also a function of the prescribed subsurface N2O concentration (Figure 11). Some of the largest surface anomalies (∼7 nM annual mean) are predicted in the Peruvian upwelling zone, where high subsurface N2O concentrations are co-located with moderate to strong upwelling rates. In comparison, observed surface concentrations of ∼20 nM extending ∼100 km from shore have been reported along much of the coastline from 8°S to 15°S during a non-peak upwelling period [Codispoti et al., 1992]. Strong surface annual mean ΔN2O maxima of up to 8 nM are also predicted along much of the coast of Mexico and Central America, where surface anomalies reaching 4–9 nM have been observed (Figure 4). Since upwelling in this region is relatively weak (Figures 9 and 10), the surface anomalies are due largely to the high subsurface N2O concentrations in the O2-deficient waters of the eastern tropical North Pacific, which are commonly >40 nM at 100 m and have been measured as high as 115 nM [Cohen and Gordon, 1978; Nevison et al., 2003]. Along the northwestern coast of Africa, highest surface ΔN2O (4–5 nM annual mean) is predicted from ∼12°N to 22°N, somewhat equatorward of the upwelling maxima in Figures 9 and 10, but coincident with the subsurface O2 minimum [Locarnini et al., 2002] and consistent with surface observations at 14°N (Figure 4). Along the southwestern coast of Africa, highest surface ΔN2O (∼6 nM annual mean) is predicted near ∼16°S where subsurface O2 minima overlap with strong upwelling rates. We are not aware of observations from this region to compare to the upwelling model prediction. The model predicts mean coastal ΔN2O anomalies of ∼3 nM in the Arabian Sea. In comparison, surface ΔN2O maxima of 5–7 nM have been observed off the coast of Oman (Figure 4) and anomalies of up to ∼25 nM have been observed off the coast of Somalia during the July–August peak of the southwest Monsoon [De Wilde and Helder, 1997]. Along the southwest coast of India, surface anomalies of ∼16 nM were observed during the southwest monsoon, but these declined to near-equilibrium levels during the intermonsoon season [Lal and Patra, 1998].
 To summarize, the predicted annual mean ΔN2O anomalies appear generally consistent with observed values, although the observations represent a snapshot in time that are not necessarily comparable to annual mean values (Figures 3, 4, and 11). In some regions of extreme subsurface O2 deficiency (i.e., Peru, Mexico, and the Arabian Sea), where potentially huge concentrations of N2O can be brought to the surface, the model may significantly underestimate the surface ΔN2O anomaly by a factor of 3 or more.
 Since many coastal regions have not been sampled for N2O, and since high coastal concentrations often decay to near-equilibrium levels away from the coast (Figures 3 and 4), global surface N2O climatologies based on relatively sparse transect data may overlook a number of important coastal source areas. When transect data are extrapolated and filled to create global maps, these omissions can be propagated to entire coastlines. In the N95 method, this problem is aggravated by the tendency of the extrapolation algorithm to assign unsampled gridpoints the value of the nearest zonal neighbor. On north-south running coastlines, which include the eastern boundary regions that comprise the world's major sites of coastal upwelling, this feature of the algorithm will tend to underestimate coastal N2O. The algorithm also tends to smooth out sharp spatial gradients and thus may not capture the abrupt rise of N2O close to the coast (Figures 3 and 4). For all these reasons, the N95 annual composite climatology probably neglects large surface anomalies along the western coasts of much of Central and South America, Africa, and western India (Figure 12). Not surprisingly, the wind-only upwelling model predicts ∼3–8 times more N2O flux from these regions than the N95 method (Table 2).
 In contrast to the regions mentioned above, the N95 method appears to estimate the total coastal N2O flux from the 35°–50° latitude band along the Pacific Northwest coastline more or less correctly, based on the constraints imposed by the atmospheric flux and upwelling models (Tables 1 and 2). However, the apparent agreement may be somewhat fortuitous, since, as noted earlier, the temporal distribution of the flux does not capture the large pulses associated with upwelling events (Figure 7). The spatial distribution also appears to be poorly represented. The pockets of high coastal ΔN2O in the annual composite climatology are centered around a few clusters of expedition data. The high values do not extend continuously along the coastline and are absent in the latitude band from ∼38°N to 43°N, which both upwelling models suggest is an important N2O source region (Figures 8 and 12). Again, the lack of continuity represents a failure of the N95 extrapolation algorithm along north-south running coastlines.
