Optimized Tropical Cyclone Wind Retrievals from ASCAT, OceanSAT-2, and QuikSCAT

Abstract: We utilize an artificial neural network technique (ANN) developed by [4] to produce optimized scatterometer wind vector fields around tropical cyclones. This technique retrieves optimized 12.5-km resolution wind speeds from scatterometer measurements in tropical cyclone (TC) conditions including in typical rainy conditions in such storms. ANN is similar to other neural network wind retrieval methods described in [5,6]. We apply ANN to two years of ASCAT scatterometer data and two years of OceanSAT-2 data, as well as data throughout the ten-year QuikSCAT mission as already reported in [4]. Retrieved wind data for QuikSCAT and OceanSAT-2 have been made public at http://tropicalcyclone.jpl.nasa.gov. Two years of ASCAT TC wind retrievals will be publicly released in early 2014. The ANN technique was trained on H*WIND analyses that included aircraft reconnaissance data in 2005 Atlantic basin tropical cyclones. Wind directions in the ANN retrievals were not changed from the current (version 3) JPL global wind vector product [2] that utilized the neural network rain correction for low to moderate winds from [5]. The ANN technique uses several types of scatterometer measurements as inputs to determine the wind speed, including backscatter, brightness temperature obtained from the noise channel of the scatterometer, and the nominal scatterometer wind speed. The ANN technique was shown to retrieve accurate wind speeds up to 40 m/s when compared with aircraft reconnaissance data, including GPS dropwindsondes and Stepped-Frequency Microwave Radiometer surface wind speed measurements, and global best track maximum wind speeds (Figure 1). The ANN technique removes positive biases with respect to best track intensity in the developing phase of tropical cyclones that occur in the nominal (version 2) JPL QuikSCAT product and reduces negative biases with respect to best track intensity that occur in the nominal product (both versions 2 and 3) during the most extreme period of the lifetime of intense TCs. The wind regime with the most notable improvement is 20-40 m/s (40-80 knots), with more modest improvement for higher winds. The improvement at lower winds is comparable to that achieved previously by the version 3 JPL global rain-corrected product. The net effect of all the wind speed improvements is a much better measurement of TC intensity in the ANN scatterometer winds than what has been previously available. When compared with wind speed data from aircraft reconnaissance flights in Atlantic hurricanes, the QuikSCAT ANN product exhibited a 1-2 m/s positive overall bias and a 3 m/s mean absolute error (Figure 1). The random error and systematic positive bias in the ANN scatterometer wind product is similar to that of the NOAA Hurricane Research Division H*WIND analyses when aircraft data are available for assimilation [3], [1]. This similarity may be explained by the fact that H*WIND data are used as ground truth to fit the coefficients used by the new technique to map radar measurements to wind speed and any systematic biases in the H*WIND training dataset could be reflected in the ANN retrievals. Similar comparisons for the OceanSAT-2 and ASCAT products will be shown. Also, we will further quantify biases between H*WIND, SFMR, dropwindsondes, and scatterometer winds in order to better characterize the scatterometer wind products and their utility for investigating the role of tropical cyclones in long term oceanic and atmospheric processes. The research described in this abstract was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. References: [1]DiNapoli, S. M., Bourassa, M. A., & Powell, M. D. (2012). Uncertainty and Intercalibration Analysis of H* Wind. J. Atmos. and Ocean. Tech., 29(6), 822-833, 2012. [2]Fore, A.G., Stiles, B.W., et al, “Point-wise Wind Retrieval and Ambiguity Removal Improvements for the QuikSCAT Climatological Data Set,” Accepted for publication in IEEE TGARS, doi:10.1109/TGRS.2012.2235843, 2013. [3]Powell, M. D., Sam H. Houston, et al, “The HRD real-time hurricane wind analysis system,” J. Wind Eng. and Ind. Aero. 77&78 pp 53-64, 1998.[4]Stiles, B.W., R. E. Danielson, et al. “Optimized Tropical Cyclone Winds from QuikSCAT: A Neural Network Approach,” submitted to IEEE TGARS, 2013.[5]Stiles, B. W., and R.S. Dunbar, "A Neural Network Technique For Improving The Accuracy Of Scatterometer Winds In Rainy Conditions," IEEE TGARS, Vol 48 , No. 8, pp. 3114-3122. 2010.[6] Stiles, B.W.; Hristova-Veleva, S.M.et al, "Obtaining Accurate Ocean Surface Winds in Hurricane Conditions: A Dual-Frequency Scatterometry Approach," IEEE TGARS , vol.48, no.8, pp.3101-3113, 2010. Figure 1: Comparison of QuikSCAT ANN wind speeds to other wind data sources: Top panel: 2-D Histogram of best track maximum speed vs. maximum speed estimated from QuikSCAT ANN data: Bottom left panel compares both QuikSCAT ANN wind field and H*WIND fields to Stepped Frequency Microwave Radiometer Bottom right panel compares both wind fields to GPS dropwindsondes. The number of pair-wise co-locations N, mean bias error (MBE), and mean absolute error (MAE) are reported for each case. Blue lines are H*WIND comparisons. Red lines are QuikSCAT ANN comparisons.