Development and validation of a SAR wind emulator

It is realised that the spatial coverage and irregular revisit time of SAR sensors is not adequate for direct use of SAR-retrieved wind in operational weather forecasting or for assimilation into numerical forecast models. Therefore a new way is sought to take benefit from the obvious strength of SAR for high resolution wind speed retrieval. The approach investigated in this study is to develop a transfer function which can prescribe (emulate) a high resolution wind field from a given lower resolution weather forecast model field. Such a transfer function (or emulator) can be trained to “learn” the micrometeorological conditions of specific regions by co-locating a large historical dataset of model and SAR derived windfields. Once developed, this emulator can then be used in real time, or even forecast situation, to emulate the high resolution wind fields from the coarse model output, without the need for additional real time SAR data. The potential gain of such an emulator is largest in coastal regions with complex topography, where variations in wind speed and direction may be large on a fine scale. In this study, two regions with complex fjord-topography along the Norwegian coast are selected as study areas. Wind retrieval from Envisat ASAR Enivsat ASAR images are read and calibrated with Matlab software developed at NERSC. The backscatter images are downscaled from a pixel size of 75 metres to about 535 metres with an anti-aliasing lowpass filter (“resample” function in Matlab with N=10) to reduce noise, speckle and data volume. From these “quicklooks”, wind speed is calculated in this study with the CMOD4 algorithm (Stoffelen at al., 1997), where the auxiliary wind direction is taken from the NCEP Global Forecast System (GFS). The GFS model is initialised every 6 hours, and provides forecast output for every 3 hours. The 0 hour (initialization) or 3 hour forecast field, whichever is closest in time to the SAR image, is used for the SAR wind inversion. The pixel size of NCEP GFS is 0.5 degrees, and the wind direction is interpolated (component by component) to the ASAR grid with linear interpolation before the SAR wind retrieval.NERSC Tehnical report no. 304. ESRIN/Contract no. 18709/05/I-L