Air Temperature at Ocean Surface Derived from Surface ... - terrapub

0 downloads 0 Views 1MB Size Report
On the other hand, the latter may be related to sensitivity of saturated vapor curve to air temperature. Sensible heat fluxes are estimated by using derived surface ...
Journal of Oceanography Vol. 51, pp. 619 to 634. 1995

Air Temperature at Ocean Surface Derived from Surface-Level Humidity MASAHISA KUBOTA* and AKIRA SHIKAUCHI** School of Marine Science and Technology, Shimizu, Shizuoka 424, Japan (Received 8 December 1994; in revised form 18 April 1995; accepted 20 April 1995)

A new method deriving surface air temperature from specific humidity is proposed. Surface atmospheric pressure and relative humidity in addition to specific humidity are necessary in order to derive surface air temperature. Assuming effects of variation of atmospheric pressure and relative humidity are small, climatological values are used for those values. Derived surface air temperature is compared with in situ surface air temperature. A cross-correlation coefficient is high and the rms error is small. However, the agreement between them varies spatially. The errors are largest in the eastern equatorial region and highlatitudes. The former may be caused by a large sampling error and remarkable interannual variation related to ENSO phenomena. On the other hand, the latter may be related to sensitivity of saturated vapor curve to air temperature. Sensible heat fluxes are estimated by using derived surface air temperature and compared with that by in situ data. For the whole North Pacific, a cross-correlation coefficient, a mean error and an rms difference are 0.89 W m–2, 0.58 W m–2 and 8.03 W m–2, respectively. 1. Introduction Momentum flux transfer between atmosphere and ocean is a main driving force for ocean circulation, while heat transfer between them is important for atmospheric circulation. Global estimates of heat transfer is crucial for understanding the mechanism of climatic variability in a combined system consisting of atmosphere and ocean. The knowledge of global estimates of the surface heat flux requires many physical parameters in the atmospheric and ocean boundaries for the global scale. However; it is not easy to estimate global surface heat fluxes using in situ ocean observation data. Since our interests focus on the heat budget in a global scale in most cases, the sparsity and inhomogeneity of in situ can be a big problem. Using satellite data is only the way to overcome the problem. There are some problems to estimate turbulent heat fluxes because we need to obtain and combine several physical parameters. This is different from the case of momentum flux which can be obtained from one remote sensor such as a microwave scatterometer. Moreover, it is difficult to obtain every necessary physical parameters from satellite data at present. Using a fairly comprehensive bulk formula (e.g., Liu et al., 1979) we need wind speed (v), sea surface temperature (Ts), air temperature (Ta), saturation specific humidity (Qs) and specific humidity (Qa) at sea surface. Since Qs is calculated using Ts which can be easily obtained from an infrared *Present affiliation: Center System Research, University of Tokyo, 4-6-1 Komaba, Meguro-ku, Tokyo 153, Japan. **Present affiliation: Sanyo Techno Marine, Inc., Nihonbashi Horidome-cho, Chuo-ku, Tokyo 103, Japan.

620

M. Kubota and A. Shikauchi

remote sensor such as NOAA/AVHRR, an estimate of Qs is possible using satellite data. On the other hand, it is not so easy to directly obtain Qa using satellite data. Liu (1986) derived a statistical relation between monthly mean Qa and precipitable water (V) using 17 years of radiosonde reports from 46 midocean meteorological stations. The statistical relation was evaluated by Hsu and Blanchard (1989) using measurements during 13 field experiments distributed over the global ocean. The results indicated that the relation could be used with monthly mean and also instantaneous soundings. Therefore, not only Qs but also Qa can be calculated by satellite data alone. Our final problem is to calculate Ta. Generally, Ta can be obtained from a microwave sounder such as NOAA/High Resolution Infrared Sounder (HIRS). However, there are very few studies evaluating NOAA/HIRS estimating Ta because of the sparse spatial resolution and the accuracy. Therefore, it is important to develop a technique to calculate Ta using satellite data in order to obtain sensible heat flux. Thadathil et al. (1993) derived a statistical relation to get surface level air-temperature from satellite observations of precipitable water. However, it is not clear whether the relation is applicable for a global extent or not because the statistical relation is based on regional observation and is not evaluated by independent observational data in the study. It is the main objective of the present study to develop a technique to estimate Ta. In the present study, the possibility to obtain Ta from satellite data alone will be examined. The data and the method will be described in Section 2. In Section 3, validation of the estimated Ta will be done by comparing to observed Ta. Sensible heat flux obtained by using the estimated Ta will be presented and compared with that by observed data in Section 4. Simple sensitivity analysis concerning the estimated Ta and sensible heat flux is carried out in Section 5. The conclusion will be given in Section 6. 2. Data and Method Sea surface temperature (Ts), surface air temperature (Ta), scalar wind (W), atmospheric pressure (P), relative humidity (R) and specific humidity (Qa) in Monthly Summaries Trimmed Groups (MSTG) from COADS Release 1 (Slutz et al., 1985) are used in the present study. Although data are available beginning in 1854, data during 1970–1979 alone are chosen here because the number of the data was higher in this period (Woodruff et al., 1987). Temporal and spatial resolutions are monthly and 2° × 2°, respectively. Ta is derived from Qa, which is given by applying the Liu (1986)’s statistical relation to precipitable water, in the following way (e.g., Blanc, 1985; Iwasaka and Hanawa, 1990). First, vapor pressure (ea) can be calculated from Qa by

