• Alishouse, J. C., , S. Snyder, , J. Vongsathorn, , and R. R. Ferraro, 1990: Determination of oceanic total precipitable water from the SSM/I. IEEE Trans. Geosci. Remote Sens., 28 , 811866.

    • Search Google Scholar
    • Export Citation
  • Chou, S. H., , R. M. Atlas, , C. L. Shie, , and J. Ardizzone, 1995: Estimates of surface humidity and latent heat fluxes over oceans from SSM/I data. Mon. Wea. Rev., 123 , 24052425.

    • Search Google Scholar
    • Export Citation
  • Chou, S. H., , C. L. Shie, , R. M. Atlas, , and J. Ardizzone, 1997: Air–sea fluxes retrieved from Special Sensor Microwave Imager data. J. Geophys. Res., 102 , C6. 1270512726.

    • Search Google Scholar
    • Export Citation
  • Chu, J. T., , J. N. Chen, , and L. Y. Xu, 2006: Study of the algorithm of calculating turbulent heat fluxes at air–sea interface and applying it to the neighboring seas of China. Oceans Limnol., in press.

    • Search Google Scholar
    • Export Citation
  • Dourras, D., , L. Eymard, , and W. T. Liu, 2002: A neural network to estimate the latent heat flux over oceans from satellite observations. Int. J. Remote Sens., 23 , 24052423.

    • Search Google Scholar
    • Export Citation
  • Fairall, C., , E. F. Bradley, , D. P. Rogers, , J. B. Edson, , and G. S. Yong, 1996: Bulk parameterization of air–sea fluxes for Tropical Ocean-Global Atmosphere Coupled-Ocean Atmosphere Response Experiment. J. Geophys. Res., 101 , 37473764.

    • Search Google Scholar
    • Export Citation
  • Goodberlet, M. A., , and C. T. Swift, 1992: Improved retrievals from the DMSP wind speed algorithm under adverse weather conditions. IEEE Trans. Geosci. Remote Sens., 30 , 10761077.

    • Search Google Scholar
    • Export Citation
  • Goodberlet, M. A., , C. T. Swift, , and J. C. Wilkson, 1989: Remote sensing of ocean surface winds with the Special Sensor Microwave/Imager. J. Geophys. Res., 94 , 1454714555.

    • Search Google Scholar
    • Export Citation
  • Hollinger, J., , R. Lo, , G. Poe, , R. Savage, , and J. Pierce, 1987: The Special Sensor Microwave/Imager user’s guide. NRL Tech. Rep., Naval Research Laboratory, Washington, DC, 120 pp.

  • Krasnopolsky, V. M., , L. C. Breaker, , and W. H. Gemmill, 1994: Development of a single “All-Weather” neural network for estimating ocean surface winds from the Special Sensor Microwave Imager. NMC/NOAA Tech. Note, OPC Contribution 94, 66 pp.

  • Krasnopolsky, V. M., , L. C. Breaker, , and W. H. Gemmill, 1995: A neural network as a nonlinear transfer function model for retrieving surface wind speeds from the Special Sensor Microwave Imager. J. Geophys. Res., 100 , 1103311045.

    • Search Google Scholar
    • Export Citation
  • Krasnopolsky, V. M., , W. H. Gemmill, , and L. C. Breaker, 1999: A multi-parameter empirical ocean algorithm for SSM/I retrievals. Environmental Modeling Center, National Centers for Environmental Prediction Tech. Rep., Washington, DC, 36 pp.

  • Krasnopolsky, V. M., , W. H. Gemmill, , and L. C. Breaker, 2000: A neural network multiparameter algorithm for SSM/I ocean retrievals: Comparisons and validations. Remote Sens. Environ., 73 , 133142.

    • Search Google Scholar
    • Export Citation
  • Liu, G. R., , C. C. Liu, , and T. H. Kuo, 2001: A contrast and comparison of near-sea surface air temperature/humidity from GMS and SSM/I data with an improved algorithm. IEEE Trans. Geosci. Remote Sens., 39 , 21482157.

    • Search Google Scholar
    • Export Citation
  • Liu, Q., , and C. Simmer, 1998: Retrieval algorithms for Special Sensor Microwave/Imager (SSM/I). Proc. SPIE, 3503 , 313321.

  • Petty, G. W., 1993: A comparison of SSM/I algorithms for the estimation of surface wind. Proc. Shared Processing Network DMSP SSM/I Algorithm Symp., Monterey, CA, NRL.

