Meet a RC researcher- Mary Jo Brodzik

At the National Snow and Ice Data Center at the University of Colorado, Boulder, Mary Jo Brodzik works with data from passive microwave sensors on Earth-observing satellites. These satellites collect data that can be used to study oceanography, wind speed on the surface of the ocean, soil moisture, seasonal snow cover, and polar sea ice. Until recently, climate modelers thought that ice sheets were static; only in the past 20 years have advances in glaciology shown that the ice sheets change and interact with the oceans and atmosphere. The information that the microwave data provides on liquid and crystal forms of water is now used to track climate change. Brodzik explains, “In September, when you hear on the news that the sea-ice minimum in the Arctic has hit a new low, that news article is coming out of this data center.”

Since 1978, the microwave data has been processed at 25-kilometer resolution. Brodzik says, “Up to now, there’s only been one number for a pixel that covers an area from Boulder to Denver and from the Front Range to I-25.” Her current project focuses on reprocessing the data, taking advantage of overlapping footprints between the original samples to improve the spatial resolution. “The supercomputer is the only way we can get that done,” Brodzik explains. “It’s highly, highly computationally expensive to run this image reconstruction technique that we’re doing, and we’re running a lot of data through it.”

Over the course of her career, Brodzik has seen software evolve dramatically. After studying mathematics as an undergraduate student, she worked as a defense contractor in the ‘80s and early ‘90s, developing software for Earth-observing defense satellites. She says, “That was when I got excited about looking at data from satellites, tracking them in orbit, and learning about the astrodynamics involved. That really got me going.” When Brodzik began working with observational satellites at the National Snow and Ice Data Center in 1993, her computer had four gigabytes of memory and took up her entire desk. At the time, she had to take certain shortcuts when processing the data due to the limits on computational power. She says, “When we got that data ready to go, we had to commandeer any Sun workstations that we could from the people down the hall and run their machines at night. Even with four workstations running around the clock, it took us years just to get caught up. This is what the supercomputer allows us to do overnight.”

Brodzik began using the Janus supercomputer in 2014. Until recently, most of her work on Janus involved testing, running daily regressions, and porting data into the supercomputer. Now, she has reached the stage at which she can examine a prototype year from the 38-year data set. Brodzik and her colleague, Molly Hardman, are using two processing techniques on each of two input data sets for the prototype year, an endeavor which will require from 200,000 to 300,000 hours on the supercomputer. Going forward, they will use a single technique and a single processing method, significantly cutting down the required number of computational hours. Once processed, the data can be used to inform computer simulations. Hardman says, “What data like ours can do is provide for models and simulations in the historical context. Scientists can compare their models to our data and see how accurate the models are.” Brodzik adds, “For climate change, we have to have simulations. We can’t possibly understand the future otherwise.”

 

To utilize the supercomputer, Brodzik and Hardman rely on the Research Computing staff for expertise. “They’ve been awesome,” Brodzik says. “There’s no way that we have the time to learn how to run a machine like this. That’s why Tim [Brown] and Pete [Ruprecht] are so critical to us.” Hardman adds, “They’re really responsive, and they learn just enough of the details of our project to get the job done, and that is critical. Not everyone is capable of doing that. And a lot of technical people are very arrogant about what they know, and there’s none of that from Tim and Pete.” High-performance computing requires a skill set so advanced that modern researchers cannot themselves become experts in it; instead, they rely on collaboration with experts such as those at Research Computing. “Big science and big data needs a lot of different people who are willing to talk to each other without teaching each other their expertise,” Brodzik says. “The days of scientists being geeks in dark closets and working by themselves are over.