Monday, 28 August 2023: 2:30 PM
Great Lakes BC (Hyatt Regency Minneapolis)
The Saccadic Phased Array Radar Sampling (SPARS) algorithm, a novel technique for steering a phased-array radar (PAR) beam in sampling meteorological targets is introduced here. SPARS is inspired by the mechanism of saccades in human vision, whereby the eye samples a scene by rapidly targeting the retina toward different features of interest, taking narrow “snapshots” during momentary pauses (fixations), which are then synthesized by the brain into a coherent larger picture and higher-level understanding. Each individual snapshot, by necessity, has a narrow field of view, as evolution has followed a strategy of conserving high-resolution photoreceptors by concentrating them in a small region of the central human retina (the fovea) and using a form of agile adaptive sampling to steer them toward the most important features in the environment. Repositioning can take place in tens of milliseconds and reach angular speeds of up to 700 deg/sec. This approach is in stark contrast to the regularly spaced grid of the CCD photodetectors across the entire field of view of a digital camera, or a reflector-based weather radar performing a fixed raster scan strategy.
The core objective of SPARS, as with human vision, is to optimally match available remote sensing resources to external information content that is of interest to the observer. Our current implementation has three functional parts. The first part consists of a very fast ongoing raster scan (peripheral vision) that blankets the full observational sector in azimuth and elevation with a coarse grid of short dwell-time beams. Every few seconds, all data from the grid are ranked and sorted by descending relevance according to an “interest metric” to identify the best beam direction, which is then scheduled to be the next fixation point. Presently, the interest metric used is reflectivity magnitude. Concurrently, the second part of our implementation targets high spatial resolution longer-dwell beams in an RHI and sector PPI intersecting at the most recently determined fixation point (analogous to vision by the human fovea). In general, the SPARS fovea can be given any desired spatial shape. The budgeting of beam time between peripheral and fixation vision is configurable and currently partitioned 20 - 80% respectively, by interleaving peripheral and fixation vision, allowing high-resolution sampling to be continuously targeted toward the most interesting features as they evolve. The third part of our implementation augments the peripheral vision beams thus far described with a smaller set of additional beams having the same properties, but in a finer grid local to the most recent fixation point. Their purpose is to provide this region with an advantage in the determination of the next fixation point. This induces a tendency to track valued targets, emulating the “smooth pursuit” mechanism of human vision.
We present initial observations obtained by SPARS during both cold and warm season conditions having a mix of convective and stratiform characteristics. Additionally, we discuss our vision for the future, and the clear potential synergy between SPARS and machine learning techniques.
The core objective of SPARS, as with human vision, is to optimally match available remote sensing resources to external information content that is of interest to the observer. Our current implementation has three functional parts. The first part consists of a very fast ongoing raster scan (peripheral vision) that blankets the full observational sector in azimuth and elevation with a coarse grid of short dwell-time beams. Every few seconds, all data from the grid are ranked and sorted by descending relevance according to an “interest metric” to identify the best beam direction, which is then scheduled to be the next fixation point. Presently, the interest metric used is reflectivity magnitude. Concurrently, the second part of our implementation targets high spatial resolution longer-dwell beams in an RHI and sector PPI intersecting at the most recently determined fixation point (analogous to vision by the human fovea). In general, the SPARS fovea can be given any desired spatial shape. The budgeting of beam time between peripheral and fixation vision is configurable and currently partitioned 20 - 80% respectively, by interleaving peripheral and fixation vision, allowing high-resolution sampling to be continuously targeted toward the most interesting features as they evolve. The third part of our implementation augments the peripheral vision beams thus far described with a smaller set of additional beams having the same properties, but in a finer grid local to the most recent fixation point. Their purpose is to provide this region with an advantage in the determination of the next fixation point. This induces a tendency to track valued targets, emulating the “smooth pursuit” mechanism of human vision.
We present initial observations obtained by SPARS during both cold and warm season conditions having a mix of convective and stratiform characteristics. Additionally, we discuss our vision for the future, and the clear potential synergy between SPARS and machine learning techniques.