ea = PQa / ( 0.622 + Qa ).

(1)

On the other hand, Ta is given from ea and R by the following equation. Ta = −273.2 − 5419.285 / ( ln ea − ln R − ln 6.11 − 19.836).

(2 )

Therefore, Ta can be calculated in principle by P, Qa and R. Specific humidity (Qa) can be derived from precipitable water data observed by a microwave radiometer by using the statistical formulae by Liu (1986). However, because P and R cannot be directly observed by a satellite, Ta cannot be calculated by satellite data alone. Therefore, it is a critical issue to determine surface

Air Temperature at Ocean Surface Derived from Surface-Level Humidity

621

pressure and relative humidity. It should be noted that an effect of P on ea is not so large compared with that of Qa because of relatively small fluctuation of P in space and time. Moreover, lnR seems to be much smaller than lnea, ln6.11 and 19.836. Therefore, we may calculate exact surface air temperature even by using climatological mean values for both P and R. Also, it should be noted that precipitable water can be estimated from satellite data with an accuracy of about 0.2 g cm–2 which is comparable to the accuracy achieved from radiosonde profiles (Abbot and Chelton, 1991). More than 20 bulk formula with different transfer coefficients are proposed to estimate turbulent flux. Blanc (1985) estimated turbulent flux by using ten published bulk transfer coefficient schemes with more than 2600 sets of shipboard observations. He showed that the scheme-to scheme differences results in a typical maximum variation of 70% for an average sensible heat flux determination of ±25 W m–2. The bulk transfer coefficient scheme proposed by Kondo (1975) is used in the present study. The scheme is based on roughness Reynolds number observations combined with the Owen-Thomson (1963) theory of heat and mass transfer above a rough solid surface. 3. Validation of Derived Surface Air Temperature First, derived surface air temperature (Ta′) is compared with observed surface air temperature (Ta) for 10-year average values. Here, monthly Ta′ is derived from monthly Qa, P and R, and then the monthly Ta′ data are averaged for the 10 years. Figure 1 shows a scatter diagram between derived values and observed values. The correlation coefficient is quite high, 0.996. Also, the rms error is small, 0.83°C. This result indicates that Ta averaged over a long time period can be estimated with enough accuracy if we can obtain data of Qa, P and R. Figure 1 also shows that the rms error is larger in a low-temperature region from 0°C to 10°C where Ta′ is larger

Fig. 1. Comparison of the observed surface air temperature (Ta) and derived surface air temperature (Ta′) for the 10-years mean value.

622

M. Kubota and A. Shikauchi

than Ta. Next, applying of the present method to climatological monthly mean values is investigated. Climatological monthly mean Ta′ data are derived from each monthly derived Ta′ by using each monthly P, Qa and R for the period of 1970–1979. Figure 2 shows scatter diagrams between Ta and Ta′ for January, April, July and October. We can easily find the high correlation values and the small rms error as shown in Table 1. It should be noted that the values for boreal winter are not as good as that for other seasons. This feature is consistent with results from the scatter diagrams, which shows that the differences between both values are large in the low-temperature region. Above-mentioned results indicate that derivation of the 10-year averaged and the climatological monthly mean Ta is successful using Qa, P and R observed at the same time. However, it should be noted that we must actually use climatological values for P and R because we cannot obtain those values from satellite data. Therefore, it is necessary to validate the derived Ta by comparing with observed Ta. We calculate monthly Ta by using (a) monthly-mean data, (b) climatological monthly-mean

Fig. 2. Same as in Fig. 1 except for the climatological monthly mean values. (a) January, (b) April, (c) July, and (d) October.

