Radar rainfall: Separating signal and noise fields to generate meaningful ensembles

For the purpose of generating meaningful stochastic ensembles of radar estimates of rainfall, a relatively simple and objective method of separating a radar rainfall image into signal and noise is described. An alternative noise field, with the same spectrum as the original noise, can then be simulated and combined with the signal field of each successive image, to generate an ensemble member for performing sensitivity studies. The method is based on identifying the appropriate wavelength in the power spectrum which defines the variance threshold used to separate noise from signal. The algorithm is explained and figures illustrate the efficacy of the procedure.

Research Highlights
► Take a sequence of radar rainfall images as error-free as possible (clutter etc).
► Fourier transform each image to obtain 2-D spectrum.
► Select a horizontal threshold to define and separate noise and signal spectra.
► Stochastically replicate alternative noise spectra from original.
► Combine simulated noise with original spatial signal to get ensemble member.