Fig. 1

Data processing flow of modeling and estimating SST-dependent GMF for HY-2A scatterometer"

Fig. 2

Distributions of speed and SST of the matching dataset. (a) mean of SST for different wind speeds, and (b) mean of wind speed for different SSTs"

Fig. 3

Topological architecture of neural network geophysical model with SST input. The input: p represents polarization, v represents wind speed, sinχ and cosχ respectively represent the sine and cosine of relative wind direction, and SST represents sea surface temperature. The output of σ0 represents the normalized radar cross-section (NRCS) from the sea surface"

Fig. 4

HY-2A scatterometer GMF without SST input. (a) HH polarization, and (b) VV polarization. The heavy and fine lines correspond to NGMF and NSCAT2, respectively"

Fig. 5

SST-dependent GMF for HY-2A scatterometer. (a) HH polarization, and (b) VV polarization. Each cluster of curves in the figure is correspondent to a certain speed, from 4 to 14 m·s-1 by an interval of 2 m·s-1"

Fig. 6

Mean of normalized radar cross-section varying with SST. (a) HH polarization, and (b) VV polarization. Dots and lines represent the mean of normalized radar cross-section from estimated data and TNGMF, respectively"

Fig. 7

Variation span of normalized radar cross-section affected by SST varying with wind speed"

Fig. 8

Temperature sensitivity varying with wind speed"

Fig. 9

TNGMF and NGMF wind speed bias mean varying with wind speed, for SSTs at 5 ℃ (a), 10 ℃ (b), 15 ℃ (c), 20 ℃ (d), 25 ℃ (e), and 30 ℃ (f)"

Fig. 10

TNGMF and NGMF wind speed bias mean varying with SST, for wind speed from 4 to 15 m·s-1"

Fig. 11

TNGMF and NGMF wind speed (a) and direction (b) RMSE varying with SST"

Fig. 12

TNGMF and NSCAT2 wind speed bias mean varying with wind speed, for SSTs at 5 ℃ (a), 10 ℃ (b), 15 ℃ (c), 20 ℃ (d), 25 ℃ (e), and 30 ℃ (f)"

Fig. 13

TNGMF and NSCAT2 wind speed bias mean varying with SST, for wind speed from 4 to 15 m·s-1"

Fig. 14

TNGMF and NSCAT2 wind speed (a) and direction (b) RMSE varying with SST"