  • Stogryn, A. P., , C. T. Butler, , and T. J. Bartolac, 1994: Ocean surface wind retrievals from Special Sensor Microwave Imager data with neural networks. J. Geophys. Res., 99 , 981984.

    • Search Google Scholar
    • Export Citation
  • Thiria, S., , C. Mejia, , F. Badran, , and M. Crepon, 1993: A neural network approach for modelling nonlinear transfer functions: Application for wind retrieval from spaceborne scatterometer data. J. Geophys. Res., 98 , 2282722842.

    • Search Google Scholar
    • Export Citation
  • Weng, F., , and N. C. Grody, 1994: Retrieval of cloudy liquid water using the Special Sensor Microwave Imager (SSM/I). J. Geophys. Res., 99 , 2553225551.

    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., 1992: Measurement of oceanic wind vector using satellite microwave radiometers. IEEE Trans. Geosci. Remote Sens., 30 , 960972.

    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., 1997: A well calibrated ocean algorithm for SSM/I. J. Geophys. Res., 102 , 87038708.

  • Xia, Y., 2001: Sea-surface temperature and sea-surface wind-speed retrievals from spaceborne radiometer measurements. Ph.D. dissertation, University of Massachusetts—Amherst, 105 pp.

  • Xin, W., , L. Zhou, , D. L. Wang, , and X. Y. Xiong, 2002: Application and Design of Artificial Neural Networks in MATLAB. (in Chinese). Science Press, 333 pp.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The location of the NDBC and TAO buoys.

  • View in gallery

    The topological configuration of the NN.

  • View in gallery

    The SSM/I–buoy comparison (within 10 km and 0.5 h) for (a) surface WS, (b) SST, (c) Ta, and (d) Td over global oceans for 11 624 data points from 1997 to 2002. Here, A is the NN output and T is the buoy measurement.

  • View in gallery

    The comparison of the converted RH from the NN retrieval and the buoy measurement for 11 624 data points.

  • View in gallery

    The comparison of the SSM/I and buoy heat fluxes for 8326 data points: (a) sensible and (b) latent heat flux.

  • View in gallery

    The SSM/I (dotted line) and the Xisha (solid line) heat flux comparisons for 71 data points: (a) sensible and (b) latent heat flux.

  • View in gallery

    The heat fluxes from SSM/I data using an NN vs heat fluxes from Xisha data: (a) sensible and (b) latent heat flux.

  • View in gallery

    The daily mean fields of the SSM/I (a) surface WS, (b) SST, (c) Ta, and (d) surface Td from the NN for 2 Jan 2000.

  • View in gallery

    (Continued)

  • View in gallery

    The daily mean fields of the SSM/I (a) sensible and (b) latent heat fluxes from the NN for 2 Jan 2000.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 33 33 3
PDF Downloads 17 17 1

Neural Network Retrieval of Ocean Surface Parameters from SSM/I Data

View More View Less
  • 1 Institute of Oceanology, Chinese Academy of Sciences, Qingdao, China
© Get Permissions
Full access

Abstract

A new algorithm based on the multiparameter neural network is proposed to retrieve wind speed (WS), sea surface temperature (SST), sea surface air temperature, and relative humidity (RH) simultaneously over the global oceans from Special Sensor Microwave Imager (SSM/I) observations. The retrieved geophysical parameters are used to estimate the surface latent heat flux and sensible heat flux using a bulk method over the global oceans. The neural network is trained and validated with the matchups of SSM/I overpasses and National Data Buoy Center buoys under both clear and cloudy weather conditions. In addition, the data acquired by the 85.5-GHz channels of SSM/I are used as the input variables of the neural network to improve its performance. The root-mean-square (rms) errors between the estimated WS, SST, sea surface air temperature, and RH from SSM/I observations and the buoy measurements are 1.48 m s−1, 1.54°C, 1.47°C, and 7.85, respectively. The rms errors between the estimated latent and sensible heat fluxes from SSM/I observations and the Xisha Island (in the South China Sea) measurements are 3.21 and 30.54 W m−2, whereas those between the SSM/I estimates and the buoy data are 4.9 and 37.85 W m−2, respectively. Both of these errors (those for WS, SST, and sea surface air temperature, in particular) are smaller than those by previous retrieval algorithms of SSM/I observations over the global oceans. Unlike previous methods, the present algorithm is capable of producing near-real-time estimates of surface latent and sensible heat fluxes for the global oceans from SSM/I data.

