Julia Poska | March 22, 2019
Hydrologists can never completely predict when flooding will strike. Conservationists can never be sure how chemical spills will impact fish populations, nor can anyone really foretell how extreme the effects of climate change will be. That’s why environmental researchers need statisticians like Kate Cowles.
“One of the hallmarks of statistical work is assessing realistically how much uncertainty remains,” said the University of Iowa professor of statistics and biostatistics.
Cowles’ was first introduced to environmental statistics when another environmental statistician, her early mentor Dale Zimmerman, called on her expertise in Bayesian statistics. Together they calibrated four methods of measuring water held in snow across the western U.S..
“Indeed I learned an enormous amount from Dale and really got hooked on the environmental and spatial,” she said. and “I’ve pretty much been working in that area ever since.”
Cowles began her career as a piano teacher…how did she get here today? Listen to her describe her fascinating journey.
Notably, Cowles was director of GEEMaP (Geoinformatics for Environmental and Energy Modeling and Prediction), a value-added graduate program funded by the National Science Foundation. Before it ended last summer, GEEMaP brought together faculty and graduate students in fields like statistics, civil and environmental engineering, mechanical and industrial engineering, computer science, geoinformatics and geography.
The problems they discussed and solved exposed students to real-world problems and gave them a strong grounding in statistics and geographic information systems (GIS), Cowles said. Every project promoted interdisciplinary collaboration.
Cowles said a class she teaches on Bayesian statistics, her specialty, also resonates well with engineering students. The Bayesian approach allows users to quantify what they do and do not know and update their understanding as more information comes in. Cowles believes it parallels the way engineers think and lends itself well to engineering problems. She is always excited to advise engineering students and further promote collaboration with statisticians.
“I think that it is crucially important for those two types of data analysts to work together and communicate with each other,” Cowles said.
Because environmental datasets are often measured over both space and time, researchers in fields like agriculture and meteorology must account for spatial correlation. As the first law of geography states, “everything is related to everything else, but near things are more related than distant things.”
Hear about a possible application for Cowles and her student’s spatial correlation software.
Calculating that relationship requires complex statistics, but failing to account it properly can lead to faulty conclusions.
“Statistical methods that help us draw the right conclusions for complex data like that are becoming more and more important,” Cowles said.
One of Cowles’ graduate students is developing software that “mops up that spatial correlation,” making things easier for non-statisticians making predictions based on spatial data.
Processing such enormous datasets is slow work, however. In many cases, engineering methods like machine learning are faster than statistical methods, which Cowles said creates tension between disciplines.
Listen to Cowles explain how she hopes to speed up complex spatial processing.
Another large part of her work focuses on activating underutilized graphical processing units inside computers to do many simple computations simultaneously, which can speed up the processing of such data.
“Statisticians need to catch up, because engineers and environmental scientists cannot wait for a long time for results of their analyses!” she said.
***This post is part of “CGRER Looks Forward,” a blog series running every other Friday. We aim to introduce readers to some of our members working across a wide breadth of disciplines, to share what the planet’s future looks like from their perspective and the implications of environmental research in their fields. ***