The ocean surface heat budget comprises radiative

Transcription

1 A NEW AIR SEA INTERACTION GRIDDED DATASET FROM ICOADS WITH UNCERTAINTY ESTIMATES By dav i d i. Be r ry a n d el i Z a B e t h c. Ke n t A new method of analyzing weather reports from merchant ships leads to improved in situ air sea heat flux datasets and information about the uncertainty of the flux estimates. Detail of mean sea surface temperature over the period 1973 to see Fig. 1 for more information. The ocean surface heat budget comprises radiative and turbulent components: the radiative components are solar shortwave and thermal longwave, both of which are strongly affected by cloud; the turbulent heat flux components are the direct transfer of sensible heat and the evaporative transfer of latent heat. Although direct measurements of the turbulent fluxes are possible, they remain a research activity due to the high cost of the measurements. Measurement of the radiative components is easier but still challenging on long-term deployments. As a result, global marine flux datasets are typically constructed using bulk estimates of the mean meteorological parameters and flux parameterizations, known as bulk formulas. Various sources exist for the bulk meteorological parameters, ranging from model output and satellite observations to

2 direct in situ measurements such as those made by the Voluntary Observing Ships (VOS) and moored and drifting buoys. Each source has its own advantages and disadvantages. Satellite estimates can give improved spatial coverage but cannot accurately recover all the variables required to calculate the fluxes. Flux datasets based on atmospheric reanalysis models currently have a low spatial resolution but high temporal resolution. This low spatial resolution may lead to problems in coastal and high variability regions. The VOS observations are known to contain biases, to be of variable quality, and to have uneven sampling, with the observations concentrated in the Northern Hemisphere midlatitudes and major shipping lanes. Although the VOS observations can be of variable quality, they are well characterized in terms of random uncertainty (Kent and Berry 2005), estimates of bias (e.g., Kent et al. 1993; Berry et al. 2004; Kent and Taylor 1996, 2006), and metadata giving information on observing methods (Kent et al. 2007). Averaging over observations from many individual VOS reduces the errors due to any biases on any particular ship, such as an individual calibration offset. A new monthly mean air sea interaction gridded dataset is described and compared to high-quality measurements of the surface fluxes from research moorings. This new dataset, the National Oceanography Centre Southampton (NOCS) Flux Dataset v2.0 (NOCS v2.0) is a major update of the NOCS Flux Dataset v1.1 [NOCS v1.1; often referred to as the SOC Flux Climatology (Josey et al. 1999)]. NOCS v2.0 is based on VOS observations from the International Comprehensive Ocean Atmosphere Data Set (ICOADS; Woodruff et al. 1998; Worley et al. 2005) and is presented on a 1 spatial grid for the period Monthly fields of the meteorological variables and fluxes, and their associated uncertainty estimates, are available for download (at For regions where sampling permits, daily fields will be made available on request. AFFILIATIONS: Be r ry a n d Ke n t National Oceanography Centre Southampton, Southampton, Southampton, United Kingdom CORRESPONDING AUTHOR: David Berry, Ocean Observing and Climate, National Oceanography Centre Southampton, European Way, Southampton, SO14 3ZH, United Kingdom dyb@noc.soton.ac.uk The abstract for this article can be found in this issue, following the table of contents. DOI: /2008BAMS In final form 3 November American Meteorological Society Dataset Development. The bulk formulas. The bulk formulas used to calculate the surface fluxes from meteorological observations were reviewed by the [World Climate Research Programme (WCRP)/ Scientific Committee on Oceanic Research (SCOR)] Working Group on Surface Fluxes (Taylor 2000). Since this review the COARE 3.0 algorithm (Fairall et al. 2003) for the turbulent fluxes was published and has been shown to perform well using fluxes directly measured on research vessels (e.g., Brunke et al. 2002). However, for application to VOS data the COARE 3.0 algorithm, which explicitly models the effect of both any surface warm layer and the cool skin effect, is problematic. To correctly implement the algorithm, the daily evolution of solar radiation and wind speed are required, and it is unclear what effect approximating these values for the gridded daily values would have on the flux estimates. We therefore take the same approach as Josey et al. (1999) and use the bulk formulas of Smith (1980, 1988) for the turbulent fluxes, Reed (1977) for net shortwave flux, and Clark et al. (1974) for the net longwave flux [Eqs. (1) (4)] Q H = ρc p C h u(t sea T air ) (1) Q E = ρlc e u(q sea q air ) (2) Q SW = (1 α)q c (1 0.62n θ n ) (3) Q LW = εσ SB T 4 (0.39 sea 0.05e1/2 )(1 λn 2 ) + 4εσ SB T 3 (T T ) (4) sea sea air Here, Q H, Q E, Q SW, and Q LW are the sensible, latent, net shortwave, and net longwave heat fluxes, respectively; ρ is the density of air; c p is the specific heat capacity of air at constant pressure; L is the latent heat of vaporization; C h and C e are the stability- and height-dependent transfer coefficients for sensible and latent heat, respectively; u, T air, and q air are the scalar wind speed, air temperature, and humidity at 10 m, respectively; T sea is the sea surface temperature; q sea is 98% of the saturation specific humidity at T sea ; Q c is the clear-sky solar radiation; n is the daily mean fractional cloud cover; θ N is the local noon solar elevation; α is the albedo of the sea surface; ε is the emittance of the sea surface; σ SB is the Stefan Boltzmann constant; e is the water vapor pressure; and λ is a latitude-dependent cloud cover coefficient. Uncertainties in the fluxes are caused by either errors in the input data or deficiencies in the bulk formulas. The uncertainty in the fluxes due to uncertainties in the input data is calculated from gridbox 646 may 2009