Air Temperature at Ocean Surface Derived from Surface-Level Humidity

623

Fig. 3. Maps of correlation coefficients between (a) Ta and Ta′ or (b) Ta and Ta″. Contour interval is 0.05. The shaded region indicates the correlation coefficient larger than 0.95. A hanning filter is applied two times to correlation coefficients.

624

M. Kubota and A. Shikauchi

data for P and R, and compare the obtained values with observed Ta. For convenience, Ta derived by using original P and R data is represented by Ta′, while that using climatological mean data for P and R is denoted by Ta″. Maps of correlation coefficients and rms errors are given in Figs. 3 and 4, respectively. The correlation coefficients between Ta and Ta′ are remarkable high and larger than 0.95 in most areas except marginal seas northwest of the North Pacific and the

Fig. 4. Same as Fig. 3, except for rms errors. Contour interval is 0.1. The shaded region indicates the rms error larger than 1.5°C.

Air Temperature at Ocean Surface Derived from Surface-Level Humidity

625

Table 1. Correlation coefficients and rms errors between climatological monthly mean Ta and derived Ta. Derived Ta is obtained from monthly mean values of P, Qa and R.

Rms error (°C) Correlation coefficient January April July October Average

1.17 0.61 0.49 0.66 0.83

0.9928 0.9980 0.9974 0.9967 0.9960

Fig. 5. A density map of ocean observations of W(Ts-Ta) in 1979.

equatorial areas. Also rms errors are quite low and less than 0.4°C in most areas in the case of using original data. On the other hand, the correlation coefficients between Ta and Ta″ calculated by using climatological monthly mean values for P and R are larger than 0.95 for the North Pacific, and not so high and so variable for the tropical region. This low correlation coefficient in the tropical region may be associated with the large sampling error because data density in the tropical region is quite low. A density map for W(Ts-Ta) data in 1979 is given in Fig. 5 as an example. The total observation number is larger than 250 in the northern North Pacific, while that is less than 50 in the equatorial Pacific except coastal regions and along ocean routes. The effect of the sampling error in the tropical region is notable not only in Fig. 3, but also in Fig. 4(b). The

626

M. Kubota and A. Shikauchi

contour strongly overlie the main ocean routes such as from Panama to Australia. Next, we investigate time variability of a spatial correlation and rms difference obtained by averaging in the analyzed region, i.e. North Pacific (Fig. 6). The correlation coefficients between Ta and Ta′ are quite high, larger than 0.995 and the rms differences are quite small, about 0.5 degree. Additionally, the correlation coefficients between Ta and Ta″ are still high, between 0.985 and 0.995 and the rms differences are small, 1.0 degree. These values appear to give enough accuracy for estimating heat fluxes. Also, it should be noted that seasonal variations of correlations between Ta and Ta′ of being high in winter and low in summer are remarkable. Such seasonal variability is not so clear in the rms differences. The correlation value is strongly dependent on the spatial variability in Ta and Ta′. Therefore, the remarkable seasonal variation may be caused by the large contrast of surface air temperature in winter between high and low latitudes.

Fig. 6. Time variation of (a) spatial correlation coefficients and (b) rms differences. 䊊 indicates a value between Ta and Ta′. + indicates a value between Ta and Ta″.

Air Temperature at Ocean Surface Derived from Surface-Level Humidity

627

4. Sensible Heat Flux In the previous section, derived surface air temperatures were compared with observed surface air temperature. Since one purpose of deriving surface air temperature is to estimate sensible heat flux by using the surface air temperature, it is necessary to validate the obtained sensible heat flux. Sensible heat fluxes are calculated by using three kinds of surface air temperature, and compared with each other. Moreover, the comparison is carried out for four regions: region A (20°S–20°N); region B (20°N–50°N); region C (50°N–80°N); and the whole region (20°S–80°N). In the first case, sensible heat flux calculated by using original Ta is compared with that by surface air temperature derived from original P, Qa and R data, i.e. Ta′. The results are given in Table 2. The data number in regions A and B are comparable, while the data number in region C is extremely small. This result indicates accuracy of the present method of deriving surface air temperature by P, Qa and R. For the whole North Pacific, the correlation coefficient is 0.89 and a mean error is 0.58 W m–2 and a rms difference is 8.03 W m–2. Though the rms difference seems to be larger than most value of sensible heat flux, the smallness of the mean error may be encouraging for applying the present method to calculate global sensible heat flux. In the equatorial region, a mean error is the largest and an rms difference is the smallest. The small rms difference may be due to small variability there. Second, sensible heat flux calculated by using Ta″ is compared with that by using Ta′. Table 3 presents comparison results, which indicate an error about sensible heat flux derived by using Ta″. In this case, a mean error is remarkably small in all regions. However, correlation coefficients for the equatorial region is small compared with other regions. This is due to the climatological value not being so useful because variability in the equatorial region has a remarkable bi-modal character depending on ENSO phenomena. Moreover, it should be noted Table 2. A comparison between the sensible heat flux by original data and that by using surface air temperature derived from original P and R. Region Data number Corre. coeff. Mean diff. (W m–2 ) Rms diff. (W m –2 )