Corresponding author address: Dr. Yijun He, Institute of Oceanology, Chinese Academy of Sciences, Qingdao 266071, China. Email: heyj@ms.qdio.ac.cn

Abstract

A new algorithm based on the multiparameter neural network is proposed to retrieve wind speed (WS), sea surface temperature (SST), sea surface air temperature, and relative humidity (RH) simultaneously over the global oceans from Special Sensor Microwave Imager (SSM/I) observations. The retrieved geophysical parameters are used to estimate the surface latent heat flux and sensible heat flux using a bulk method over the global oceans. The neural network is trained and validated with the matchups of SSM/I overpasses and National Data Buoy Center buoys under both clear and cloudy weather conditions. In addition, the data acquired by the 85.5-GHz channels of SSM/I are used as the input variables of the neural network to improve its performance. The root-mean-square (rms) errors between the estimated WS, SST, sea surface air temperature, and RH from SSM/I observations and the buoy measurements are 1.48 m s−1, 1.54°C, 1.47°C, and 7.85, respectively. The rms errors between the estimated latent and sensible heat fluxes from SSM/I observations and the Xisha Island (in the South China Sea) measurements are 3.21 and 30.54 W m−2, whereas those between the SSM/I estimates and the buoy data are 4.9 and 37.85 W m−2, respectively. Both of these errors (those for WS, SST, and sea surface air temperature, in particular) are smaller than those by previous retrieval algorithms of SSM/I observations over the global oceans. Unlike previous methods, the present algorithm is capable of producing near-real-time estimates of surface latent and sensible heat fluxes for the global oceans from SSM/I data.

Corresponding author address: Dr. Yijun He, Institute of Oceanology, Chinese Academy of Sciences, Qingdao 266071, China. Email: heyj@ms.qdio.ac.cn

1. Introduction

The Special Sensor Microwave Imager (SSM/I) was first flown on the Defense Meteorological Satellite Program (DMSP) F8 satellite in June 1987 (Hollinger et al. 1987). Since then, six SSM/I sensors have been launched successfully, and currently there are three sensors onboard the DMSP F1315. Our analysis is based on the observations by the SSM/I onboard the DMSP F14 between 1997 and 2002. The scan direction of the SSM/I on DMSP F14 is from left to right with a spatial resolution of 25 km. The SSM/I provides dual-polarization measurements at 19.55, 37.0, and 85.5 GHz and a vertical-polarization measurement at 22.235 GHz, with a 1-K accuracy of microwave energy in units of radiometric brightness temperature.

The SSM/I has gained widespread applications in the ocean and atmosphere system in the past decades. SSM/I observations have been used to obtain long time series of geophysical parameters over the global oceans such as water vapor, liquid water, rainfall, and ice concentration, which are very important to the global hydrologic cycle. They can also be used to retrieve ocean surface parameters such as sea surface wind, sea surface temperature, sea surface air temperature (Ta), and relative humidity. These parameters, in turn, give estimates of latent and sensible heat fluxes, freshwater flux, and surface stress (Liu and Simmer 1998; Krasnopolsky et al. 1999; Wentz 1997).

The surface sensible and latent heat fluxes over the global oceans are important to a wide variety of atmospheric and oceanic problems. They have been used to drive ocean general circulation models and to validate coupled ocean–atmosphere simulations for forecasting purposes (Chou et al. 1997). The majority of the existing works, however, focus on the monthly means of global latent and sensible heat fluxes. Surface heat fluxes of high spatial and temporal resolution derived from satellite measurements of surface wind, SST, humidity, and specific humidity are currently lacking in the literature.

There are mainly three types of retrieval algorithms for SSM/I: the physical, empirical retrieval, and neural network (NN) algorithm. Wentz (1992, 1997) developed a physical algorithm for retrieving the near-surface wind speed (WS), water vapor, and cloud liquid water and the line-of-sight wind over the global oceans. This method uses a large amount of in situ data for parameterizations and therefore has high accuracy. Nonetheless, this method requires SST as an external input and consequently cannot retrieve parameters such as wind speed at real time. The empirical retrieval algorithms have been developed separately for various geophysical parameters such as surface wind speed (Goodberlet et al. 1989; Goodberlet and Swift 1992; Petty 1993), water vapor (Alishouse et al. 1990), cloudy liquid water (Weng and Grody 1994), near sea-surface air temperature/humidity (Chou et al. 1995; Liu et al. 2001), etc. These empirical algorithms are usually derived by matching the SSM/I brightness temperature with the buoy and/or radiosonde measurements.

