12B.8 Prototype Precipitation Profiling Algorithms for the Tomorrow-R1 and Tomorrow-R2 Radars

Thursday, 31 August 2023: 9:45 AM
Great Lakes A (Hyatt Regency Minneapolis)
Stephen J. Munchak, Tomorrow.io, Boston, MA; and E. Nelson and R. Roy

On April 14, 2023, the Tomorrow-R1 Radar Pathfinder satellite launched aboard the SpaceX Transporter 7 mission. An identical satellite, Tomorrow-R2, is scheduled for launch in June. The Pathfinder satellites are focused on testing the radar and pulse sampling strategies on orbit, and feature a state-of-the-art, all-solid-state Ka-band (35.5-36 GHz) transceiver coupled to a rigid reflector antenna yielding a fixed pencil-beam geometry. The spacecraft will be maneuvered to collect data at a wide range of scan angles, nominally sampling precipitation profiles within a 40 degree field of view in the cross-track plane. These will be followed by additional satellites with wide-swath scanning capabilities, currently under development for launch in 2024.

Prototype level 2 precipitation profiling algorithms have been developed for use with the Pathfinder data. Two pathways are being pursued for these retrievals: 1) an optimal estimation approach (1DVAR) and 2) a quantile regression neural network (QRNN). The 1DVAR algorithm is similar to profiling algorithms developed for the TRMM Precipitation Radar (PR), GPM Dual-frequency Precipitation Radar (DPR), and Cloudsat Profiling Radar (CPR), and is primarily constrained by effect of path-integrated attenuation (PIA) on the normalized radar surface backscatter cross-section (NRCS). At Ka-band, the 1DVAR approach works well for light and moderate precipitation rates (< 10 mm/hr), but is under-constrained at heavier precipitation rates when the near-surface reflectivity profile and NRCS are completely attenuated, and becomes entirely dependent on a priori assumptions regarding the precipitation profile in these cases.

Because the QRNN approach implicitly uses profile characteristics related to the particle size distribution, it provides a higher correlation to the training data than the GPM Ka-only retrievals, and can provide unbiased estimates even at high precipitation rates with only Ka-band observations as input. It has a further advantage of providing well-calibrated uncertainty estimates, however, a drawback is that it retains the biases of the training dataset, and the retrieved profile may not necessarily be physically consistent with the observed reflectivity profile. Techniques to reconcile the QRNN and 1DVAR approaches are under development, and application to Tomorrow-R1 and Tomorrow-R2 observations will be presented.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner