Below are brief descriptions of research areas in which I am active or that I would like to explore in the near future — there’s no specific order, but active projects are listed first. The main tools and datasets I use to pursue each of these interests are given before the description (notice any common threads?). After the description, any current funding and relevant publications are noted — most of my papers address multiple topics and are listed accordingly.
Please get in touch if you’re interested in pursuing any of these as a collaborator!
Scaling properties of clouds and precipitation
Tools used: scaling analysis, multifractal analysis, aircraft in situ observations, large eddy simulations
The collisional growth processes that produce precipitation are nonlinear which means that, in a sufficiently large volume, neglecting small-scale variability and using only grid-mean quantities to calculate process rates results in systematic underestimation. Many modern GCM microphysics parameterizations account for this variability by integrating over assumed PDFs (e.g. gamma or lognormal) of input variables such as cloud liquid water mixing ratio, drop concentration, etc. The width of the PDFs varies with spatial scale and convective regime, though, so specifying constant PDF width can lead to biased results, especially if a variable mesh model grid is used.
To better understand these issues from an observational perspective, we analyzed the scaling properties of cloud and rain liquid water content (LWC) from high-frequency (500 Hz) aircraft measurements and quantified cloud-rain covariability as a function of length scale (Witte et al., 2019). The analysis showed that both cloud and rain LWC exhibit scaling behavior (i.e. a power law relationship of power spectral density with length scale) from tens of centimeters to tens of kilometers, which implies that variance, covariance and therefore correlation at a particular spatial scale can be derived analytically. But there are other factors than length scale that impact covariability: cloud drop number concentration, vertical location of the observing platform in cloud, and mesoscale cloud organization (open vs. closed cells) also modulate cloud-rain correlation.
While variances and correlation are sufficient to describe variability in the PDF-based approach, a “smooth” description of precipitation is not realistic. In nature, cloud and rain water fields are both rough (statistically nonstationary) and intermittent (sparse with a tendency to singularity), and these properties are essential for producing extreme precipitation. Multifractal analysis quantifies roughness and intermittency and I applied this technique to LES-generated cloud/rain fields to understand why simulations with bulk microphysics agree so poorly with observations. Our results show that bulk microphysics overconstrains precipitation development and results in rain that is too “cloud”-like; that is, overly smooth and insufficiently intermittent. Allowing more degrees of freedom produces more realistic rain fields, but at far greater computational expense. An alternative solution is to stochastically perturb bulk process rates, which we will explore next. A longer-term goal is to formulate a multifractal statistical simulator for clouds and rain using either wavelets or a hybrid additive-multiplicative cascade model.
Publications: Witte et al. (2022); Witte et al. (2019)
–
Coupling dynamics and microphysics in a unified convection/turbulence scheme
Tools: single column modeling, global atmospheric models
A major trend in the last two decades has been the development of a unified representation of moist convection in global models, typically using an assumed probability density function (PDF) based approach (e.g. CLUBB). A major limitation is that this approach is not conducive to simulating deep convection because there is no mechanism for countergradient transport. The most common family of analytical PDFs used in current parameterizations, Gaussian mixtures, can produce sufficient skewness to represent shallow cumulus but it is not suitable for more vigorous air motion. Another limitation is that convective and cloud processes are considered separately, with different PDFs used in the convective and stratiform microphysics schemes and no consideration of sub-grid variability for deep convection – a glaring inconsistency present even in the newest versions of NCAR’s CAM and DOE’s E3SM.
The eddy diffusivity-mass flux (EDMF) framework uses an eddy diffusivity (ED) approach for the non-convective portion of a model column while the convective fraction is described by a number of steady-state mass flux (MF) plumes. The MF plumes can be thought of as a representative sample of the convective “tail” of the multivariate PDF of vertical velocity, temperature and total water mixing ratio. Coupling cloud macro- and microphysics to dynamics in this framework is a matter of 1) using consistent statistical assumptions for the dynamical, thermodynamical and microphysical calculations in both ED and MF components and 2) developing a framework for precipitation physics in the mass flux plumes. At present, I am working on implementing EDMF into three different climate models (DOE E3SM, NCAR CESM, and GFDL AM) and examining how our scheme interacts with other model physics modules to optimize the EDMF implementation. In addition, I am adapting the ED (stratiform) microphysics to respond to variable grid spacing (i.e. scale awareness) based on previous observational work.
