SURFACE FLUXES
259
The near surface humidity in the NCEP reanalysis is too high (Fig. 9),
with the relative humidity, γ, seldom less than 80%. Large et al. (1997)
were able to match the Kent et al. (1993) observed annual cycle of
latent heat flux in the North Atlantic only after reducing the NCEP
specific humidity by a factor of 0.93. In the western tropical Pacific,
a comparison of NCEP relative humidity to TAO mooring data in the
equatorial Pacific shows a year-round positive bias of ≈ 2 − 3% (Wang
and McPhaden 2001). Plausible reasons for this bias are that ship based
ocean humidity measurements that are assimilated into the reanalysis
are more likely too wet than too dry (Kent et al. 1993), and that the
NCEP model evaporates too much (Smith et al. 2001) and transports
too little vertically out of the boundary layer. The excess evaporation
is accomplished despite the low wind speeds and moist near surface by
an excessively large C E (Smith et al. 2001). The TAO data suggests
reducing NCEP relative humidity by 3% in the tropics and this is used
as a minimum correction at all latitudes. Figure 9 shows that NCEP
relative humidity exceeds the SOC climatology almost everywhere outside the tropical band, with local differences exceeding 14%. The SOC
humidities should be most reliable in the northern Hemisphere, because
of the greater number of ship reports and the extensive work of Kent
et al. (1993) in the North Atlantic. Therefore, in the north a smoothed
version of the zonally averaged difference is used as a correction wherever
it exceeds the minimum 3%. In the absence of any further information,
and because of the rough symmetry in Fig. 9 about the equator, a similar correction is applied in the south, but with much less confidence. It
is a maximum of about 6% at 50 ◦ S. The net result is the NCEP bias
δγ shown as a function of latitude by the dashed line in Fig. 9. Overall, lowering the relative humidity increases the latent heat flux loss by
13W/m 2 (Table 3).
Over most of the globe NCEP surface air temperatures are not corrected. However, comparison of NCEP temperatures with weather station and drifting buoy data from Antarctica reveals that a persistent very
cold bias exists in the reanalysis product at extreme southern latitudes,
especially in the winter. In the Arctic, the POLES (Polar Exchange
at the Sea Surface) project has combined buoy and land station temperature data within an optimum interpolation scheme (Rigor et al.,
2000). Over the Arctic cap north of 70 ◦ N , the annual average POLES
and NCEP air temperatures differ randomly by only about ±0.5 ◦ C, but
NCEP air-temperatures are corrected monthly by the mean monthly
climatological difference between POLES and NCEP. The 12 monthly
corrections (January through December) are 0.49, 0.06, -0.73, -0.89,
-0.77, -1.02, -1.99, -0.91, 1.72, 2.30, 1.81 and 1.06 ◦ C.
259
The near surface humidity in the NCEP reanalysis is too high (Fig. 9),
with the relative humidity, γ, seldom less than 80%. Large et al. (1997)
were able to match the Kent et al. (1993) observed annual cycle of
latent heat flux in the North Atlantic only after reducing the NCEP
specific humidity by a factor of 0.93. In the western tropical Pacific,
a comparison of NCEP relative humidity to TAO mooring data in the
equatorial Pacific shows a year-round positive bias of ≈ 2 − 3% (Wang
and McPhaden 2001). Plausible reasons for this bias are that ship based
ocean humidity measurements that are assimilated into the reanalysis
are more likely too wet than too dry (Kent et al. 1993), and that the
NCEP model evaporates too much (Smith et al. 2001) and transports
too little vertically out of the boundary layer. The excess evaporation
is accomplished despite the low wind speeds and moist near surface by
an excessively large C E (Smith et al. 2001). The TAO data suggests
reducing NCEP relative humidity by 3% in the tropics and this is used
as a minimum correction at all latitudes. Figure 9 shows that NCEP
relative humidity exceeds the SOC climatology almost everywhere outside the tropical band, with local differences exceeding 14%. The SOC
humidities should be most reliable in the northern Hemisphere, because
of the greater number of ship reports and the extensive work of Kent
et al. (1993) in the North Atlantic. Therefore, in the north a smoothed
version of the zonally averaged difference is used as a correction wherever
it exceeds the minimum 3%. In the absence of any further information,
and because of the rough symmetry in Fig. 9 about the equator, a similar correction is applied in the south, but with much less confidence. It
is a maximum of about 6% at 50 ◦ S. The net result is the NCEP bias
δγ shown as a function of latitude by the dashed line in Fig. 9. Overall, lowering the relative humidity increases the latent heat flux loss by
13W/m 2 (Table 3).
Over most of the globe NCEP surface air temperatures are not corrected. However, comparison of NCEP temperatures with weather station and drifting buoy data from Antarctica reveals that a persistent very
cold bias exists in the reanalysis product at extreme southern latitudes,
especially in the winter. In the Arctic, the POLES (Polar Exchange
at the Sea Surface) project has combined buoy and land station temperature data within an optimum interpolation scheme (Rigor et al.,
2000). Over the Arctic cap north of 70 ◦ N , the annual average POLES
and NCEP air temperatures differ randomly by only about ±0.5 ◦ C, but
NCEP air-temperatures are corrected monthly by the mean monthly
climatological difference between POLES and NCEP. The 12 monthly
corrections (January through December) are 0.49, 0.06, -0.73, -0.89,
-0.77, -1.02, -1.99, -0.91, 1.72, 2.30, 1.81 and 1.06 ◦ C.