kirknorth / roapy Goto Github PK

Mapping a large number of radar gates (e.g., 1M+) and querying a kd-tree for the 400+ nearest neighbours quickly puts a lot of pressure on computer memory resources, particularly when computing the weighted average (e.g., np.ma.average) between these points for numerous radar fields. Unfortunately, this type of setup with this many points to consider is often required when mapping scanning weather radar data. Therefore we need to optimize this process, either through the use of buffers or more likely through heterogeneous techniques like OpenCL or multiprocessing.

I believe there is a potential memory leak in the Weight class. When mapping multiple KVNX volumes I notice that memory usage steadily increases. Even on my machine which has 32 GB of RAM this has become a big enough issue where swap memory starts being accessed.

Add support for multipass objective analysis schemes. According to Majcen et al. (2008) a multipass objective analysis of radial velocity data prior to dual-Doppler wind synthesis is probably worth the added computational cost, and that these improvements are even more apparent in higher-order fields such as vorticity and divergence.

Majcen, M., P. Markowski, Y. Richardson, D. Dowell, and J. Wurman, 2008. Multipass objective analyses of Doppler radar data. J. Atmos. Oceanic Technol.

We need to better handle radar beam propagation and provide radar pointing information (e.g., azimuth and elevation) in the output grid. As of now we parameterize radar beam propagation with the 4/3 Earth radius model, which is fine, but from this we need to derive the pointing information at each grid point.