The neural network method is an empirical retrieval algorithm based on a highly connected array of elementary processors called neurons. Thiria et al. (1993) introduced the neural network method as an alternative to nonlinear regression to capture the nonlinear behavior of radiation transfer function without prior knowledge of its nonlinearity. Then a single-parameter neural network algorithm was developed to retrieve the sea surface wind speed from SSM/I data (Stogryn et al. 1994; Krasnopolsky et al. 1994, 1995). Krasnopolsky et al. (1999) developed a multiparameter neural network algorithm to retrieve sea surface temperature, water vapor, liquid water, and wind speed over the ocean simultaneously. This neural network consists of 5 input parameters, 1 hidden layer with 12 neurons, and 4 output parameters. Xia (2001) used a similar method to retrieve sea surface temperature, wind speed, relative wind direction, and water vapor from SSM/I data. Based on a neural network, Dourras et al. (2002) proposed a method to calculate mesoscale instantaneous latent heat flux from SSM/I observations, which gives rms errors ranging from 25 to 45 W m−2.

The objective of this paper is to retrieve the sea surface wind speed, sea surface temperature, sea surface air temperature, and relative humidity (RH) over the global oceans simultaneously at real time. Through a bulk method, these geophysical parameters are then used to estimate the global sea surface latent and sensible heat flux (Chou et al. 1997; Chu et al. 2006).

This paper is arranged as follows: in section 2, data are described for the neural network retrieval of surface geophysical parameters. In section 3, a neural network model is constructed for retrieving four surface parameters, WS, SST, Ta, and RH. The bulk method for estimating the surface fluxes of latent and sensible heat flux from these retrieved geophysical parameters is also discussed in this section. In section 4, the neural network retrievals of these parameters are compared with buoy observations. The retrieved surface heat flux is then compared with both buoy observations and observations at Xisha Island in the South China Sea. In section 5, the daily mean fields of WS, SST, Ta, dewpoint temperature (Td), and sensible and latent heat fluxes on 2 January 2000 are demonstrated. Finally, the major findings and conclusions of this study are summarized in section 6.

2. Data

The training and validation data of the neural network were derived from the matchups of SSM/I observations and buoy measurements. The buoy data were obtained from the National Data Buoy Center (NDBC) and the Tropical Atmosphere Ocean Project (TAO). We selected 71 NDBC buoys located in the northwestern Pacific, Gulf of Mexico, northwestern Atlantic, and near Hawaii and 62 TAO buoys located in the equatorial Pacific and equatorial Atlantic. The time interval for the data was from 1997 to 2002. These selected buoys are at least 30 km from the coast to avoid the contamination of the sea surface brightness temperature by land (Goodberlet et al. 1989). The buoy locations are shown in Fig. 1.

These buoys provide hourly measurements of WS, SST, Ta, and Td. Wind speeds at various heights above the sea surface, measured by anemometers on these buoys, are converted to an equivalent wind speed at 19.5 m above the sea surface using the formula (Wentz 1997)
i1520-0493-135-2-586-e1
where
i1520-0493-135-2-586-eq1
and w(H) and w(19.5) are the wind speeds at heights of H and 19.5 m, respectively. The dewpoint temperature is then converted to the RH using the relationship
i1520-0493-135-2-586-e2
The SSM/I sensor nearly covers the global oceans within 2–3 days. In this paper, we use seven SSM/I channels of brightness temperature to retrieve the geophysical parameters. We choose the SSM/I data in clear and cloudy weather conditions. The weather conditions are based on the following criteria for the brightness temperatures of four SSM/I channels (Thiria et al. 1993):
i1520-0493-135-2-586-e3
i1520-0493-135-2-586-e4

The buoy measurements and the SSM/I observations are paired within a spatial interval of 10 km and a time interval of 30 min. The matchup parameters include the longitude and the latitude of the buoy, the buoy measurements of WS, SST, Ta, RH, and the seven brightness temperatures measured by the SSM/I. As a result, 23 247 pairs of SSM/I overpasses and buoy data were obtained in both clear and cloudy weather conditions. These pairs were then randomly divided into datasets A and B, which consist of 11 623 and 11 624 pairs of data, respectively. Dataset A was used to train the neural network while dataset B was used to verify the retrieval results of the neural network. There were 18 889 data pairs in clear weather conditions and 4358 pairs in cloudy weather conditions.

