Leveraging polarimetric radar observations to learn about rain microphysics: An exploration of parameter estimation, uncertainty quantification, and observational information content with BOSS
Open Access
- Author:
- Reimel, Karly
- Graduate Program:
- Meteorology and Atmospheric Science
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 16, 2021
- Committee Members:
- Steven Greybush, Major Field Member
Julio Urbina, Outside Unit & Field Member
Matthew Kumjian, Chair & Dissertation Advisor
Anthony Didlake, Major Field Member
Marcus van Lier-Walqui, Special Member
David Stensrud, Program Head/Chair - Keywords:
- meteorology
rain
radar
polarimetric radar
microphysics
Bayesian inference
Markov chain Monte Carlo
BOSS
drop size distribution - Abstract:
- The accuracy of numerical weather and climate models depends on our ability to represent complicated atmospheric processes in a simplified way. Processes that act on hydrometeors are particularly important for computing rain rates and accurately evolving cloud systems within models. Unfortunately, these microphysical processes act on scales that are much smaller than model resolution, as computational costs do not allow for each hydrometeor to be explicitly tracked. Additionally, we lack fundamental understanding of microphysical processes because they are difficult to observe or reproduce in a laboratory setting. Microphysics schemes parameterize rates for specific processes, such as drop evaporation, but their inherent assumptions lead to model uncertainty that is often difficult to understand and quantify. To improve the representation of microphysical processes within numerical weather and climate models, there is a need to identify and reduce scheme uncertainty and improve our understanding of each process using observations. The Bayesian Observationally Constrained Statistical-Physical Scheme (BOSS) is a recently-developed bulk rain microphysics scheme with no predefined drop size distribution (DSD) shape and few assumptions made about the process rate formulations. Process rates are written as generalized power functions of a flexible choice in prognostic DSD moments (related to bulk quantities such as mass concentration). The corresponding parameters are constrained directly with observation using Markov chain Monte Carlo (MCMC), allowing BOSS to learn microphysical information from observations while simultaneously quantifying parametric uncertainty. The process rate formulations in BOSS can be made systematically more complex, which allows us also to track down sources of structural uncertainty. Thus the BOSS/MCMC framework provides an avenue to further our understanding of complicated microphysical processes while improving how they are represented within numerical weather models. Polarimetric radar observations are an ideal choice for constraining the BOSS parameters because vertical gradients in the polarimetric radar variables (ZH, ZDR, and KDP) exhibit unique signatures that indicate which rain process (evaporation, coalescence, breakup) is dominant within a rainshaft. In this study, we constrain the BOSS parameters with polarimetric radar observations by pairing the BOSS/MCMC framework with a polarimetric forward operator. We test the framework using a detailed bin microphysics scheme as “truth” to generate the constraining observations synthetically. The constraining observations include profiles of ZH, ZDR, and KDP, and fluxes of prognostic DSD moments at the surface. An error analysis shows that BOSS produces process rate profiles similar to those of a bin scheme. We explore the parametric and structural uncertainty of BOSS through testing how the number of constraining observations and choice in prognostic moments affect the posterior probability distributions of the BOSS parameters. Analyzing correlations between the estimated parameters is one way to indicate sources of uncertainty within the model. We then explore the physical meaning behind these parameters and provide evidence that our framework teaches BOSS rain physics relationships that are consistent with those found in nature. These results suggest that the posterior parameter PDFs estimated using synthetic observations can be used to retrieve microphysical information from real observations and bolsters our confidence that BOSS can eventually learn about rain physics from real observations. To enable BOSS to learn about rain physics from real observations, we develop a DSD moment retrieval that estimates the BOSS prognostic moments and quantifies the corresponding uncertainty when a value of ZH, ZDR, and KDP is used as input. These retrieved values are used to initialize BOSS rainshafts at model top. Tests indicate that the retrieval performs best when the moments being retrieved are closely related to the polarimetric radar observations. To ensure BOSS learns physics from real radar signatures, rather than noise within the radar observed fields, we develop a method to quantify the uncertainty of ZH, ZDR, and PhiDP in stratiform rain observations recorded by WSR-88D radars. Estimating KDP observation errors is difficult because KDP is estimated from noisy PhiDP fields. Numerous algorithms exist that estimate KDP from PhiDP, each with differing methodologies. Using a series of known-truth tests, we demonstrate that the estimated KDP field is dependent on the algorithm chosen and the parameters defined within each algorithm. We use this analysis to inform our choice in KDP estimation algorithm when pairing BOSS with real observations. The work in this study culminates in a proof of concept in which quantitative process rate information is estimated from observations of Hurricane Matthew over a 12-hour period. We use the DSD moment retrieval to estimate the prognostic moment values at model top and evolve observed rainshafts using BOSS microphysics, using the posterior parameter PDF produced with synthetic observations serving as constraint. The BOSS process rates indicate that coalescence is the dominant process, consistent with our expectations after analyzing the vertical gradients in the observed polarimetric radar fields. BOSS simulated polarimetric radar profiles are reasonably consistent with observed profiles, further bolstering our confidence in the framework and demonstrating that BOSS can be used as a tool to gain quantitative process level information from observations. Furthermore, this proof of concept shows that the framework can be used in the future to constrain the BOSS parameters with real polarimetric radar observations, such that BOSS learns physics directly from real rain observations.