3 uncertainties (Gleckler and Weare 1997). Uncertainties in the bulk formulas remain poorly known (Taylor 2000) and have not been included in the uncertainty estimates presented with this dataset. Source data. The VOS observations used in NOCS v2.0 are from version 2.4 of ICOADS. Although ICOADS also contains observations from moored and drifting buoys and fixed platforms, these are excluded from our analysis. Further research is required to characterize the uncertainty in these measurements and to determine the best way of incorporating them into the flux dataset. Exclusion of these data also allows their use for validation of NOCS v2.0. Only observations within 4.5 standard deviations of the climatological monthly mean value, as determined from the ICOADS trimming flags, are used. Additionally, observations shown to be mislocated were excluded (Kent and Challenor 2006). In addition to the observations themselves, information on the observing methods, sensor type, and instrumentation height are required for the flux calculation and bias estimation. These observation metadata are collated by the World Meteorological Organisation (Kent et al. 2007) and are merged with ICOADS, following Josey et al. (1999). In some cases it was possible to estimate missing metadata. For example, observation methods depend on the country of recruitment and observation heights vary with the ship type, which may be known (Kent et al. 2007). Bias adjustment and bias uncertainty. Each VOS observation will contain bias. Any element of bias that is specific to a particular ship (e.g., an instrument calibration error) is assumed to contribute to the overall random error for a particular grid box. A particular grid box will typically contain observations from several ships, so the contribution to the overall gridbox uncertainty by this type of bias will be reduced by averaging. Other biases may be characteristic of a particular instrument type or may only occur under certain environmental conditions. For example, measurements of the sea surface temperature using buckets thrown over the side of the ship can be biased cold when the air is much colder than the sea because of cooling of the water in the bucket. Similarly, estimates of the sea surface temperature based on the temperature of the water in the engine room intake can be biased high because of the warming of the pipes by the engines. Even when these biases can be estimated and removed, a residual uncertainty will exist because of imperfections in the correction. In the cases where we know the observations may be biased but know neither the magnitude nor the sign of the bias, the bias uncertainty in the observation will be large. This contribution to the overall gridbox uncertainty is not reduced by averaging. In well-sampled regions the bias uncertainty dominates the total uncertainty estimates. The estimates of bias, their corrections, and the residual bias uncertainties are discussed below for each of the input variables in turn. Wi n d s pe e d. ICOADS VOS wind speeds are either measured using an anemometer or visually estimated from the sea state and converted to a speed using a Beaufort equivalent scale (Kent and Taylor 1997). The methods of measurement preferred by the VOS have changed over time, with the use of anemometers becoming more common (Thomas et al. 2008), and the average measurement height has increased (Kent et al. 2007). Visual wind estimates have been adjusted to account for biases in the Beaufort scale used to report the data (Kent and Taylor 1997) following Lindau (1995). Anemometer wind speeds are adjusted to a standard level of 10 m above sea level using the wind profile relation of Smith (1980) and known measurement heights, where available (Kent et al. 2007). Where heights were unknown, the defaults were based on a 2 area monthly gridded dataset of anemometer heights. Comparisons of adjusted visual and anemometer winds confirmed the conclusion of Thomas et al. (2008) that additional adjustments are required to visual winds to improve agreement with adjusted anemometer winds. The additional adjustment was applied to individual visual wind speed estimates using a simple scaling factor. Prior to the end of 1985 the factor is 1, at the start of 2000 the factor is 0.95, and values in the intervening period are found by linear interpolation. The residual bias uncertainty in the mean wind speed from each method was estimated to be 0.2 m s 1. Su r f a c e a i r t e m p e r at u r e. Observations of air temperature onboard ship are adjusted for biases due to solar heating of the ship and sensor environment following Berry et al. (2004), who estimate the residual bias uncertainty at 0.2 C. The observations are then adjusted to a standard level, chosen as 10 m above sea level (Rayner et al. 2003). The profile relations of Smith (1980,1988) are used with known or estimated measurement heights in the same way as for wind speed [see previous section]. Sea surface temperature. Sources of bias in SST are discussed in detail in Kent and Taylor (2006). Kent and AMERICAN METEOROLOGICAL SOCIETY may