3. Methodology

a. Retrieval of the WS, SST, Ta, Td, and RH

In this section, a neural network method is developed to retrieve four surface geophysical parameters—wind speed, sea surface temperature, sea surface air temperature, and relative humidity. The neural network is a fully connected, forward feeding, multilayered perceptron that consists of one input layer, one hidden layer, and one output layer. The input layer has seven neurons that correspond to the brightness temperatures given by the seven channels of the SSM/I. The hidden layer has 30 neurons and the output layer has 4 neurons that correspond to 4 output parameters—WS, SST, Ta, and Td. The output parameter, Td, is then converted to RH using Eq. (2). The configuration of the neural network is shown in Fig. 2.

The brightness temperatures of the seven SSM/I channels and the WS, SST, Ta, and Td in dataset A were used as the input variables and targets to train the neural network using the Levenberg–Marquardt (LM) back-propagation algorithm (Xin et al. 2002). This algorithm is a second-order nonlinear optimization technique. It is more efficient and produces better results than other training methods. The neural network was trained for 200 steps.

b. The computation of heat flux

In this study, the sensible and latent heat fluxes at the air–sea interface over the global oceans were derived using an improved bulk scheme (Fairall et al. 1996; Chu et al. 2006). The sensible and latent heat fluxes were computed from the SSM/I measurements of WS, SST, Ta, and RH at 10 m above the sea surface (mab). The NN retrieval of SSM/I wind speed at 19.5 mab was converted to the wind speed at 10 mab using a relationship similar to Eq. (1); that is,
i1520-0493-135-2-586-e5
where
i1520-0493-135-2-586-eq2
and w(H) and w(10) are the wind speeds at heights of H and 10 m, respectively.

The RH was computed from the SSM/I measurement of Td, using Eq. (2). Thus, all of the geophysical parameters required to compute surface latent and sensible heat fluxes over the global oceans can be derived from the SSM/I brightness temperatures using the neural network. The accuracy of the retrieved geophysical parameters and sensible and latent heat fluxes will be discussed in detail in the next section.

4. Comparisons and validations

a. Validation of the neural network retrievals

To assess the performance of the present neural network model, we validate the neural network retrieval of wind speed, sea surface temperature, surface air temperature, and surface dewpoint temperature from SSM/I data against the buoy observations. With the criteria specified in section 2, the 11 624 pairs of SSM/I and buoy data in dataset B for the global oceans from 1997 to 2002 are used in the validation. For each SSM/I–buoy data pair, the brightness temperatures of seven channels were used as input variables of the neural network and the four geophysical parameters, WS, SST, Ta, and Td, were obtained as output variables from the model.

Figure 3 shows the comparisons of SSM/I and buoy measurements of wind speed, sea surface temperature, surface air temperature, and surface dewpoint temperature over the global ocean from 1997 to 2002. The SSM/I wind speed has a bias of −0.26 m s−1, an rms error of 1.48 m s−1, and a correlation coefficient of 0.84 with buoy measurements (Fig. 3a). The rms error of SSM/I wind speed is smaller than the rms error of 2.1 m s−1 given by Wentz (1997) and that of 1.7 m s−1 by Krasnopolsky et al. (2000). The retrieved wind speeds range from 0 to 20 m s−1, and 97.3% and 91.9% of the retrieved wind speeds have errors less than 3 and 2 m s−1, respectively.

Among the 11 624 data pairs, only 5 retrieved wind speeds are less than zero, which constitutes about 0.04% of all retrieved wind speeds. The best linear regression between NN retrievals of wind speed and the buoy measurements is A = 0.705T + 1.95, where A is the wind speed of NN and T is the buoy wind speed.

The SSM/I sea surface temperature has a bias of 0.07°C, an rms error of 1.54°C, and a correlation coefficient of 0.94 in comparison with buoy measurements. Here, 96.4% and 92.5% of the retrieved SST have errors less than 3° and 2°C, respectively. The SSM/I surface air temperature has a small bias of 0.16°C, an rms error of 1.47°C, and a correlation coefficient of 0.95, with 96.8% and 92.6% of the results of Ta having errors less than 3° and 2°C, respectively. Figure 3b shows that the best linear regression between the neural network retrieval of SST and the buoy SST is A = 0.893T + 2.8, where A is the SST of the NN and T is the SST of the buoy, and the best linear fit of Ta is A = 0.913T + 2.22 (Fig. 3c), where A is the Ta of the NN and T is the Ta of the buoy.

