"Deep Convection, Lightning, and Inertio-Gravity Waves: An Analysis of " by Stephen L. Solimine

ORCID

https://orcid.org/0000-0002-5322-6041

Date of Award

Spring 2025

Language

English

Embargo Period

5-9-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

College/School/Department

Department of Atmospheric and Environmental Sciences

Program

Atmospheric Science

First Advisor

Liming Zhou

Committee Members

Paul Roundy, Christopher Thorncroft, David Turner

Keywords

Congo Basin, Deep Convection, Inertio-Gravity Waves (WIG), Mesoscale Convective Systems (MCSs), Lightning-Precipitation Relationships, Machine Learning in Atmospheric Science

Subject Categories

Atmospheric Sciences | Climate | Meteorology

Abstract

The lack of an extensive in situ observation network throughout equatorial Africa makes it difficult to monitor the Congo climate system. Model and reanalysis products have been used to fill informational gaps but often produce disparate results. To improve model representation of deep convection over the Congo, it’s important to first understand the influence of various thermodynamic and dynamic forces that may be underrepresented. Although satellite based remote sensing retrievals and reanalysis have their own set of limitations, they can offer a great deal of insight into the climate of regions otherwise lacking in observational networks.

The first section of this dissertation explored potential methods for improving satellite derived precipitation measurements through the examination of lighting-precipitation relationships in the Congo. Utilizing retrievals from high-resolution Tropical Rainfall Measuring Mission (TRMM) during its 1998 – 2013 lifespan, lightning and precipitation measurements associated with observed echoes were isolated and assigned to one of four categories of intense convective- stratiform echo types. Results show that only 2.7% of observed echoes were classified as intense convective-stratiform, yet they produced 36.6% of observed lightning flashes and 27.4% of estimated rain totals. Significant spatial correlations were also found between total rainfall and intense convective-stratiform rain (coefficient r = 0.56). Linear relationships between lightning and echo rain rates are shown to depend heavily on the convective category. As a result, a simple linear regression cannot be made for all intense convective echoes. However, lightning can be used to retrieve a lower-bound approximation with respect to convective rain rates.

Next, a study was conducted into the origins of intense convective events analyzed in the previous section. The role of quasi-two-day westward propagating inertio-gravity (WIG) waves were proposed as a major mechanism of deep convective development and precipitation variability in the Congo Basin. A space-time Fourier transform was applied to Cloud Archive User Services (CLAUS) derived brightness temperatures (Tb) and filtered for WIG waves. Comparisons between wave phase and TRMM-PR echo retrievals were used to perform a statistical analysis on the preferred “temporal zones” of formation relative to wave passage associated with intense convective-stratiform echo types. A regression analysis between WIG filtered Tb and ERA5 variables was then conducted to investigate the strength and nature of the environmental response. This analysis was conducted for both the Congo and a region of the Indian Ocean to gain a deeper perspective on differences in wave behavior between tropical continental and maritime regions. Lastly, the modulation of precipitation by WIG waves in both regions was assessed via Multi-Source Weighted-Ensemble Precipitation (MSWEP) retrievals. Results showed a strong connection between the most intense convective events and WIG waves, with between 60 – 80%, depending on the stage of convective development, occurring withing a 24-hour period around peak wave intensity. In addition to differences in environmental response to wave passage, significant proportions of estimated rain totals were found to occur in the vicinity of these events with 40% falling in a vast section of the Congo.

The final section utilized a supervised machine learning model - eXtreme Gradient Boosting (XGBoost) - to explore differences in key atmospheric indicators of WIG wave intensity between the Congo and Indian Ocean. A suite of ERA5 atmospheric variables were used to train the model against variations in WIG filtered Tb, after which SHapley Additive exPlanations (SHAP) tools were used to rank feature importance based on their global attribution scores. The model was run and averaged over specific time windows prior to the timing of peak WIG intensity to assess which atmospheric variables are most strongly influenced by WIG wave formation. By comparing results between domain regions, key indicators of wave build-up and subsequent deep convective activity were identified. The model showed impressive ability in distinguishing between land and oceanic WIG wave-related atmospheric processes by learning the unique thermodynamic and dynamic signatures of their impending formation. Despite a strong dynamical influence of WIG waves regardless of region, the model ranked thermodynamic and convective instability parameters, such as temperature and CIN, as the most important features determining wave strength in the Congo, while dynamic fields such as divergence and large-scale circulations dominated over the Indian Ocean.

License

This work is licensed under the University at Albany Standard Author Agreement.

Download

Share

COinS