Alex Haberlie is an Assistant Professor in the Department of Geography and Anthropology at Louisiana State University. He is a climate scientist who uses geographic information systems, machine learning, and weather models to explore topics related to deep, moist convection (i.e., thunderstorms) and their affiliated hazards.
PhD in Geography, 2018
Northern Illinois University
MSc in Geography, 2014
Northern Illinois University
BSc in Computer Science, 2010
University of Wisconsin - Platteville
Understanding how snowstorms may change in the future is critical for estimating impacts on water resources and the Earth and socioeconomic systems that depend on them. Here we use snowstorms as a marker to assess the mesoscale fingerprint of climate change, providing a description of potential changes in winter weather event occurrence, character and variability in central and eastern North America under a high anthropogenic emissions pathway. Snowstorms are segmented and tracked using high-resolution, snow water equivalent output from dynamically downscaled simulations which, unlike global climate models, can resolve important mesoscale features such as banded snow. Significant decreases are found in the frequency and size of snowstorms in a pseudo-global warming simulation, including those events that produce the most extreme snowfall accumulations. Early and late boreal winter months show particularly robust proportional decreases in snowstorms and snow water equivalent accumulations.
This research uses image classification and machine learning methods on radar reflectivity mosaics to segment, classify, and track quasi-linear convective systems (QLCSs) in the United States for a 22-yr period. An algorithm is trained and validated using radar-derived spatial and intensity information from thousands of manually labeled QLCS and non-QLCS event slices. The algorithm is then used to automate the identification and tracking of over 3000 QLCSs with high accuracy, affording the first, systematic, long-term climatology of QLCSs. Convective regions determined by the procedure to be QLCSs are used as foci for spatiotemporal filtering of observed severe thunderstorm reports; this permits an estimation of the number of severe storm hazards due to this morphology. Results reveal that nearly 32% of MCSs are classified as QLCSs. On average, 139 QLCSs occur annually, with most of these events clustered from April through August in the eastern Great Plains and central/lower Mississippi and Ohio River Valleys. QLCSs are responsible for a spatiotemporally variable proportion of severe hazard reports, with a maximum in QLCS-report attribution (30%–42%) in the western Ohio and central Mississippi River Valleys. Over 21% of tornadoes, 28% of severe winds, and 10% of severe hail reports are due to QLCSs across the central and eastern United States. The proportion of QLCS-affiliated tornado and severe wind reports maximize during the overnight and cool season, with more than 50% of tornadoes and wind reports in some locations due to QLCSs. This research illustrates the utility of automated storm-mode classification systems in generating extensive, systematic climatologies of phenomena, reducing the need for time-consuming and spatiotemporal-limiting methods where investigators manually assign morphological classifications.
This research applies an automated mesoscale convective system (MCS) segmentation, classification, and tracking approach to composite radar reflectivity mosaic images that cover the contiguous United States (CONUS) and span a relatively long study period of 22 years (1996–2017). These data afford a novel assessment of the seasonal and interannual variability of MCSs. Additionally, hourly precipitation data from 16 of those years (2002–17) are used to systematically examine rainfall associated with radar-derived MCS events. The attributes and occurrence of MCSs that pass over portions of the CONUS east of the Continental Divide (ECONUS), as well as five author-defined subregions—North Plains, High Plains, Corn Belt, Northeast, and Mid-South—are also examined. The results illustrate two preferred regions for MCS activity in the ECONUS: 1) the Mid-South and Gulf Coast and 2) the Central Plains and Midwest. MCS occurrence and MCS rainfall display a marked seasonal cycle, with most of the regions experiencing these events primarily during the warm season (May–August). Additionally, MCS rainfall was responsible for over 50% of annual and seasonal rainfall for many locations in the ECONUS. Of particular importance, the majority of warm-season rainfall for regions with high agricultural land use (Corn Belt) and important aquifer recharge properties (High Plains) is attributable to MCSs. These results reaffirm that MCSs are a significant aspect of the ECONUS hydroclimate.
This research evaluates the ability of image processing and select machine learning algorithms to identify mesoscale convective systems (MCSs) in radar reflectivity images for the conterminous United States. The process used in this study is comprised of two parts: segmentation and classification. Segmentation is performed by identifying contiguous or semi-contiguous regions of deep, moist convection that are organized on a horizontal scale of at least 100 km. The second part, classification, is performed by first compiling a database of thousands of precipitation clusters, and then subjectively assigning each sample one of the following labels: 1) midlatitude MCS; 2) unorganized convective cluster; 3) tropical system; 4) synoptic system; and 5) ground clutter and/or noise. The attributes of each sample, along with their assigned label, are used to train three machine learning algorithms: Random Forest, Gradient Boosting, and XGBoost. Results using a testing dataset suggest that the algorithms can distinguish between MCS and non-MCS samples with high specificity and sensitivity. Further, the trained algorithm predictions are well-calibrated, allowing reliable probabilistic classification. The utility of this two-step procedure is illustrated by generating spatial frequency maps of automatically identified precipitation clusters that are stratified using various reflectivity and probabilistic prediction thresholds. These results suggest that machine learning can add value by limiting the amount of false-positive (non-MCS) samples that are not removed by segmentation alone.