The SSM/I dewpoint temperature has a bias of 0.21°C, an rms error of 1.51°C, and a correlation coefficient of 0.94 (Fig. 3d), with 96.4% and 93.0% of the results of Td having errors less than 3° and 2°C, respectively. The best linear fit of Td is A = 0.913T + 1.88, where A is the Td of the NN and T is the Td of the buoy. Using Eq. (2), the dewpoint temperature was converted into the relative humidity. Figure 4 shows that the NN retrieval of relative humidity has a bias of −0.48 and an rms error of 7.85 in comparison with buoy measurements.

b. Validation of the surface heat flux retrievals

1) Comparisons with buoy observations

We compare the SSM/I heat fluxes with the buoy measurements from 1997 to 2002 to assess the retrieval accuracy of the present method. The TAO data consisting of 8326 data pairs under both clear and cloudy weather conditions are used in the validation. The TAO buoys are located mainly in the tropical Pacific and Atlantic Oceans (see Fig. 1).

Figure 5 shows the comparisons of the SSM/I sensible and latent heat fluxes with the buoy measurements from 1997 to 2002. The SSM/I sensible heat flux agrees well with the buoy measurements with a bias of 0.2 W m−2 and an rms error of 4.9 W m−2. The SSM/I latent heat flux has a bias of 0.3 W m−2 and an rms error of 37.85 W m−2. Figure 5a shows that the SSM/I sensible heat flux ranges from −5 to 15 W m−2, with a mean value of 3.89 W m−2, whereas the buoy sensible heat flux ranges from 0 to 15 W m−2 with a mean value of 3.69 W m−2. Figure 5b shows that both the SSM/I and buoy latent heat fluxes range from 0 to 300 W m−2 and that they have mean values of 85.20 and 84.87 W m−2, respectively.

2) Comparisons with Xisha data

The Xisha data were collected at the air–sea flux observational station at Xisha Yong-xing Island (16°50′N and 112°20′E) of the South China Sea and used to validate the SSM/I heat fluxes. The Xisha station is located about 300 m away from the island to exclude the land effects from the observations. The station gives measurements of WS (10 m), SST, Ta, and RH every half hour. There were 71 passes of SSM/I over the Xisha station from 1 May to 31 May 2002. The brightness temperatures of SSM/I were used as input variables to the neural network described in section 3, and the NN output variables were used to compute the heat fluxes.

Figures 6 and 7 show the comparisons of 71 SSM/I sensible and latent heat fluxes with the Xisha observations. It is evident from Figs. 6a and 7a that the SSM/I sensible heat fluxes underestimate the Xisha observations. This is likely due to the small mean values of sensible heat flux (3.41 W m−2 for the SSM/I and 5.20 W m−2 for the Xisha data) and the large variances (between 2 and 12 W m−2) arising from the weak wind (the mean wind speed was 5.15 m s−1 for the Xisha data). The rms error and bias between the SSM/I estimates and Xisha observations of sensible heat fluxes are 3.21 and 0.003 W m−2, respectively.

As shown in Figs. 6b and 7b, there is good agreement between the SSM/I and Xisha estimates of latent heat fluxes. There is only a small bias of 0.0003 W m−2 and an rms error of 30.54 W m−2 in the SSM/I latent heat flux.

5. Fields of SSM/I surface parameters

Figure 8 shows the NN-retrieved mean fields of WS, SST, Ta, and Td for 2 January 2000 from the SSM/I brightness temperature. High wind speeds (>12 m s−1) are found in the northwest Atlantic and Pacific Oceans. High wind speeds up to 16 m s−1 are also found in the Southern Ocean. Figures 8b –d show that the retrieved SST, Ta, and Td have a mean value of about 30°C in the tropical oceans.

Figure 9 shows the daily mean fields of sensible and latent heat fluxes calculated from the NN retrievals of WS, SST, Ta, and Td. The sensible and latent heat fluxes are small over the tropical oceans. The maximum sensible heat and latent fluxes occur at the northwestern Atlantic and Pacific Oceans because the offshore winds in these regions bring cold, dry continental air over the warm Kuroshio and the Gulf Stream (Chou et al. 1997).

6. Conclusions

A neural network method was proposed to retrieve the sea surface parameters of wind speed, sea surface temperature, surface air temperature, and surface dewpoint temperature from SSM/I observations. The brightness temperatures of the seven SSM/I channels with a spatial resolution of 25 km were used as the input variables for the neural network. The sensible and latent heat fluxes were computed from the SSM/I measurements of wind speed, sea surface temperature, surface air temperature, and surface relative humidity using a bulk method. The relative humidity was converted from the SSM/I dewpoint temperature.