 The total coastal upwelling source of 0.2 Tg N/yr predicted by the wind-only upwelling model (Table 2) is a best-guess estimate that must be regarded as highly uncertain. The estimate is more or less directly proportional to the subsurface ΔN2O100 input, which carries an ∼±30% uncertainty [Nevison et al., 2003]. In addition, observations suggest that some N2O may be upwelled from as deep as 200 m, where the ΔN2O concentration can be at least twice as high as at 100 m [De Wilde and Helder, 1997; Lueker et al., 2003]. The total upwelling model source estimate also decreases by ∼25% when wcrit is doubled or when nback is reduced to 0 and decreases by 15% when the Wanninkhof transfer velocity k is replaced by the Nightingale et al.  formula. Including Ekman pumping due to the divergence of wind stress curl (equation (13)) in the upwelling model increases the total N2O flux by ∼10%. This increase can be partitioned into 10% increases in the Arabian Sea and off the northwest coast of Africa, and ∼30% from the 5°N–35°N region of the eastern North Pacific, which are areas known for high wind stress curl driven upwelling [Bakun and Nelson, 1991; Brock and McClain, 1992; Smith, 1994; Kessler, 2002]. In the remaining coastal regions in Table 2, wind stress curl increases the predicted N2O flux by only a few percent. The combined root mean square of all the above uncertainties suggests a total uncertainty of at least 70% in the Table 2 source estimates.
 While including wind stress curl-driven Ekman pumping makes little difference for the Pacific Northwest coastal region in the wind-only upwelling model presented here, a recent study with a 9-km resolution model suggests that finescale wind stress curl, especially near major coastal promontories, generates more upwelling in the California Current than alongshore wind stress [Pickett and Paduan, 2003]. Interestingly, this study found that traditional calculations, which consider only alongshore wind stress but which use input data centered ∼50 km offshore, where winds tend to be stronger than at ∼10 km, predict relatively accurate total upwelling rates. The ∼50-km resolution NCEP/QSCAT wind stress data used in our study are centered 25–50 km from shore [Milliff and Morzel, 2001]. In sensitivity tests, we found that the upwelling velocity w due to alongshore wind stress was much more strongly correlated to observed buoy SST and Trinidad Head atmospheric N2O anomalies than the upwelling velocity wwc due to wind stress curl.
4.4. Arabian Sea
 The Arabian Sea has long been suggested as a major source of N2O to the atmosphere due to its large regions of O2-deficient subsurface water, which support high N2O production [Law and Owens, 1990; Naqvi and Noronha, 1991; Bange et al., 1996b]. A number of studies have emphasized the strong seasonality in the Arabian Sea N2O source, which is related in large part to wind-driven upwelling during the February–March northeast monsoon and the July–August southwest monsoon. These studies have estimated that the total N2O source from the Arabian Sea may be as high as 0.6 Tg N/yr, with a most probable range of 0.2–0.5 Tg N/yr [Lal and Patra, 1998; Patra et al., 1999; Bange et al., 2001]. While the northwestern Arabian Sea was emphasized as a particularly strong source region in early analyses, Naqvi et al.  recently called attention to the southwestern Indian continental shelf, where extreme N2O levels as high as 533 nM occur in shallow near-surface waters, due to a combination of natural hydrography and anthropogenic nitrate pollution. Using the mean observed surface N2O concentration of 37.6 nM over the June–December 1999 period in which the phenomenon was observed, Naqvi et al.  estimated that the narrow shelf region alone, covering 180,000 km2 could emit 0.06–0.39 Tg N2O-N or more to the atmosphere.
 The N95 method estimates a source of only 0.16–0.20 Tg N2O-N/yr from the entire Arabian Sea (Table 2). This estimate is low compared to other published values in part because the large N2O supersaturations observed on the eastern side of the Arabian Sea [Bange et al., 2001] are not represented in the N95 ΔN2O climatology (Figure 12c). The wind-only upwelling model suggests that the N95 method may underestimate the flux from the perimeter of the Arabian Sea by a factor of 3 (Table 2) or possibly more, since wind stress-curl driven upwelling may extend farther out than 100 km from the coast, the largest width assumed in our study [Brock and McClain, 1992; De Wilde and Helder, 1997]. Furthermore, the values of ΔN2O100 assumed by the upwelling model are based on an empirical parameterization that does not reproduce the extremely high N2O concentrations observed by Naqvi et al.  along the west Indian coast in 1999. We conclude that the N95 method likely underestimates the Arabian Sea N2O source by at least 0.03 Tg N/yr and probably as much as 0.25 Tg N/yr or more, especially when anthropogenic pollution is taken into account.
