Neural Networks for Pattern Recognition
Neural Networks for Pattern Recognition
Self-Organizing Maps
Hi-index | 0.00 |
We present a new methodology, which provides the sea surface rain rate by inverting the 85 GHz channel measurements (TB) of the SMM/I and TMI microwave radiometers. This high frequency channel has the advantage of a spatial resolution close to the size of rain cells. We used a neural network whose inputs were the vertical and horizontal polarized 85 GHz TBs, the output being the rain rate. The learning dataset was made of downscale ECMWF atmospheric parameters and the corresponding brightness temperatures computed through the use of the radiative transfer equations. The computed rain rate compared well with the standard SSM/I algorithm both at global and regional scales. The comparison with the rain rates retrieved by TRMM radar, with similar pixel areas, showed a correlation coefficient higher than 0.97. A major difficulty was finding suitable observations to validate our results.