The neural network was trained and validated using NDBC and TAO buoy measurements from 1997 to 2002. The rms errors for the NN-retrieved WS, SST, Ta, and Td are 1.48 m s−1, 1.54°C, 1.47°C, and 1.51°C, respectively. The rms errors between SSM/I and buoy measurements of sensible and latent heat fluxes are 4.9 and 37.85 W m−2, respectively, whereas between SSM/I and Xisha data, they are 3.21 and 30.54 W m−2, respectively. The daily fields of WS, SST, Ta, Td, and the sensible and latent heat fluxes over the global oceans are used to obtain the spatial variations of these parameters.

The rms errors of the present algorithm are smaller than those of previous SSM/I retrieval algorithms over the global oceans. The present neural network model may be further improved by (i) incorporating the effects of the wind directions, (ii) including high wind speeds and the mid-/high-latitude events in the training dataset, and (iii) adding the atmosphere parameters, such as water vapor, rainfall, and liquid water, to the outputs of the neural network.

Acknowledgments

The authors thank the two reviewers for helpful suggestions, Drs. Dongliang Yuan and Qingping Zou for their revisions, and the National Data Buoy Center and the Pacific Marine Environmental Laboratory for providing the NDBC and TAO buoy data. The authors also thank the Global Hydrology Resource Center for providing the SSM/I brightness temperature data. This work was supported by the National High Technology Program through Grant 2001AA633060.

REFERENCES

  • Alishouse, J. C., , S. Snyder, , J. Vongsathorn, , and R. R. Ferraro, 1990: Determination of oceanic total precipitable water from the SSM/I. IEEE Trans. Geosci. Remote Sens., 28 , 811866.

    • Search Google Scholar
    • Export Citation
  • Chou, S. H., , R. M. Atlas, , C. L. Shie, , and J. Ardizzone, 1995: Estimates of surface humidity and latent heat fluxes over oceans from SSM/I data. Mon. Wea. Rev., 123 , 24052425.

    • Search Google Scholar
    • Export Citation
  • Chou, S. H., , C. L. Shie, , R. M. Atlas, , and J. Ardizzone, 1997: Air–sea fluxes retrieved from Special Sensor Microwave Imager data. J. Geophys. Res., 102 , C6. 1270512726.

    • Search Google Scholar
    • Export Citation
  • Chu, J. T., , J. N. Chen, , and L. Y. Xu, 2006: Study of the algorithm of calculating turbulent heat fluxes at air–sea interface and applying it to the neighboring seas of China. Oceans Limnol., in press.

    • Search Google Scholar
    • Export Citation
  • Dourras, D., , L. Eymard, , and W. T. Liu, 2002: A neural network to estimate the latent heat flux over oceans from satellite observations. Int. J. Remote Sens., 23 , 24052423.

    • Search Google Scholar
    • Export Citation
  • Fairall, C., , E. F. Bradley, , D. P. Rogers, , J. B. Edson, , and G. S. Yong, 1996: Bulk parameterization of air–sea fluxes for Tropical Ocean-Global Atmosphere Coupled-Ocean Atmosphere Response Experiment. J. Geophys. Res., 101 , 37473764.

    • Search Google Scholar
    • Export Citation
  • Goodberlet, M. A., , and C. T. Swift, 1992: Improved retrievals from the DMSP wind speed algorithm under adverse weather conditions. IEEE Trans. Geosci. Remote Sens., 30 , 10761077.

    • Search Google Scholar
    • Export Citation
  • Goodberlet, M. A., , C. T. Swift, , and J. C. Wilkson, 1989: Remote sensing of ocean surface winds with the Special Sensor Microwave/Imager. J. Geophys. Res., 94 , 1454714555.

    • Search Google Scholar
    • Export Citation
  • Hollinger, J., , R. Lo, , G. Poe, , R. Savage, , and J. Pierce, 1987: The Special Sensor Microwave/Imager user’s guide. NRL Tech. Rep., Naval Research Laboratory, Washington, DC, 120 pp.

  • Krasnopolsky, V. M., , L. C. Breaker, , and W. H. Gemmill, 1994: Development of a single “All-Weather” neural network for estimating ocean surface winds from the Special Sensor Microwave Imager. NMC/NOAA Tech. Note, OPC Contribution 94, 66 pp.