4 Kaplan (2006) estimated bias in the North Atlantic for the period 1975 to 1994 using pairs of nighttime SST observations, each containing one observation made using a bucket and one observation made using an engine intake. Although Kent and Kaplan (2006) identified corrections that could be applied to bucket and engine intake SST observations, these are not immediately applicable to ICOADS. Further research is required to extend the bias adjustments to daytime observations and to observations collected at all wind speeds. The SST data are therefore presently left unadjusted. The estimate of the residual bias uncertainty is the greater of 0.15 C and 0.1 T sea T air C. Sur f a c e h u m i d i t y. Humidity measurements over the ocean are usually made using wet and dry bulb thermometers housed in either screens or whirling psychrometers (Kent et al. 2007). It has been shown that humidities measured using screens are biased high compared to those from whirling psychrometers, probably due to inadequate ventilation of the screens (Kent et al. 1993; Berry and Kent 2005). Comparison of data from each source shows that consistency is improved by adjusting humidity observations made using screens with a reduction of 4% in specific humidity. Where no measurement method was available for an observation, a partial adjustment was made based on the known fraction of observations made by screens within the same 10 area. The residual bias uncertainty is estimated to be 0.2 g kg 1. Both measurement types were adjusted to a standard height of 10 m as for air temperature. Su r f a c e p r e s s u r e. For the period 1973 to 2006 there is no evidence for a systematic bias in reported air pressure. This suggests that the adjustments required to allow for instrument height above sea level have been applied correctly by the observers. Dataset construction background. Our aim is to produce a gridded dataset of surface fluxes from observations that are distributed unevenly in space and time, containing both random uncertainties and biases, and to characterize the uncertainty in the fluxes for each resulting grid box. To achieve this challenging objective, the characteristics of the measurements, the calculation methods, and the fields we wish to estimate must all be carefully considered. Any observational bias will propagate directly into a bias in the fluxes. Observations are therefore adjusted for bias and the residual bias uncertainty is estimated, as described earlier. Random uncertainty in observations will also propagate through the bulk formulas to give a random uncertainty in the fluxes. Also, because the bulk formulas are non-linear, normally distributed random errors in the input variables can lead to biased flux estimates. It is therefore important to reduce any biases in the input data and also to minimize any random uncertainty. Estimates of the random errors in individual observations have been estimated using the semi-variogram method following Kent and Berry (2005). The random error values used for the different variables are listed in Table 1. An obvious method of reducing random uncertainty is to use averages of the input variables in the flux calculation. However, the correlation of the synoptic variability needs to be maintained when calculating the flux fields (e.g., Ledvina et al. 1993; Josey et al. 1995). For example, cold air outbreaks over the Atlantic from the east coast of North America are characterized by cold dry air and high wind speeds over a relatively warm ocean, leading to high latent and sensible heat loss from the ocean. Averaging will reduce these correlations, leading in this example to an underestimate of the heat loss. As a result, there is a need to balance the reduction in the random errors through averaging with maintaining the correlation between the variables. Most early datasets (e.g., Hsiung 1986; Oberhuber 1988) produced monthly mean fields of the input variables and then calculated the fluxes from these monthly fields. More recent datasets (e.g., da Silva et al. 1994; NOCS v1.1) calculated fluxes for each observation and then averaged the fluxes to give a monthly mean value. Table 1. Uncertainty estimates used in the construction of the NOCS v2.0 dataset. Variable Random uncertainty Residual bias uncertainty Pressure (hpa) m wind speed (m s 1 ) m air temperature ( C) m specific humidity (g kg 1 ) SST ( C) 1.2 max[0.15, 0.1(sst-at)] Cloud cover (octas) may 2009