4.5. Comparison of Upwelling Source to Previous Estimates
 The wind-only upwelling model estimate of the N2O flux associated with coastal upwelling generally agrees well with the one previous estimate that distinguished coastal upwelling zones from other coastal areas [Bange et al., 1996a] (Table 3). Comparison to other estimates is difficult because there is no standard definition of “coastal area” across the various studies and because the conceptual basis of the estimates differs widely [Capone, 1991; Bange et al., 1996a; Seitzinger and Kroeze, 1998]. For example, Seitzinger and Kroeze  estimated annual continental shelf N2O production based on assumptions about rates of microbial nitrification and denitrification and the N2O yields thereof. Their methodology does not explicitly consider upwelling and the timing and mechanisms of air-sea transfer. Continental shelves are defined in their study as coastal areas of <200 m depth. By this criterion, many shelves extend well beyond the maximum 100-km Rossby Radius of deformation assumed in the current study, while other shelves are much narrower than the minimum 40 km [Sverdrup et al., 1942]. Applying the 200-m criterion (using a 1° × 1° bathymetric data set [Gates and Nelson, 1975a, 1975b]) to the N95 method, we estimate a continental shelf source of 0.08 Tg N/yr, nearly an order of magnitude less than Seitzinger and Kroeze's  estimate. Capone's  estimate, which is the highest in Table 3, is also biologically based and involves multiplying global rates of nitrification and denitrification by assumed N2O emission coefficients, which are independent of upwelling and gas transfer. Capone  assumed a high yield of N2O from denitrification (5%) compared to Seitzinger and Kroeze , who estimated the yield at only 0.3%. Bange et al.'s [1996a] methodology explicitly considers air-sea transfer velocities, although their results may have been biased by a handful of extremely high supersaturation anomalies observed in some polluted estuaries [Seitzinger and Kroeze, 1998]. The comparison in Table 3 highlights the fact that other factors besides upwelling may contribute to N2O production and high surface anomalies in coastal waters, including large inputs of riverborne nitrogen from agricultural and industrial pollution [Bange et al., 1996a; Seitzinger and Kroeze, 1998; Naqvi et al., 2000]. Thus the physically based upwelling models presented here may represent only part of the picture needed to quantify the true “coastal” N2O source.
Table 3. Comparison of Coastal N2O Source Estimates
 Independent calculations, using a flux model based on observed atmospheric N2O and an upwelling model based on observed wind and SST data, predict N2O fluxes of similar timing and magnitude from the Pacific Northwest coastline between 41°N and 50°N. A simplified version of the upwelling model, driven primarily by observed winds, predicts that the world's major eastern boundary regions may contribute on the order of 0.2 ± >70% Tg N2O-N/yr. This estimate could increase by a factor of 2 or more if extreme N2O concentrations associated with anthropogenic pollution events are taken into account. The coastal flux of ∼0.2 Tg N/yr represents a significant fraction (∼5%) of the whole-ocean N2O source, estimated here at ∼4 Tg N/yr using traditional gas transfer methods, but also falls well within the range of uncertainty in the total source.
 Despite their transient nature, upwelling events tend to leave residual excess amounts of N2O in surface waters, which are evident even in annual climatologies. Thus traditional gas transfer calculations based on annual composite climatologies of the surface N2O anomaly may effectively predict the correct magnitude of the coastal N2O source, although the temporal and spatial distribution of the flux may be poorly represented. This is true for coastlines like the Pacific Northwest, which are well represented in the climatological database. Other coastal upwelling regions, including the western coasts of South America and South Africa, are less well sampled. As a result, traditional gas transfer calculations may underestimate the coastal upwelling source from these regions by a factor of 3–8.
 We acknowledge NASA grants NAG5-4023 and NAG5-8180 for AGAGE support. C. D. N. thanks NSF grant OCE0096404 for research support. We are grateful to Steve Worley and Stephen Smith for assistance with model input data and to Dudley Chelton for helpful discussions. We thank Marian Westley, Brian Popp, and Steve Barnes for help in collecting the Kilo Moana surface N2O data.