20°S–20°N 182328 0.95 2.23 3.74

20°N–50°N 50°N–80°N 20°S–80°N 36966 0.87 –1.46 11.36

4308 0.93 –1.39 14.45

321429 0.89 0.58 8.03

Table 3. A comparison between the sensible heat flux by using derived surface air temperature from climatological P and R data and from original P and R data. Region Data number Corre. coeff. Mean diff. (W m–2 ) Rms diff. (W m –2 )

20°S–20°N 182328 0.66 0.22 9.02

20°N–50°N 50°N–80°N 20°S–80°N 136966 0.91 0.17 9.99

2135 0.94 0.12 13.91

321429 0.86 0.16 9.36

628

M. Kubota and A. Shikauchi

that a climatological value is not reliable because there are very few data in the equatorial region as shown in Fig. 5. Finally, sensible heat flux calculated by using Ta is compared with that by Ta. This result shown in Table 4 includes both errors caused by derived surface air temperature by P, Qa and R and by using climatological values for P and R. Still, mean error for the whole region is small, 0.79 W m–2. The mean error for the equatorial region is positive, while mean errors for the midand high-latitudes are negative. It is concluded that these errors are caused by the deriving method because the same features are found in Table 2. On the other hand, the rms error increases with the latitude and that for the high-latitude region is considerably large, 23.04 W m–2. This large value may be in part due to the smallness of the data number there. However, it is considered that

Table 4. A comparison between the sensible heat flux by using original data and surface air temperature derived from climatological P and R. Region Data number Corre. coeff. Mean diff. (W m–2 ) Rms diff. (W m –2 )

20°S–20°N 182328 0.63 2.45 9.68

20°N–50°N 50°N–80°N 20°S–80°N 136966 0.80 –1.31 14.96

2135 0.86 –1.80 23.04

321429 0.76 0.79 12.22

Fig. 7. Relationship of the slope of the saturated vapor pressure curve as a function of temperature.

Air Temperature at Ocean Surface Derived from Surface-Level Humidity

629

there is a more essential problem for deriving air surface temperature in the high-latitudes. Shikauchi (1994) pointed out that the accuracy of deriving temperature is quite sensitive in the low-temperature region because of the characteristics of the saturated vapor pressure curve. Figure 7 shows the relationship of the slope of the saturated vapor pressure curve as a function of temperature. We can easily understand that the sensitivity increases as the temperature increases. This means the accuracy of the predicted air-temperature is worse for the lowtemperature region. This may be the cause of the large rms error for the high-latitude region. 5. Sensitivity Analysis In the previous section we obtained encouraging results, especially for the overall region. However, several problems still exist in the results, for example, the large rms errors for the highlatitude region and the low correlation coefficient for the equatorial region.Therefore, simple sensitivity analysis is carried out. Results shown in Table 3 are reflected by the effectiveness of using climatological monthly mean values for P and R. Therefore, it is important to evaluate the sensitivity of derived air surface temperature to P and R. Substituting equation (1) into (2) we can obtain a relation of P, Qa, R and Ta. The partial derivatives of Ta about P and R are calculated from the obtained equation. The results give the magnitude of the partial derivative about R to be about 100 times as large as that of P. However, what is important is not only the partial derivatives but the deviations, which are given by

∆Ta =

∂ Ta ∂T ∆P + a ∆R. ∂P ∂R

(3)

