NOAA plans to build a Geostationary Lightning Mapper (GLM) whose objectives are providing continuous, full-disk
lightning measurements for storm warning and science applications. Due to limited telemetry bandwidth, much of the
detection processing will be done autonomously.
Since the contractor is responsible for the autonomously generated output, which is detection reports - not images, we
took a design approach that did not stop with a signal to noise calculation but instead simultaneously considers the
effects of hardware configurations and algorithm choices. Key requirements for GLM are the probability of detection
(PD) and probability of false alarm (PFA). Our approach allows us to provide a system with the best PD and PFA
performance and the best value. We have accomplished this by developing an analytical model that can find "knees-in-the
curve" in our hardware configuration selections and an algorithm prototype that provides realistic end-to-end
performance. These tools allow us to develop an optimal system since we have a good handle on realistic performance
prior to launch.
Our tools rely on descriptions of lightning phenomena embodied in probability densities we developed for the amplitude,
temporal and spatial distribution of lightning optical pulses. The "analytic model" uses tabulated integration formulae
and conventional numerical integration to implement an analytical solution for the PD estimate. The average PD is
quickly computed, making the analytic model the choice for rapid evaluation of sensor design parameter effects.
The "algorithm prototype" utilizes simulation, consisting of data cubes of time elapsed imagery containing lightning
pulses and structured backgrounds, and prototyped detection and false alarm mitigation algorithms to estimate PD and
PFA. This approach provides realistic performance by accounting for scene spatial structure and apparent motion.
We discuss the design and function of these tools and show results indicating the variation of PD and PFA performance
with changes in sensor and algorithm parameters and how we use these tools to improve our instrument design
capabilities.
|