5 In brief, the improvements that have been made in NOCS v2.0 compared to NOCS v1.1 include the following: The period of record has been extended to cover the 34 years from 1973 to VOS observations are bias adjusted based on recent research; updates have been made to the bias adjustments for air temperature, humidity, sea surface temperature, and wind speed. NOCS v1.1 used the successive correction method (da Silva et al. 1994) to smooth simple gridbox averages of fluxes calculated from individual observations. Kent et al. (2000) showed that successive correction, which does not account for uncertainty in the input fields, is not suitable for smoothing fields with strong variations in uncertainty. NOCS v2.0 therefore uses optimal interpolation with individual observations that are characterized by uncertainty estimates. Each grid box and time step in the resulting optimally interpolated NOCS v2.0 includes uncertainty estimates that represent one standard error of the mean for each of the random, bias, and total uncertainty components. Bias uncertainties in the meteorological variables are propagated through the bulk formulas to give a bias uncertainty for each flux component following Gleckler and Weare (1997). Finally, NOCS v1.1 used a climatological monthly mean ice mask, whereas NOCS v2.0 uses weekly ice data. NOCS v2.0 construction methodology. The optimal interpolation (OI) scheme used is based on the scheme developed by Reynolds and Smith (1994) and by Lorenc (1981). OI is performed on the individual observations, relative to a first guess field, and normalized by the uncertainty in the first guess. The full details on the application of this scheme to the VOS for flux estimation can be found in Berry and Kent (2009; manuscript submitted to Int. J. Climatology, hereafter BK). The main changes made to the standard Reynolds and Smith scheme are as follows: 1) First guess: Following Reynolds and Smith (1994) the first guess field was based on the previous day s analysis. However, in data-sparse regions there may be an extended period of time without data; hence, we need to be careful to maintain the annual cycle in these regions. As a result, the first guess field has been set to the previous day s analysis incremented to allow for daily changes in annual cycle estimated from climatological means. 2) First guess uncertainty: The uncertainty in the first guess due to day-to-day variability has been calculated by allowing the uncertainty in the previous day s analysis to increase toward the climatological standard deviation with a 3-day e-folding scale. After several days without data, the first guess uncertainty will be close to the climatological standard deviation for that grid box. 3) Random uncertainty: Estimates of the random uncertainty in individual observations are required to determine their weights in the OI. The values used are given in Table 1. Errors in the observations are assumed to be uncorrelated for all VOS observations. 4) Spatial correlation scales: The spatial correlation terms required for the OI have been estimated using a Gaussian function with e-folding scales of 300 km. This length scale was chosen to balance the capture of the synoptic variability with the limited number of observations. 5) Ice mask: A weekly ice mask, based on Reynolds et al. (2002), has been used to exclude those regions covered by ice from the analysis. For each daily analysis the ice mask is interpolated to give a daily value, and any grid box in which sea ice concentration exceeds 25% is not analyzed. If a grid box becomes ice free, the appropriate climatological mean value for that grid box is used as the first guess and the climatological standard deviation as the first guess error. The weekly ice data are only available from 1982; prior to 1982 climatological weekly values for sea ice concentration, interpolated to daily values, are used. Daily uncertainty estimates are produced as part of the OI for each variable, grid box, and time step. These are then used, through propagation of errors (e.g., Taylor 1997; Gleckler and Weare 1997), to estimate the uncertainties in the fluxes due to random errors and sampling. The residual bias uncertainty is estimated separately, again using propagation of errors, to estimate the uncertainty in the daily fluxes due to unknown biases. The total uncertainty is then the sum of the random and bias uncertainties added in quadrature. Monthly mean fields and uncertainty estimates. Monthly mean fields have been calculated as the arithmetic mean of the daily analysis values within the month. To give realistic uncertainty estimates in the monthly mean values, the correlation between the different daily analyses needs to be taken into account. Similarly, when averaging onto a larger spatial grid, the correlation between grid boxes needs to be taken AMERICAN METEOROLOGICAL SOCIETY may

6 Fig. 1. Mean values over the period 1973 to 2006: (a) surface pressure; (b) scalar wind speed; (c) SST; (d) specific humidity; (e) air temperature; (f) cloud cover; (g) air temperature bias adjustment; (h) air temperature total monthly uncertainty; (i) mean monthly standard deviation of daily air temperatures; (j) standard deviation of monthly mean air temperatures. 650 may 2009

7 into account. The temporal correlation will depend on the impact of the first guess field on the data fields. In well-sampled regions the weight of the observations, relative to the first guess, will be high and the temporal correlation in the uncertainties low. However, when there are few input observations, information from the first guess will persist and the uncertainties will be highly correlated. The temporal correlation is estimated based on the relative impact of the observations to the first guess field on the daily analyses. The spatial correlation between the uncertainty estimates has been derived using the same 300-km spatial correlation scale used in the OI. Once the correlations between each day and grid box going into an average are known, the uncertainty in the average can be estimated using propagation of errors (e.g., Taylor 1997). For further details, see BK. The NOCS v2.0 Dataset. Mean meterological fields. Figure 1 shows the mean fields, after adjustment for bias and to a standard measurement height of 10 m where