It may be expected that ∆P and ∆R are strongly regional dependent. Figure 8 shows a map of the partial derivative of Ta about R and the deviation which is defined as the partial derivative of Ta about R times standard deviation of R at each grid point. The partial derivative of Ta about R is large in the mid-latitude region and small in the high-latitudes and the eastern tropical region. It is interesting that the maximum value can be found in the eastern subtropics northeast of the Hawaii Islands. Kubota (1994) pointed out this region to be related to strong Ekman convergence associated with the Subtropical High. On the other hand, distribution of the deviation is quite different from that of the partial derivative. The deviation is extremely large in the equatorial upwelling region and shows a complicated feature, while that in the mid- and high-latitude regions is flat except along the coast. The large value in the tropical region may be due to large variability of R depending on the seasonal and interannual El Nino phenomena (Giese and Cayan, 1993). The large sensitivity of derived Ta to R explains the large mean error and the low correlation coefficient for the equatorial region shown in Table 4. The sensitivity of the calculated sensible heat flux to the derived surface air temperature given by ∆Qh/Ta is also important. Maps of the partial derivative of Qh about Ta and the deviation defined as the partial derivative by a standard deviation of Ta at each grid point are shown in Fig. 9. Figure 8(a) suggests that the partial derivative of the Qh about Ta is relatively large in the high latitudes. However, the distribution of the deviation shown in Fig. 9(b) is similar to that for the Ta about R. The deviation is large in the eastern tropical region in this case. This is also due to the sea surface temperature in the tropical region drastically changes corresponding to the El Nino phenomena. This is clarified by the distribution of standard deviations of Ta shown in Fig. 10.

630

M. Kubota and A. Shikauchi

Fig. 8. A map of (a) the partial derivative of Ta about R and (b) the deviation of Ta defined as the partial differential of Ta times a standard deviation of R. Contour interval is (a) 0.05°C and (b) 0.9°C.

Air Temperature at Ocean Surface Derived from Surface-Level Humidity

631

Fig. 9. A map of (a) the partial derivative of Qh about Ta and (b) the deviation of Ta defined as the partial differential of Qh times a standard deviation of Ta. Contour interval is (a) 0.2 W m–2 °C–1 and (b) 20 W m–2.

632

M. Kubota and A. Shikauchi

Fig. 10. A map of the standard deviation of Ta. Unit is °C. Contour interval is 1.0°C.

The above results show that the estimate of the sensible heat flux in the eastern equatorial Pacific by using the present method is highly sensitive by the large variability for Ta and R related to the El Nino phenomena. On the other hand, the sensitivity results cannot explain the large rms differences in the high latitude regions shown in Table 4. Probably this may be related to the characteristics of the saturated vapor pressure curve pointed out in Shikauchi (1994). Variation of water vapor pressure is less sensitive to temperature variation in the low-temperature region as mentioned before. This suggests the inaccuracy of derivation of low air-temperature by the present method. 6. Conclusions In the present study a new method to obtain the air temperature above the ocean from surfacelevel humidity is proposed. The method is based on the saturated vapor pressure curve. In order to estimate sensible heat flux, surface atmospheric pressure and relative humidity are necessary in addition to air temperature. Since, unfortunately these data cannot be obtained from satellite, climatological values are used for these data. The effectiveness of the present method was verified by using COADS including various ocean observation data such as P, Ts, Ta, Qa and R. Here, monthly averaged data, i.e. MSTG from COADS Release 1 are used because Liu’s statistical relation between Qa and V is applicable for time scales greater than two weeks (Liu et al., 1991). Shikauchi (1994) examined the validity of sensible heat flux obtained by the method in the present study for time scale varying from one day to one month and found that reliability of the estimated sensible heat flux is high for the averaging-period of more than 10 days. Surface air temperature can be derived by the present method even if climatological values for P and R are used. The temporal correlation coefficients between derived and observed surface air temperatures are larger than 0.95 for the North Pacific. However, these are not so high for the tropical region depending on the large sampling error. On the other hand, the spatial correlation coefficient are between 0.985 and 0.995 and the rms differences are about 1°C. These values are

