Radar data in Met.3D for developers

This section gives an overview of how weather radar data is handled internally in Met.3D.

Radar Pipeline

Actors load radar data as an MNWPActorVariable. On top of other variable properties, radar variables also have MRadarProperties, and MRadarRegridProperties if they are regridded, for settings specific to radar data and regridding.

The radar variable data selected in the add variable dialog is loaded by MDWDHDF5RadarReader into an MRadarGrid object. This grid comprises the data of all elevations of one variable/location/time step, and contains an instance of MRadarElevation for each elevation. MRadarElevation holds the data and metadata specific to this elevation sweep.

Rendering

The radar actor iterates over all elevation in a radar grid and renders its grid cells as boxes. For each corner, elevation angle, azimuth angle and range along the radar beam are converted to longitude, latitude and pressure in the geometry shaders. The equations implemented in radarToLonLatPressure in radar_utils.fx.glsl to convert a point P are explained in the bachelor’s thesis 3D Visualization of Weather Radar Data in Met.3D by Susanne Fuchs in section 2.1.6 Determining the position of the target.

Fig. 35 Determining the pressure/height coordinate of P.

Fig. 36 Determining longitude and latitude of P.

Regridder: Radar to regular lon lat remapping

To use all of Met.3D’s actors for numerical weather prediction data, radar data can be interpolated to a regular lat lon grid, see Remapping from radar grid to regular lat lon grid.

The approach used is a nearest neighbour interpolation that transforms each grid point in the regular grid to radar coordinates and evaluates which radar grid cell contains this point. It is implemented in inverseTransformationInterpolation in MRadarGrid and also in lonLatWorldZToRadarIndices in volume_radar_utils.fx.glsl.

Height \(z_p\) is transformed to elevation angle \(\alpha\). For that, the transformation \(\alpha -> z_p\) described in the last section is solved for \(\alpha\):

(with d = direct distance between radar antenna and P)

\[\alpha = \arcsin(\frac{(R_e + h_0)^2 + d^2 - (R_e + z_p)^2}{2 * (R_e + h_0) * d}) - \frac{\pi - \beta}{2}\]

Longitude and latitude are converted to an azimuth angle in function gcAzimuth_deg by calculating the bearing angle as described in https://www.movable-type.co.uk/scripts/latlong.html in section Bearing.

The range bin is received by calculating the length of the curved ray and dividing it by the distance between two bins.

\[\begin{split}\beta &= \arcsin(d / (2 * R_R)) * 2 \\ rayDist &= \beta * R_R \\ binIndex &= floor((rayDist - rangeStart)/ binDistance)\end{split}\]

Direct Volume Rendering

DVR in the raycaster actor can be used for radar data regridded to a regular grid or radar data on the original grid. The first works the same as for NWP data. In the second case, to sample the radar volume at a world position, sampleRadarVolumeAtPos in volume_radar_utils.fx.glsl also uses the inverse transformation in lonLatWorldZToRadarIndices to find the closest radar cell. It uses a nearest neighbour approach.