Figure Captions:

 

Figure 1. Spatial domain of the satellite-derived estimation of surface air temperature (TA) and specific humidity (Qa). The stars denote the locations of the moored buoys in the TAO and PIRATA domains as well the WHOI buoy in the Arabian Sea.  Temporal Domain for available SSM/I sensors.

 

Figure 2a. Schematic representation of the artificial neural network (ANN) architecture used to estimate TA. The first layer contains the input parameters of total precipitable water (TPW), sea surface temperature (SST), surface wind (Wind), and cloud liquid water (CLW) whereas the output layer contains the target TA. The architecture is denoted as 4-15-10-1x and contains fifteen nodes in the first hidden layer and 10 nodes in the second hidden layer.

 

Figure 2b. Schematic representation of the artificial neural network (ANN) architecture used to estimate QA. The first layer contains the input parameters of total precipitable water (TPW), sea surface temperature (SST), surface wind (Wind), cloud liquid water (CLW), and rain rate (RR) whereas the output layer contains the target QA. The architecture is denoted as 5-15-6-1x and contains fifteen nodes in the first hidden layer and 6 nodes in the second hidden layer.

 

Figure 3. Validation of TA estimated with ANN model and buoy observations for validation samples in the TAO and PIRATA domains (top left and right). Sample sizes (N), correlation coefficient, mean bias (satellite minus buoy) (°C) and rms errors (°C) are indicated in the figure. The comparison of NCEP/NCAR TA (2 m) and buoy observations (3 m) are shown in the bottom.

 

Figure 4. Same as in Fig 3, but for validation of Qa. Units are in g kg-1.

 

Figure 5. Same as in Fig 3, but for validation of LHF. Units are in W m-2.

 

Figure 6a. Spatial distribution of Ta rms (°C) errors between satellite estimates and buoy observations over 12 years in the TAO domain.

 

Figure 6b. Spatial distribution of Qa rms (g kg-1) errors between satellite estimates and buoy observations over 12 years in the TAO domain.

 

Figure 6c. Spatial distribution of number of observations used to calculate rms in Figures 6a and 6b.

 

Figure 6d. Spatial distribution of NCEP-ANN Ta and Qa rms errors over the TAO domain.

 

Figure 6e. Spatial distribution of LHF rms (W m-2) errors between satellite estimates and buoy observations over 12 years in the TAO domain.

 

Figure 6f. Spatial distribution of SHF rms (W m-2) errors between satellite estimates and buoy observations over 12 years in the TAO domain.

 

Figure 6g.Spatial distribution of number of observations used to calculate rms in Figures 6d and 6e.

 

Figure 6h.Spatial distribution of NCEP-ANN LHF and SHF rms errors over the TAO domain.

 

Figure 7a-h.  Same as in Fig. 6a-h, but for moored buoys in the PIRATA domain.

 

Figure 8.  15 year mean and standard deviation of satellite-derived Ta field.

Period: October 1987 to September 2002.

 

Figure 9.  15 year mean and standard deviation of satellite-derived Qa field.

Period: October 1987 to September 2002.

 

Figure 10. 15 year mean and standard deviation of satellite-derived LHF field.

Period: October 1987 to September 2002.

 

Figure 11. 15 year mean and standard deviation of satellite-derived SHF field.

Period: October 1987 to September 2002.

 

Figure 12. Relative sensitivity of input parameters on resulting LHF and SHF rms with respect to buoy data for ANN and NCEP inputs for the TOGA TAO domain.

 

Figure 13. Relative sensitivity of input parameters on resulting LHF and SHF rms with respect to buoy data for ANN and NCEP inputs for the Pirata domain.