appropriate, for each of the input variables averaged over the full 34 years of the NOCS v2.0 dataset (Figs. 1a f). Also shown is an example of the bias adjustments applied to air temperature. The adjustment of the measured air temperature to the standard height of 10 m (not shown) typically increases the air temperature because the measurements are typically made at heights greater than 10 m (Kent et al. 2007). The subsequent bias adjustment for solar heating (Berry et al. 2004) acts to decrease the air temperature and is greatest when the incoming solar radiation is largest (Fig. 1g). The size of the bias adjustment is modulated by the relative wind speed over the ship, which acts to reduce the effect of the solar heating. The monthly mean total uncertainty in air temperature (Fig. 1h) is strongly dependent on sampling and shows small uncertainties (~0.25 C) in the main shipping lanes and larger values in regions that are less well sampled. The uncertainty also includes the effects of any undersampling of natural variability and the residual bias uncertainty. Figures 1i and 1j show two estimates of the variability in air temperature: Fig. 1i shows the mean standard deviation of daily fields within a month (the intramonthly standard deviation) and Fig. 1j shows the standard deviation of the monthly mean (the intermonthly standard deviation). The standard deviation in the monthly means is dominated by the annual cycle. Typically, the air temperature variance within a month is about one fifth of the variance of the monthly means. However, highly variable regions, such as the Gulf Stream, show a relatively higher contribution of variability on the shorter time scales. In a few regions, such as the northeast tropical Pacific, where variability is dominated by the Gulf of Tehuantepec wind jet, the day-to-day variance exceeds that of the annual cycle. Surface fluxes. Figure 2 shows each component of the heat flux, calculated from the input fields shown in Figs. 1a f. The sign convention is positive for heat gain by the ocean (the shortwave flux) and negative for heat loss by the ocean (typically the longwave, sensible, and latent heat fluxes). The shortwave radiation (Fig. 2a) shows the expected latitudinal dependence modulated by the effect of clouds (Fig. 1f), varying from near zero at the poles to over 275 W m 2 in the tropics. The range of variation in annual mean longwave is less (Fig. 2b), from 13 to 68 W m 2, dominated by variations in cloud cover (Fig. 1f) and SST (Fig. 1c). The latent heat flux (Fig. 2c) shows large heat loss in the tropics, typically 100 to 140 W m 2, as expected from the SST and Clausius Clapeyron relation, but with the peak annual mean heat loss in the region of the western boundary currents (the Gulf Stream in the Atlantic and the Kuroshio in the Pacific) where annual mean heat loss exceeds 220 W m 2. Sensible heat loss (Fig. 2d) is smaller than the latent heat loss, especially in the tropics where mean sensible heat loss is typically around 10 W m 2. Peak annual mean sensible heat loss in the western boundary current regions exceeds 50 W m 2. The annual mean total heat flux is shown in Fig. 2e and its uncertainty in Fig. 2f. The seasonal variation of the total heat flux is shown in Figs. 2g j. Over the period of NOCS v2.0 ( ) the global imbalance of the annual mean net heat flux is 24 W m 2, comparable to that in NOCS v1.1 and da Silva et al. (1994). An imbalance of this magnitude is unphysical: Grist and Josey (2003) estimated that every W m 2 of heat input would warm the upper 1000 m of the ocean by around 0.1 C per decade. The Intergovernmental Panel on Climate Change Fourth Assessment (Solomon et al. 2007) estimated that heating of the upper 700 m of the ocean of 0.1 C between 1961 and 2003 had occurred, suggesting that the global imbalance should be on the order of 1 W m 2 over this period. As in NOCS v1.1, the magnitude of the global imbalance varies over the period of the dataset: for NOCS v2.0 the imbalance reduces from around 40 W m 2 in the early 1970s to around 20 W m 2 by This trend in net heat flux is much reduced if only relatively well-sampled regions are considered: in the Northern Hemisphere extratropics the trend AMERICAN METEOROLOGICAL SOCIETY may

8 Fig. 2. Climatological mean values over the period 1973 to 2006: (a) shortwave radiation; (b) longwave radiation; (c) latent heat flux; (d) sensible heat flux; (e) total heat flux annual mean; (f) uncertainty in net heat flux; (g) total heat flux Northern Hemisphere winter [December February (DJF)]; (h) total heat flux spring [March May (MAM)]; (i) total heat flux summer [June August (JJA)]; (j) total heat flux autumn [September November (SON)]. Positive values indicate heat gain by the ocean. 652 may 2009

