Series No. 467
Subject: Ocean Surface Wind Speeds retrieved from SSM/I on board DMSP Satellites
THIS IS A NEW BULLETIN ON THIS SUBJECT
Created 02/21/2000; revised 04/07/2004.
This bulletin, prepared by Mr. W. H. Gemmill, Mr. L. D. Burroughs, Ms. V. M. Gerald and V.M. Krasnopolsky of the Ocean Modeling Branch (OMB), Environmental Modeling Center (EMC), National Centers for Environmental Prediction (NCEP), describes "NCEP improved" ocean surface wind speed data at 20 m above the sea surface, columnar liquid water data, and columnar water vapor data retrieved from the SSM/I sensor on the DSMP satellites, based on a methodology developed by Krasnopolsky, which is being prepared for distribution through AWIPS, FOS, and GTS in BUFR format.
These data are expected to become operational in late March 2000. They will be issued 4 times a day and contain 6 hours of data centered on the synoptic hours. The issuance times will be approximately 0400, 1000, 1600, and 2200 UTC. Data from all the currently operational DSMP satellites will be included. Currently, there are 3 operational satellites (f11, f13, and f14) in polar orbit. The orbit times are approximately 102 minutes with two orbits per day over any given area (one ascending and one descending) per satellite. A passive microwave sensor (SSM/I) is used with 7 channels of brightness temperature data available. Each satellite covers a swath of 1400 km with a footprint resolution of 40 km at a spacing of 25 km.
OCEAN SURFACE WIND SPEEDS RETRIEVED FROM SSM/I ON BOARD DSMP SATELLITES
W. H. Gemmill, V.M. Krasnopolsky, L. D. Burroughs, and V. M. Gerald
The Defense Meteorological Satellite Program (DMSP) satellite series are polar orbiters. which were first launched in 1987. Wind speed data at 20 m above the sea surface are derived from a 7 channel passive microwave sensor (SSM/I) and have been available to NCEP since the early 1990s. At present there are three DMSP satellites (f11, f13 & f14) in orbit. The SSM/I measures "brightness temperatures" from the ocean surface and intervening atmosphere from seven channels. From these brightness temperatures ocean surface wind speed, columnar water vapor, columnar liquid water, rain rate, sea surface temperature and sea-ice concentrations can be derived. These channels are 19 gHz (H, V), 22 GHZ (V), 37 GHZ (H, V) and 85 GHZ (H, V) where H is the horizontal polarization, V is the vertical polarization. Each satellite provides data that cover a wide swath (1400 km) with a footprint resolution of 40 km at a spacing of 25 km. Table 1 presents a summary of the characteristics of the SSM/I specifications. The global distribution of data for a six hour period is shown in Fig. 1.
2. ALGORITHM DEVELOPMENT
A number of algorithms or transfer functions have been developed which convert the various SSM/I channel brightness temperatures into surface wind speed. Goodberlet et al. (1989) developed a single linear algorithm for use on a global basis. This algorithm as designed and tested met the specified speed accuracy criteria (2 m/sec over the range of wind speeds between 3 and 20 m/sec), but under rain free and low moisture conditions. The accuracy of wind speeds retrieved using this algorithm deteriorates rapidly in areas where rain and heavy cloud cover occur. Wind speeds are flagged as such which leave large areas of no retrievals in active weather regimes (storms or frontal regions). These areas are the very regions where wind information is critical. Further, the algorithm was developed on the basis of match ups with buoys where there were almost no wind speeds above 18 m/s. As a result, there is large uncertainty at high wind speeds above 15 m/s. A water vapor correction has been added to the linear algorithm by Petty (1993) (GSWP). This data is available in real-time through the Shared Processing Center of the U.S. Navy.
Krasnopolsky et al. (1995) selected neural networks (NN) as an alternative method to estimate surface wind speed from the SSM/I brightness temperatures. The NN approach corresponds to a general nonlinear model for the transfer function, and does not require any a priori knowledge about the particular form of the input/output relationship. The current version of the NCEP NN algorithm (OMBNN3) is a multi-variate algorithm (Krasnopolsky et al., 1996, 1999) which retrieves ocean surface winds, columnar water vapor, columnar liquid water and SST based on brightness temperatures from 5 SSM/I channels. Application of NN algorithm has led to an improvement in wind speed retrieval accuracy for clear conditions by approximately 20%; for higher moisture/cloudy conditions, the improvement
Table 1. Satellite ocean surface wind specifications for
|Type||Polar Orbiter (~102 min/orbit)|
|Areal Coverage||Twice Daily (one ascending and one descending orbit)|
|Receptors||7 microwave channels|
|Frequency Bands||19(H, V), 22(V), 37(H, V), 85(H, V) gHz, where H is horizontal polarization and V is vertical polarization|
|Number of Data Cells||64|
|Cell Footprint (km)||40|
|Wind Height (m)||20|
|Range of Wind Speeds (m/s)||3 - 25|
|Speed Accuracy||±2 m/s up to 20 m/s and ±10% above 20 m/s|
|Direction Accuracy||not applicable|
|Algorithm used||GWSP and OMBNN3|
was better when compared to the GSWP algorithm, and there was an improvement in high wind speed retrievals (> 15 m/s) as well. The increase in areal coverage, due to improvement in accuracy, was about 15% on average and higher in areas with active weather systems (see Fig. 2). Further, using the data "images" of the three variables together provides a detailed synoptic interpretation of the weather over the ocean otherwise not apparent from just one of the variables alone (Gemmill and Krasnopolsky, 1999).