Publications: Suselj et al. (2022); Smalley et al. (2022); Witte et al. (2022); Chinita et al. (2023); Glenn et al. (in prep)
Funding: NSF/NOAA Climate Process Team grant, 2019-2024; DOE Earth System Model Development grant, 2023-2026
–
Impacts of shallow convective rain on boundary layer evolution
Tools used: ground-based remote sensing, aircraft in situ observations, radiosondes, single column modeling
Mass flux parameterizations typically describe only updrafts, but convective downdrafts and associated precipitation also contribute significantly to the transport of energy and moisture in the cloud-topped planetary boundary layer. In particular, I am interested in how evaporation of precipitation in downdrafts 1) moistens the boundary layer and 2) drives cold pool formation, which is important for the development of cloud cellular structure and the transition from stratocumulus to cumulus. Arguably, cold pools are even more important for the transition from shallow to deep convection. To gain a statistically robust understanding of these phenomena, my group has compiled a long-term database of precipitating shallow convective clouds at the DOE Atmospheric Radiation Measurement program Eastern North Atlantic facility. This database has been used as a standard for comparison with single column model simulations run over a suite of shallow convective cases to evaluate which aspects of convective downdrafts are most important to incorporate into EDMF.
Publications: Jeong et al. (2022); Chandrakar et al. (2022); Jeong et al. (in review); Glenn et al. (in prep)
Funding: DOE Atmospheric System Research grants, 2019-2023; 2022-2025
–
intercomparison of in situ cloud observations with remote sensing retrievals
Tools: aircraft in situ observations; ground-based remote sensing; satellite-based remote sensing
Confidence in the performance of numerical models is gained by validating against observations, and confidence in observations is gained by obtaining comparable results from instrumentation/platforms that operate on different principles. Identifying the capabilities and limitations of different observing techniques is also necessary to make informed decisions about the appropriate dataset to use for a given scientific question. Since many of the questions I am interested in pertain to small spatial scales (especially related to variability), aircraft in situ probes are typically the starting point for my investigations. I have used aircraft observations in nearly all projects I have been involved with to date, although I am increasingly using remote sensing platforms (both ground- and satellite-based) to gain a longer-term, more global perspective.
A key question regarding in situ instrumentation is how well different probes are able to measure the width of the drop size distribution. This is a topic that surfaces every few years in the literature but has never been comprehensively addressed by the instrumentation community. I would like to perform an intercomparison of state of the art cloud probes’ spectral width measurements, which can then be used to place previous results in a broader context. One of the main findings of Witte et al. (2018) was that users of in situ observations beyond the instrumentation development community are often overly reductive in error analyses (i.e. assuming uncertainty comes only from sampling error) and neglect to account for basic differences among instruments such as different operating principles, fundamental variables measured, etc., thus organizing and presenting this information in an easily digestible format would be of great value to the broader research community.
I am also interested in better understanding how well larger footprint remote sensors (in particular, satellites) are able to measure cloud variability and in what environments the remote sensing algorithms are prone to failures that can be addressed from in situ measurements. In my opinion, the most pressing need is to validate cloud drop number concentration retrievals, which are highly sensitive to drop size. Finally, experiments such as the recent DOE ACE-ENA campaign that had nearly collocated aircraft in situ and ground-based radar cloud observations could lead to insights on how retrieved Doppler spectra correspond with drop size distribution shape, an advance that would open huge datasets to new interpretation from a microphysical perspective.
Publications: Lebsock and Witte (in press); Witte et al., (2018); Rémillard et al. (2017)
–
Precipitation formation in warm clouds
Tools: aircraft in situ observations; ground-based radar observations; large eddy simulations
The question of how rain forms in the absence of ice has remained an open area of research despite over a century of work on the problem. Small drops grow by condensation but the rate of vapor diffusion is limited by surface area; large drops grow by colliding and coalescing with smaller drops by differential gravitational settling and turbulent air motions but small drops settle slowly and are inefficient colliders. I have studied two of the proposed mechanisms to bridge this process bottleneck: turbulent enhancement of collision rates and preferential condensation on giant cloud condensation nuclei (GCCN).
In Witte et al. (2017), we evaluated theoretical collision rates using aircraft observed DSDs and showed that turbulent acceleration of collision rates cannot explain steady state DSD shape, suggesting that turbulent collision-coalescence alone cannot form precipitation in the low marine clouds sampled. We developed two case studies from the same set of observations for an LES study (Witte et al., 2019) in which we showed that the bin microphysics framework, often considered the “gold standard” to which simpler schemes are compared (for now! see next topic for more), suffers from systemic numerical errors due to separation of advection and calculation of supersaturation forcing that limit our ability to answer process-level questions about microphysical evolution. These errors were quantified in Morrison et al. (2018) and a method to correct the issue has been identified.