9 in net heat flux is close to 10 W m 2, similar to that seen by Yu and Weller (2007) in global latent heat flux over the period 1981 to Analyzed Air Sea Fluxes (OAFlux) product (Yu and Weller 2007). The OAFlux product is a synthesis of reanalyses and satellite data. Also shown is a version Accuracy of surface fluxes. There are several sources of data that can be used for validation of the surface products. Research vessels make high-quality measurements of both the direct fluxes (e.g., Fairall et al. 2003) and mean meteorology (Gould and Smith 2006). Moorings reporting research-quality measurements of meteorological variables and radiative fluxes have been made in a series of buoy deployments since 1981 by the Woods Hole Oceanographic Institution (WHOI) Upper Ocean Processes Group. More recently, data have become available via the OceanSITES project, which brings together sustained observation programs including long-term deployments from WHOI and two operational tropical moored buoy arrays: the Tropical Atmosphere Ocean (TAO)/Triangle Trans-Ocean Buoy Network (TRITON) (McPhaden et al. 1998) and the Prediction and Research Moored Array in the Atlantic (PIRATA) (Bourlès et al. 2008). As an example, we make a preliminary comparison of the fluxes from NOCS v2.0 with research-quality measurements from the WHOI archive. We only use those deployments that contain all the necessary measurements required to estimate all components of the heat budget (i.e., shortwave and longwave radiation and the mean meteorological variables required to calculate the turbulent fluxes from the bulk formulas described in section 2a). Figure 3 shows a comparison of each flux component, together with the net heat flux, for a sample 4-month deployment in the sub-arctic Nor t h At la nt ic Ocea n during the summer of 1991 (Plueddemann et al. 1995, denoted MLM91 in Fig. 4). Also shown are fluxes from three reanalysis products (Kalnay et al. 1996, denoted NCEP1; Kanamitsu et al. 2002, denoted NCEP2; Uppala et al. 2005, denoted ERA40) and the Objectively Fig. 3. Comparison of surface fluxes from a surface mooring Marine Light Mixed Layer experiment in the sub-arctic North Atlantic Ocean (59.5 N, 20.8 W), 29 April September Flux differences (product buoy) are shown for NOCS v2.0 (black), NOCS v2.0 with height adjustment but without bias adjustment (vertical lines), NCEP1 Reanalysis (horizontal lines), NCEP2 Reanalysis (upward sloping lines); ERA40 Reanalysis (downward sloping lines), and OAFlux (squares). Fig. 4. Locations of the WHOI moorings used for comparison with gridded flux products (NOCS, NCEP1, NCEP2, ERA40, and OAFlux; see text for descriptions). All of the moorings in the WHOI data archive [ whoi.edu/archives/dataarchives.html] with estimates of all four components of the heat flux were used, with the exception of those from the Coastal Ocean Processes Inner Shelf Study. The moorings are the Arabian Sea Mixed Layer Dynamics Experiment (Arabian Sea); Acoustic Surface Reverberation Experiment, 1991 (ASREX91); Coastal Mixing and Optics Moored Array, Central Mooring (CMO); Coupled Ocean Atmosphere Response Experiment (COARE); Marine Light Mixed Layers Experiment, 1991 (MLM91); Severe Environment Surface Mooring (SESMOOR); Shelf Mixed Layer Experiment (SMILE); and five buoys from the Subduction Experiment. Each mooring location is labeled with the name of the flux product ranked as being in best agreement with the mooring fluxes. AMERICAN METEOROLOGICAL SOCIETY may

10 of the NOCS v2.0 fluxes without adjustment for measurement error (but with adjustments for height and Beaufort scale). Only products with daily estimates are compared to allow close matching with the full deployment period and to account for any gaps in the buoy data. This excludes the NOCS v1.1 dataset from this comparison. Each component of the NOCS v2.0 flux, as well as the net heat flux, is within 10 W m 2 of the buoy values (Fig. 3). The net heat flux is within 2 W m 2 of that from the buoy. The estimated uncertainty in the NOCS v2.0 net heat flux is 4 W m 2. The bias adjustment applied to the NOCS v2.0 acts to improve the agreement with the buoy: prior to adjustment, the offset is 16 W m 2. The reanalysis products show biases compared with the buoy of between 12 and 21 W m 2. OAFlux has a bias in the net heat flux compared to the buoy of 9 W m 2. However, the largest differences for the OAFlux product come in the radiative components, which are satellite-based estimates (Zhang et al. 2004). Note that the OAFlux product is a combination of datasets for which the optimal weightings have been derived using the WHOI moorings, so the comparison shown here is not strictly independent. In addition, use of a different turbulent flux algorithm (COARE 3.0) in the OAFlux product may have acted to worsen comparisons with the buoy. Figure 3 shows that mean differences for the flux components can be large for the reanalysis products: the maximum mean difference seen over the period of this deployment is 38 W m 2 for shortwave from the NCEP2 product. In Fig. 3 the NOCS v2.0 fluxes compare well to those calculated from the buoy. A simple ranking analysis shows that this is a typical situation. The performance of each dataset was ranked by its absolute mean difference from the buoy for each component, as well as for the net heat flux. The NOCS v2.0 was ranked as closest to the buoy fluxes in five out of the eleven comparisons possible; the other products were each ranked closest on either one or two deployments (Fig. 4). A full comparison of the fluxes with the WHOI moorings and other sources of high-quality flux and meteorological data is planned. Differences are expected to arise from biases in the meteorological variables, the different spatial scales of the datasets, deficiencies in the bulk formulas or model physics, or problems with the mooring data themselves (Weller et al. 2008). Summary and Future Developments. NOCS v2.0, a new dataset of mean meteorological variables and the surface heat flux components calculated from the ICOADS archive of in situ meteorological observations, has been presented. The dataset is available at monthly resolution for the period 1973 to 2006 on a 1 spatial grid, although the effective resolution of the data