Kim Whitehall
(NASA, Jet Propulsion Laboratory)
GTG (Grab ’em, Tag ’em, Graph ‘em) and MapReduce in Atmospheric Sciences
What | GR Homepage Meteo Colloquium UG |
---|---|
When |
Aug 31, 2016 03:30 PM
Aug 31, 2016 04:30 PM
Aug 31, 2016 from 03:30 pm to 04:30 pm |
Where | 112 Walker Building |
Contact Name | Greg Jenkins |
Contact email | gsj1@psu.edu |
Contact Phone | (814) 865-0479 |
Add event to calendar |
![]() ![]() |
This talk provides an overview of the Grab ‘em, Tag ‘em, Graph ‘em (GTG) method - a fully automated graph-theory based algorithm for identifying mesoscale convective systems (MCSs) and the subcategory large-scale mesoscale convective complexes and mesoscale convective complexes (MCCs) in a timeseries of highly resolved infrared satellite images (Whitehall et al., 2014). Large-scaled MCSs and MCCs are meso-alpha scale features ~ 105 km2 that last between 9 to 24hrs, produce high precipitation events that significantly contribute to monthly precipitation variability, and possess an undefined capacity to alter regional mass, moisture and heat fluxes, and thus global energy and water distributions. As such, on weather and season-timescales, MCCs and large-scaled MCSs present considerable socio-economic impacts to local and international disaster response agencies (Maddox, 1980; Laing and Fritsch, 1993). This talk will explain how the algorithm works within a serial environment and distributed, parallel environment. Special attention is paid to the latter implementation which is available in SciSpark (https://scispark.jpl.nasa.gov/). The SciSpark project (AIST 2014 PI Mattmann 2015) leverages Apache Spark’s (http://spark.apache.org/) in-memory calculations and converged analytics platform to deliver the scientific Resilient Distributed Datasets (sRDDs) - a collection of earth science data elements partitioned across the nodes of a cluster that can be operated on in parallel - created by time subsetting operations on earth science datasets, and/or other operations during analysis. Such computational abilities applied to meteorological applications can significantly reduce computational time associated with data subsetting, ingestion and analysis.