Day 4

Detailed paper information

Back to list

Paper title Improvement of the Ku-band Winds in Rainy Regions by a Support Vector Machine Method
  1. Xingou Xu NSSC, CAS Speaker
  2. Ad Stoffelen KNMI
Form of presentation Poster
  • A8. Ocean
    • A8.13 Remote-sensing of Ocean Winds and Stress
Abstract text Ku-band observations from scatterometers are easier affected by rain due to their shorter wavelength than those collected at C-band, while both frequencies are commonly applied for wind scatterometers. We proposed a support vector machine (SVM) model based on the analysis of Quality Control (QC) indicators of rain screening ability, which has been validated by collocated winds from the Ku-band and C-band scatterometers: OSCAT-2 and ASCAT-B onboard the ScatSat and MetOp-B satellites respectively, together with simultaneous rain rates from the Global Precipitation Mission (GPM) products. Meanwhile, the principle of SVM was addressed for its advantages in the rain-effect correction problem. The established SVM model was evaluated by the testing set not applied in the training procedure. In the verification, where QC-accepted winds from the C-band collocations that are of QC accepted features are applied as the truth, given their low rain sensitivity.

In this research, first, the data sets are increased by including the collocations from the OCSAT-2 and the ASCAT-A scatterometers. The wind speed range applied for the model has been extended based on the recent update of the QC indicator, Joss, which is one of the inputs for the SVM. Then to validate the model, the probability density functions (pdf) and the features due to rains of the inputs due to rain are checked in more detail. The results of the SVM from the new test set, which is not applied in the training procedure, are analyzed specifically, in addition to the statistical features obtained comparing the resulting winds and the truth. Along with the pdf, cumulative density functions (CDF) are also checked. A case study is conducted with simultaneous references from the Medium Infrared advanced imager on board the Himawari-8 satellite.

We conclude that the corrected winds can provide improved quality information for Ku-band scatterometers under rain that can be vital for nowcasting applications, where the effectiveness of optimization methods based on Machine Learning for such problems is proven.
In this research, we also discuss the application of joint-SVMs for better representing wind-rain tangling problem and the possibility of resolving winds and rain rates in such model.