The accuracy of SST was not sufficient to be used as an output variable in itself since the SSM/I channels are not suitable for SST. But, there was enough SST signal in the brightness temperatures to be retrieved to contribute positively to the improvement of the wind speeds.
Statistics presented in Table 2 show RMS errors for different wind speed algorithms. Biases were similar at -0.5 m/s (GSWP) to -0.3 m/s (OMBNN3). NN algorithms obviously outperform all other algorithms in terms of standard deviations. OMBNN3 For the total performance over a wide range of wind speeds and weather situations, the OMBNN3 results appear to be the best in terms of bias, standard deviation, and for high wind speeds.
Table 2. Statistics for the operational wind speed algorithms: GSWP, and OMBNN3 for clear and clear plus cloudy conditions (in parentheses) for SSM/I satellite instruments matched with 15,000 buoy observations.
|Algorithm||Bias (m/s)||RMSE (m/s)||Wind > 15 m/s
|GSWP(1)||-0.2 (-0.5)||1.8 (2.5)||(2.7)|
|OMBNN3(2)||-0.2 (-0.3)||1.5 (1.7)||(2.3)|
Based on the improved performance of OMBNN3, ocean surface wind retrievals were incorporated into GDAS at NCEP in April 1998 (Yu et al., 1997).
3. PRODUCT DISSEMINATION
These data are presently on the Internet at http://polar.ncep.noaa.gov/marine.meteorology/marine.winds four times a day for the Northeast Pacific and Northwest Atlantic with panels for SSM/I neural network ocean surface wind speeds (Gemmill et al., 2000), columnar water vapor, and columnar liquid water data along with ERS2 wind vector data, ship and buoy data, and a sea level pressure analysis (see Fig. 3).
These data are expected to become operational in late March 2000. The WMO bulletin header is ISXX02 KWBC. They will be issued 4 times a day (0400, 1000, 1600, and 2200 UTC) and contain 6 hours of data centered on the synoptic hours. They will include the following information:
1) Satellite ID
9) Ocean surface (20 m) wind speed
10) Columnar water vapor
11) Columnar liquid water
The issuance times will be approximately 0400, 1000, 1600, and 2200 UTC. Global data from all the currently operational DSMP satellites will be included. Currently, there are 3 operational satellites (f11, f13, and f14) in polar orbit. The orbit times are approximately 102 minutes with two orbits per day over any given area (one ascending and one descending) per satellite. Each satellite covers a swath of 1400 km with a footprint resolution of 40 km at a spacing of 25 km.
Gemmill, W. H. and V. M. Krasnopolsky, 1999: The Use of SSM/I Data in Operational Marine Analysis, Wea. and Forecasting, 14, 789-800.
Goodberlet, M. A., C. T. Swift, and J. C. Wilkerson, 1989: Remote sensing of ocean surface winds with the Special Sensor Microwave/Imager, J. Geophys. Res., 94, 14,547- 14,555.
Krasnopolsky, V.M., L.C. Breaker, 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, No. C6, 11033-11045,
Krasnopolsky, V. M., W. H. Gemmill, and L. C. Breaker, 1996: A new transfer function for SSM/I based on an expanded neural network architecture. Tech Note, OMB Contr. No. 137, National Centers for Environmental Prediction, Washington, DC 38pp.
Krasnopolsky, V. M., W. H. Gemmill, and L. C. Breaker, 1999: A multi-parameter empirical ocean algorithm for SSM/I Retrievals. Can. J. Remote Sensing, 25, 486-503.
Petty, G. W., 1993: A comparison of SSM/I algorithms for the estimation of surface wind, Proc. Shared Processing Network, DMSP SSM/I Algorithm Symposium, Monterrey, CA.,8-10 June, 1993.
Yu, T.-W., M. D. Iredell, and D. Keyser 1997: Global data assimilation and forecast experiments using SSM/I wind speed data derived from a neural network algorithm. Wea. and Forecasting, 12, 859-865.
CAPTIONS - FIGURES
Figure 1. SSM/I wind speed coverage from satellite F13 for January 13, 2000 for the six hour period from 0300 to 0900 UTC.
Figure 2. Comparison of coverage between GSW and OMB NN for 2 DSMP satellites. Upper left GSW for f10; upper right OMB NN for f10; lower left GSW for f13, and lower right OMB NN for f13.
Figure 3. Satellite and other data available on the Internet: upper left - SSMI Wind Speed, upper center - SSMI Liquid Water, upper right - SSMI Water Vapor, lower left - Ship Winds, lower center - ERS2 Winds, and lower right - Surface Pressure Analysis.
1. Petty (1993)
2. Krasnopolsky et al. (1996)