  • Krasnopolsky, V. M., , L. C. Breaker, , and W. H. Gemmill, 1995: A neural network as a nonlinear transfer function model for retrieving surface wind speeds from the Special Sensor Microwave Imager. J. Geophys. Res., 100 , 1103311045.

    • Search Google Scholar
    • Export Citation
  • Krasnopolsky, V. M., , W. H. Gemmill, , and L. C. Breaker, 1999: A multi-parameter empirical ocean algorithm for SSM/I retrievals. Environmental Modeling Center, National Centers for Environmental Prediction Tech. Rep., Washington, DC, 36 pp.

  • Krasnopolsky, V. M., , W. H. Gemmill, , and L. C. Breaker, 2000: A neural network multiparameter algorithm for SSM/I ocean retrievals: Comparisons and validations. Remote Sens. Environ., 73 , 133142.

    • Search Google Scholar
    • Export Citation
  • Liu, G. R., , C. C. Liu, , and T. H. Kuo, 2001: A contrast and comparison of near-sea surface air temperature/humidity from GMS and SSM/I data with an improved algorithm. IEEE Trans. Geosci. Remote Sens., 39 , 21482157.

    • Search Google Scholar
    • Export Citation
  • Liu, Q., , and C. Simmer, 1998: Retrieval algorithms for Special Sensor Microwave/Imager (SSM/I). Proc. SPIE, 3503 , 313321.

  • Petty, G. W., 1993: A comparison of SSM/I algorithms for the estimation of surface wind. Proc. Shared Processing Network DMSP SSM/I Algorithm Symp., Monterey, CA, NRL.

  • Stogryn, A. P., , C. T. Butler, , and T. J. Bartolac, 1994: Ocean surface wind retrievals from Special Sensor Microwave Imager data with neural networks. J. Geophys. Res., 99 , 981984.

    • Search Google Scholar
    • Export Citation
  • Thiria, S., , C. Mejia, , F. Badran, , and M. Crepon, 1993: A neural network approach for modelling nonlinear transfer functions: Application for wind retrieval from spaceborne scatterometer data. J. Geophys. Res., 98 , 2282722842.

    • Search Google Scholar
    • Export Citation
  • Weng, F., , and N. C. Grody, 1994: Retrieval of cloudy liquid water using the Special Sensor Microwave Imager (SSM/I). J. Geophys. Res., 99 , 2553225551.

    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., 1992: Measurement of oceanic wind vector using satellite microwave radiometers. IEEE Trans. Geosci. Remote Sens., 30 , 960972.

    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., 1997: A well calibrated ocean algorithm for SSM/I. J. Geophys. Res., 102 , 87038708.

  • Xia, Y., 2001: Sea-surface temperature and sea-surface wind-speed retrievals from spaceborne radiometer measurements. Ph.D. dissertation, University of Massachusetts—Amherst, 105 pp.

  • Xin, W., , L. Zhou, , D. L. Wang, , and X. Y. Xiong, 2002: Application and Design of Artificial Neural Networks in MATLAB. (in Chinese). Science Press, 333 pp.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The location of the NDBC and TAO buoys.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 2.
Fig. 2.

The topological configuration of the NN.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 3.
Fig. 3.

The SSM/I–buoy comparison (within 10 km and 0.5 h) for (a) surface WS, (b) SST, (c) Ta, and (d) Td over global oceans for 11 624 data points from 1997 to 2002. Here, A is the NN output and T is the buoy measurement.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 4.
Fig. 4.

The comparison of the converted RH from the NN retrieval and the buoy measurement for 11 624 data points.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 5.
Fig. 5.

The comparison of the SSM/I and buoy heat fluxes for 8326 data points: (a) sensible and (b) latent heat flux.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 6.
Fig. 6.

The SSM/I (dotted line) and the Xisha (solid line) heat flux comparisons for 71 data points: (a) sensible and (b) latent heat flux.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 7.
Fig. 7.

The heat fluxes from SSM/I data using an NN vs heat fluxes from Xisha data: (a) sensible and (b) latent heat flux.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 8.
Fig. 8.

The daily mean fields of the SSM/I (a) surface WS, (b) SST, (c) Ta, and (d) surface Td from the NN for 2 Jan 2000.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 8.
Fig. 8.

(Continued)

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Fig. 9.
Fig. 9.

The daily mean fields of the SSM/I (a) sensible and (b) latent heat fluxes from the NN for 2 Jan 2000.

Citation: Monthly Weather Review 135, 2; 10.1175/MWR3292.1

Save