is determined by the length scale of 300 km used in the OI. NOCS v2.0 represents a major update to the NOCS v1.1 dataset (Josey et al. 1999). Improvements compared to NOCS v1.1 include: Extension to the 34-year period 1973 to 2006, Estimates of random and bias uncertainty for all variables and fluxes, Improved dataset construction method using optimal interpolation, Calculation of daily mean fluxes from daily meteorological fields to retain the correlations in synoptic variability between the variables and reduce bias due to nonlinearity of the bulk formulas, Updated bias adjustments for air temperature (Berry et al. 2004), humidity, and wind speed (bias adjustment for SST applied in NOCS v1.0 is now not applied), and Use of new ice datasets. The new dataset, like its predecessors (da Silva et al. 1994; NOCS v1.1), does not achieve a realistic heat balance over the ocean. However, the accompanying uncertainty estimates for each grid box of the input meteorological variables and flux components allows analyses to focus on regions where the quality of the data is appropriate for a particular application. The uncertainty estimates could also be used to improve inverse calculations to constrain the surface fluxes using estimates of ocean heat transport (da Silva et al. 1994; Grist and Josey 2003). Detailed comparisons of the flux estimates with previous datasets and with newer flux datasets such as OAFlux (Yu and Weller 2007), the results of atmospheric and ocean reanalyses, and high-quality measurements from moorings (Taylor et al. 2001; McPhaden et al. 1998; Bourlès et al. 2008) and research vessels (Gould and Smith 2006) are the subject of future work. Future developments to the dataset will fall into three main areas: improvements to the characterization of the input observations, improved analysis methods, and the extension to include a wider range of input data. The characterization of uncertainties is an active area of investigation and future updates will include the results of any new research. Analysis of NOCS v2.0 itself will allow improved estimation of the spatial and temporal correlation length scales and of the natural background variability estimates used in the OI scheme. It is hoped in the future to improve analysis methods and uncertainty estimation 654 may 2009

11 in very data-sparse regions. By using only VOS data, NOCS v2.0 is independent of the buoy and satellite records and has only a weak relationship to the atmospheric reanalyses. This independence is invaluable in dataset intercomparisons, and a version with these characteristics will be maintained. However, the inclusion of selected data from other sources will improve the quality of the fluxes, especially in regions poorly sampled by the VOS, and future developments will include versions of the dataset incorporating additional data. (NOCS v2.0 meteorological data and surface fluxes can be downloaded from the project Web site at Acknowledgments. This work was funded by the UK Natural Environment Research Council (NERC) Oceans2025 programme with additional funding from the NERC & Ministry of Defence Joint Grants Scheme and the Met Office through the National Centre for Ocean Forecasting. ICOADS individual reports were obtained from the Research Data Archive managed by the Data Support Section of the Computational and Information Systems Laboratory at the National Center for Atmospheric Research in Boulder, Colorado. ECMWF ERA-40 data were obtained from the ECMWF data server. NCEP1 and NCEP2 Reanalysis derived data were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, from their Web site at OAFlux data were obtained from the OAFlux project Web site at edu/. The WHOI mooring data were obtained from the WHOI Upper Ocean Processes Group Web site at uop.whoi.edu/archives/dataarchives.html. NOAA OI SST V2 ice data were obtained from the NCEP Environmental Modeling Centre ( sst_analysis/). The Ferret program, a product of NOAA s Pacific Marine Environmental Laboratory, was used for analysis and graphics in this paper ( gov/ferret/). Finally, we thank the reviewers for their help in improving this paper. References Berry, D. I., and E. C. Kent, 2005: The effect of instrument exposure on marine air temperatures: An assessment using VOSClim data. Int. J. Climatol., 25 (7), , doi: /joc.1178.,, and P. K. Taylor, 2004: An analytical model of heating errors in marine air temperatures from ships. J. Atmos. Oceanic Technol., 21, Bourlès, B., and Coauthors, 2008: The PIRATA Program: History, accomplishments, and future directions. Bull. Amer. Meteor. Soc., 89, Brunke, M. A., X. Zeng, and S. Anderson, 2002: Uncertainties in sea surface turbulent f lux algorithms and data sets. J. Geophys. Res., 102, doi: /2001jc Clark, N. E., L. Eber, R. M. Laurs, J. A. Renner, and J. F. T. Saur, 1974: Heat exchange between ocean and atmosphere in the eastern North Pacific for NOAA Tech. Rep. NMFS SSRF-682, U.S. Dept. of Commerce, 108 pp. Da Silva, A., A. C. Young, and S. Levitus, 1994: Atlas of surface marine data Vol. 1. Tech. Rep. 6, U.S. Department of Commerce, NOAA, NESDIS, 83 pp. Fairall, C. W., E. F. Bradley, J. E. Hare, A. A. Grachev, and J. B. Edson, 2003: Bulk parameterization of air sea fluxes: Updates and verification for the COARE algorithm. J. Climate, 16, Gleckler, P. J., and B. C. Weare, 1997: Uncertainties in global ocean surface heat flux climatologies derived from ship observations. J. Climate, 10, Gould, W. J., and S. R. Smith, 2006: Research vessels: Underutilized assets for climate observations. Eos, Trans. Amer. Geophys. Union, 87, Grist, J. P., and S. A. Josey, 2003: Inverse analysis adjustment of the SOC air sea flux climatology using ocean heat transport constraints. J. Climate, 20, Hsiung, J., 1986: Mean surface energy fluxes over the global ocean. J. Geophys. Res., 91 (C9), Josey, S. A., E. C. Kent, and P. K. Taylor, 1995: Seasonal variations between sampling and classical mean turbulent heat flux estimates in the North Atlantic. Ann. Geophys., 13, ,, and, 1999: New insights into the ocean heat budget closure problem and analysis of the SOC air sea flux climatology. J. Climate, 12, Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, Kanamitsu, M., W. Ebisuzaki, J. Woollen, S.