A major implication of Witte et al. (2017) is that condensation likely plays a greater role than previously thought in opening the process bottleneck. Specifically in marine settings, preferential activation of GCCN with radius r>2 µm can grow by condensation alone to r>25 µm (Jensen and Nugent, 2017), a commonly used threshold size to differentiate between cloud and precipitation particles. Using aircraft observations of sea salt GCCN and cloud and rain DSDs in shallow marine clouds, Jensen and Witte (in prep) show that the observed rate of rain drop formation (i.e. autoconversion) is underestimated by up to an order of magnitude if condensational growth on GCCN is omitted, a finding that has significant ramifications for microphysics schemes that completely neglect GCCN (i.e. the vast majority of schemes).
Publications: Zhang et al. (2021); Witte et al. (2019); Morrison et al. (2018); Witte et al. (2017); Rémillard et al. (2017)
–
development and evaluation of microphysical schemes
Tools: in situ and remote sensing observations for evaluation; large eddy simulations; simple models (e.g. single column, parcel, box)
Our ability to answer scientific questions depends on having research tools that we understand and trust. Just as different methods of observing clouds must be evaluated against each other to understand their relative strengths and weaknesses, the same applies to different methods of simulating cloud microphysics. I only discuss liquid phase microphysics here for brevity, but most of the concepts I mention are also applicable to mixed- and ice phase microphysics as well. As such, I will refer to “drops” instead of the more general term “hydrometeors.”
There are three commonly used approaches to modeling microphysics: bulk, bin and Lagrangian (sometimes referred to as the “superdroplet” method). Each approach has advantages and disadvantages, and which approach one should use depends on the specific application/scientific question. Bulk schemes are the most commonly used because they have the lowest computational cost, but the tradeoff is that they allow few degrees of freedom in representing microphysical processes. Bin and Lagrangian schemes explicitly model drop size distribution (DSD) evolution (i.e. the shape freely evolves), but this added complexity comes at a significant computational cost. There are also differences between the two size-resolving approaches: the bin method subdivides the DSD into discrete bins and is an Eulerian approach (i.e. each model gridbox has a full DSD within it) while the Lagrangian approach tracks the evolution of distinct particles (“superdroplets”) in space and time. At present, Lagrangian methods typically suffer from poor sampling statistics due to the high cost of using a large number of particles, but advantages in numerical implementation point to this approach being the most promising in the long term.
My previous work in evaluation of microphysical schemes has been with the bin approach (see publications below). Although this work has primarily been diagnostic up to this point, the next step is to begin testing approaches to fix problems we have identified. If you have a strong programming background and are willing to learn the relevant physics, this type of work may be for you. As computing resources continue to evolve, I would also like to revisit questions about warm rain with Lagrangian microphysics since it is not subject to the same numerical errors inherent to Eulerian methods.
Publications: Chandrakar et al. (2022); Witte et al., (2022); Witte et al. (2019); Morrison et al., (2018); Rémillard et al., (2017)
–
phase partitioning in mixed phase clouds
Tools: aircraft observations; ground- and space-based remote sensing observations; limited-area models
While participating in the SOCRATES field campaign over the Southern Ocean, I was struck by the incredible heterogeneity in cloud phase we would occasionally observe from the aircraft: pure liquid cloud, then through intermixed drops and needles, then a completely glaciated patch, and back to liquid — all within the span of a minute. Mixed-phase microphysical processes are quite sensitive to assumptions about the overlap of ice and liquid hydrometeors, especially in coarser-gridded models. There are typically too few low clouds over the Southern Ocean in global atmospheric models, and one cause of this can be misdiagnosis of ice fraction: especially when it is overpredicted, low clouds rain out and dissipate too quickly, leading to too much surface absorption of solar radiation. Furthermore, most spaceborne sensors have insufficient resolution to resolve the small-scale phase heterogeneity seen from the aircraft and retrievals are sensitive to phase partitioning. Another complicating factor is that low clouds are composed of multiple thin layers (<200 m) that are not visible to active sensors such as CloudSat. Characterizing this variability from observations, testing our ability to reproduce it in models, and understanding how to deal with this heterogeneity in satellite retrievals are all interesting questions for future research.
