Convective Initiation and the Surface in Model Simulations

Work conducted at the University of Washington in support of my doctorate degree with Drs. Greg Hakim and Cliff Mass

Simulating Isolated Convective Initiation

We are conducting simulations using Cloud Model 1 (CM1; George Bryan) of environments where deep convective initiation (CI) is possible in the absence of larger-scale forcing and due to boundary-layer processes alone. With only random, small temperature perturbations added at the initial time and diurnal radiation as a forcing, we have been able to successfully simulate CI in over 20 environments. By varying only the initial random temperature perturbations, we develop ensembles of simulations for each environment, allowing us to generate large samples of individual convective cells throughout their development and critically absent a prescribed forcing mechanism. We are probing these simulations for insights into the CI process and how to best observe the atmosphere prior to CI for improved prediction.

Composited Surface Anomalies Prior to CI

Using these ensemble simulations of convective initiation, we identify storm "objects" and track them through their evolution, considering the time when they first produce precipitation that reaches the surface to be the time of "initiation". We then composite anomalies of surface fields surrounding these storm objects as they develop. While some features---like positive temperature anomalies and convergent winds early in the evolution---were expected, the dominant pre-precipitation feature we note is the effect of cloud shadowing, which produces a cold anomaly on the order of several Kelvin as the convective cloud deepens. This anomaly also alters surface heat and moisture fluxes as the storm develops. Prior to CI, it is the temperature and wind anomalies that appear most dominant and feasibly observable with possible observing networks. The strength and structure of the moisture anomalies vary depending on the environment and the surface pressure anomalies during CI are generally weak. Following initiation, classic cold-pool signatures are observed.

Horizontal and Vertical Correlations

We examine ensemble correlations for surface variables to estimate the potential scale of impact for surface observations during the CI process. Absent any large-scale forcing in these simulations, the horizontal correlation length scales are generally very short--on the order of 4-5 km for most variables, with surface pressure having the longest correlation length scale. We also examine correlations in the vertical and find that surface observations, particularly temperature, could provide information about the structure of moisture within the deepening convective boundary layer and the developing convective cloud throughout the simulations. These correlations are likely larger than what would be realizable in more realistic simulations with a background environment that is not as well constrained. However, these correlations illustrate promise for surface observations to inform beyond just the near surface layer in convectively-active environments.

Assimilating Surface Observations for CI Prediction

Skill in Perfect-Model Observing System Simulation Experiments

A series of Observing System Simulation Experiments (OSSEs) were conduced using the CM1 model interfaced with the Data Assimilation Research Testbed (DART; NCAR) ensemble Kalman adjustment filter for data assimilation. Despite an "identical twin" setup for these OSSEs, we still found that a surface observing network grid with spacings of at least 4 km (and particularly at 1 km) was required to have any skill at predicting CI over a random forecast with no observations assimilated. The figure at left shows how skill at predicting storm development locations changes with increasing surface observation density. Additionally, no skill improvement was possible for forecasts initiated more than 2 hours prior to convective initiation, or before the cumulus field began to form.

Ongoing Challenges in Convective-Scale Data Assimilation

The role of ensembles in data assimilation becomes complicated at smaller spatial scales, as sampling error and strong non-linearities challenge traditional ensemble DA methods. Despite this, our OSSE experiments showed that even with this sampling error, surface observations could make correct adjustments to the ensemble background cloud field and contribute positively to convective forecast skill. The figure at right shows how surface increments (middle panel) correctly augment or reduce the ensemble mean background cloud field (top panel, shading) to produce an analysis (bottom panel) that is in better agreement with the true cloud state (black contours, bottom panel). Ongoing work is examining how to further maximize the utilization of the observation information by the ensemble during DA. I have ongoing projects considering how four-dimensional (4D) and object-based data assimilation methods may contribute to improved convective-scale analyses.

Publications and Presentations

Currently Published:
Madaus, L. and G. Hakim (2016): Observable Surface Anomalies Preceding Simulated Isolated Convective Initiation. Mon. Wea. Rev.. 144, 2265-2284. doi:

In Revision:
Madaus, L. and G. Hakim (2017): Constraining Ensemble Forecasts of Discrete Convective Initiation with Surface Observations. Mon. Wea. Rev.. In Revision. Manuscript Available on Request.

Here is a presentation I gave on the preliminary findings of this work at the 27th Severe Local Storms conference in Madison, Wisconsin (Nov 2014):