-K. Yang, J. J. Hnilo, M. Fiorino, and G. L. Potter, 2002: NCEP DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, Kent, E. C., and P. K. Taylor, 1997: Choice of a Beaufort equivalent scale. J. Atmos. Oceanic Technol., 14, , and D. I. Berry, 2005: Quantifying random measurement errors in voluntary observing ships meteorological observations. Int. J. Climatol., 25, , doi: /joc.1165., and P. G. Challenor, 2006: Toward estimating climatic trends in SST. Part II: Random errors. J. Atmos. Oceanic Technol., 23, AMERICAN METEOROLOGICAL SOCIETY may

12 , and A. Kaplan, 2006: Toward estimating climatic trends in SST. Part III: Systematic biases. J. Atmos. Oceanic Technol., 23, , and P. K. Taylor, 2006: Toward estimating climatic trends in SST. Part I: Methods of measurement. J. Atmos. Oceanic Technol., 23, ,, B. S. Truscott, and J. S. Hopkins, 1993: The accuracy of voluntary observing ships meteorological observations Results of the VSOP-NA. J. Atmos. Oceanic Technol., 10, ,, and P. G. Challenor, 2000: The effect of successive correction on variability estimates for climatological datasets. J. Climate, 13, , S. D. Woodruff, and D. I. Berry, 2007: Metadata from WMO Publication No. 47 and an assessment of Voluntary Observing Ship observation heights in ICOADS. J. Atmos. Oceanic Technol., 24, Lindau, R., 1995: A new Beaufort equivalent scale. Proc. International COADS Wind Workshop, Kiel, Germany, Institut für Meereskunde and NOAA Environmental Research Labs, Lornec, A. C., 1981: A global three-dimensional multivariate statistical interpolation scheme. Mon. Wea. Rev., 109, McPhaden, M. J., and Coauthors, 1998: The Tropical Ocean Global Atmosphere (TOGA) observing system: A decade of progress. J. Geophys. Res., 103, Oberhuber, J. M., 1988: An atlas based on the COADS data set: The budget of heat, buoyancy and turbulent kinetic energy at the surface of the global ocean. Max Planck Institute for Meteorology. Rep. 15, 20 pp. Plueddemann, A. J., R. A. Weller, M. Stramska, T. D. Dickey, and J. Marra, 1995: The vertical structure of the upper ocean during the Marine Light Mixed Layer Experiment. J. Geophys. Res., 100, Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Global analyses of SST, sea ice and night marine air temperature since the late 19th century. J. Geophys. Res., 108, 4407, doi: /2002jd Reed, R. K., 1977: On estimating insolation over the ocean. J. Phys. Oceanogr., 7, Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7, , N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15, Smith, S. D., 1980: Wind stress and heat flux over the ocean in gale force winds. J. Phys. Oceanogr., 10, Smith, S. D., 1988: Coefficients for sea surface wind stress, heat flux, and wind profiles as a function of wind speed and temperature. J. Geophys. Res., 93 (C12), Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H. L. Miller, Eds., 2007: Climate Change 2007: The Physical Science Basis. Cambridge University Press, 996 pp. Taylor, J. R., 1997: An Introduction to Error Analysis: The Study of Uncertainties in Physical Measurements. 2nd ed. University Science Books, 327 pp. Taylor, P. K., Ed., 2000: Intercomparison and validation of ocean atmosphere energy flux fields Final report of the Joint WCRP/SCOR Working Group on Air Sea Fluxes. WCRP-112, WMO/TD-1036, 306 pp., E. F. Bradley, C. W. Fairall, L. Legler, J. Schulz, R. A. Weller, and G. H. White, 2001: Surface fluxes and surface reference sites. Observing the Oceans in the 21st Century. C. Koblinsky and N. Smith, Eds., Bureau of Meteorology, Melbourne, Australia, Thomas, B. R., E. C. Kent, V. R. Swail, and D. I. Berry, 2008: Analysis of monthly mean marine winds adjusted for observation method and height. Int. J. Climatol., 28, , doi: /joc Uppala, S. M., and Coauthors, 2005: The ERA-40 re-analysis. Quart. J. Roy. Meteor. Soc., 131, Weller, R. A., E. F. Bradley, J. Edson, C. Fairall, I. Brooks, M. J. Yelland, and R. W. Pascal, 2008: Sensors for physical fluxes at the sea surface: Energy, heat, water, salt. Ocean Sci. Discuss., 5, Woodruff, S. D., H. F. Diaz, J. D. Elms, and S. J. Worley, 1998: COADS release 2 data and metadata enhancements for improvements of marine surface flux fields. Phys. Chem. Earth, 23, Worley, S. J., S. D. Woodruff, R. W. Reynolds, S. J. Lubker, and N. Lott, 2005: ICOADS release 2.1 data and products. Int. J. Climatol., 25, Yu, L., and R. A. Weller, 2007: Objectively analyzed air sea heat fluxes for the global ice-free oceans ( ). Bull. Amer. Meteor. Soc., 88, Zhang, Y., W. B. Rossow, A. A. Lacis, V. Oinas, and M. I. Mishchenko, 2004: Calculation of radiative fluxes from the surface to top of atmosphere based on ISCCP and other global data sets: Refinements of the radiative transfer model and the input data. J. Geophys. Res., 109, D19105, doi: /2003jd may 2009