Retrieving 2D Wind Field from Aliased Doppler Data by Means of Sliding Windows

In addition to estimating precipitation, Doppler weather radars are routinely used for measuring winds. The most popular wind products are a Cartesian display of the lowest-sweep Doppler data as well as Velocity Volume Processing (VVP) which produces a vertical profile of wind above the radar.

As a measurand, wind is subject to two major limitations in radar. Firstly, three-dimensional wind vectors become projected on radar beams hence only beam-directional components of wind are obtained. Second, the maximum unambiguous range of speed is limited by Nyquist velocity, a quantity inversely proportional to desired distance range.

In this paper, we present a method that retrieves the wind field in the original sweep geometry. Our approach is resembles that of VVP, involving trigonometric least-squares fitting. In this case derivatives instead of original values are fitted and Doppler dealiasing is not needed as preprocessing.

The basic idea is to assume locally uniform wind field and to find wind velocity that best explains the apparent speed change in azimuthal direction i.e. on adjacent beams. In this study, we have focused on lowest sweeps, yielding nearly-horizontal wind (u,v). The fitting is computed in original polar coordinates using a two-dimensional neighborhood window. The size of the window should be relatively large in azimuthal direction, for example 60 or 90 degrees, whereas rather small beam-directional range, say 3 kms, seems sufficient. The approximated wind velocity is obtained for every bin. The computational effort is minimized using a sliding window that accumulates a set of sums from which the wind is retrieved using a simple 2x2 matrix inversion. A quality index is readily obtained by means of the determinant.

The proposed method is rather insensitive to Doppler speed aliasing in locally uniform wind. The trick is in that the observed changes of wind speed in azimuthal direction are much smaller than the (unambiguous) speed range, hence differences can be treated unaliased. Respectively, the algorithm fails in convective precipitation, as velocity data becomes noisy within the applied window.

We illustrate the approach with examples on widespread precipitation. To some extent, the algorithm works also on areas of separate cells of weakened convection where droplet motion is driven only by advection by prevailing wind. We also discuss compositing and visualizing obtained wind fields.