Air Temperature at Ocean Surface Derived from Surface-Level Humidity

633

encouraging. Remarkable seasonal variability can be found in time variation of the spatial correlation coefficients, while it is not the case for time variation of the rms difference. Moreover, sensible heat flux estimated by using the derived surface air temperature is validated. For the whole North Pacific the cross-correlation coefficient is 0.89 and a mean error is 0.58 W m–2 and an rms difference is 8.03 W m–2. It should be noted that sensible heat flux estimated by the ocean observation data is not always accurate. Weare (1989) estimated the uncertainties in calculation of the heat fluxes at the surface of the tropical and subtropical oceans based upon the bulk formulae and archived marine weather reports. He concluded that the total uncertainty for individual observations of the sensible heat fluxes are 12 W m–2. Therefore, the rms difference shown in the present study may be comparable with the total uncertainty. Also, the smallness of a mean error appears to be important for estimating global sensible heat flux. Finally, the sensitivity of the derived air temperature to P and R and that of the derived sensible heat flux to Ta are examined. As expected, the sensitivity of the derived air temperature to R is much larger than that of P. The large sensitivity in the high-latitudes may be closely related to the characteristics of the saturated vapor pressure curve for low-temperatures. On the other hand, the large sensitivity in the eastern equatorial region may be associated with the large variability of R there depending on the seasonal and interannual El Nino phenomena. A new technique to estimate global sensible heat flux by using satellite data is developed.The results show that validity of the present method varies depending on season and region. The use of satellite data and bulk formulae are considered to be critical in order to estimate global heat flux. However, the validity of estimating heat flux should be investigated. Kubota and Shikauchi (1994) compared latent and sensible heat fluxes resulted from satellite data with those estimated by in situ data. The agreement between them remarkably varies dependent on analyzed regions. Therefore, considering characteristics of the method and the data, several kinds of data and methods should be complementally used to obtain accurate heat fluxes. Acknowledgements We are grateful to Mrs. Akio Kaneda, Hisashi Takahashi and Shoichi Mitsumori for their assistance. References Abbott, M. R. and D. B. Chelton (1991): Advances in passive remote sensing of the ocean. Rev. Geophys., Supplement, 571–589. Blanc, T. V. (1985): Variation of bulk-derived surface flux, stability, and roughness results due to the use of different transfer coefficient schemes. J. Phys. Oceanogr., 15, 650–669. Giese, B. S. and D. R. Cayan (1993): Surface heat flux parameterizations and tropical Pacific sea surface temperature simulations. J. Geophys. Res., 98, 6979–6989. Hsu, S. A. and B. W. Blanchard (1989): The relationship between total precipitable water and surface-level humidity over the sea surface: A further evaluation. J. Geophys. Res., 94, 14539–14545. Iwasaka, N. and K. Hanawa (1990): Climatologies of marine meteorological variables and surface fluxes in the North Pacific computed from COADS. Tohoku Geophys. J., 33, 185–239 Kondo, J. (1975): Air-sea bulk transfer coefficients in diabatic conditions. Bound-Layer Meteor., 9, 91–112. Kubota, M. (1994): A mechanism for the accumulation of floating marine debris north of Hawaii. J. Phys. Oceanogr., 24, 1059–1064. Kubota, M. and A. Shikauchi (1994): Sensible and latent heat flux in the North Pacific using satellite data. Proceedings of PORSEC-94, 55–60. Liu, W. T. (1986): Statistical relation between monthly mean precipitable water and surface level humidity over global oceans. Mon. Wea. Rev., 114, 1591–1602. Liu, W. T., K. B. Katsaros and J. A. Businger (1979): Bulk parameterization of air-sea exchanges of heat and water

634

M. Kubota and A. Shikauchi

vapor including the molecular constraints at the interface. J. Atmos. Sci., 36, 1722–1735. Liu, W. T., W. Tang and P. P. Niiler (1991): Humidity profiles over the ocean. J. Climate, 4, 1023–1034. Owen, P. R. and W. R. Thomson (1963): Heat transfer across rough surfaces. J. Fluid Mech., 15, 321–334. Shikauchi, A. (1994): Estimate of latent and sensible heat fluxes over the North Pacific by using satellite data. Master’s Thesis, Tokai University, 107 pp. (in Japanese). Slutz, R. J., S. J. Lubker, J. D. Hiscox, S. D. Woodruff, R. L. Jenne, D. H. Joseph, P. M. Steurer and J. D. Elms (1985): Comprehensive Ocean-Atmosphere Data Set; Release 1, NOAA Environmental Research Laboratories, Climate Research Program, Boulder, CO, 268 pp. (NTIS PB86-105723). Thadathil, P., A. Shikauchi, Y. Sugimori and M. Kubota (1993): A statistical method to get surface level airtemperature from satellite observations of precipitable water. J. Oceanogr., 49, 551–558. Weare, B. C. (1989): Uncertainties in estimates of surface heat fluxes derived from marine reports over the tropical and subtropical oceans. Tellus, 41A, 357–370. Woodruff, S. D., R. J. Slutz, R. L. Jenne and P. M. Steurer (1987): A comprehensive ocean-atmosphere data set. Bull. Amer. Meteor. Soc., 68, 1239–1250.