1 of 15

��Uncovering PFAS Contamination through Active Geospatial Search

​

​

Jowaria Khan

PhD. Student

University of Michigan

Elizabeth Bondi-Kelly

Assistant Professor, Department of CSE

University of Michigan

Alexa Friedman

Environmental Research Scientist

Environmental Working Group

Sydney Evans

Environmental Research Scientist

Environmental Working Group

David Andrews

Environmental Research Scientist

Environmental Working Group

Kaley Beins

Environmental Research Scientist

Environmental Working Group

Katherine Manz

Assistant Professor, Department of Public Health

University of Michigan

Yevgeniy Vorobeychik

Professor, Department of CSE

University in St. Louis, St. Louis

Yiting Xiao

Postdoctoral Student, Department of Public Health

University of Michigan

Anindya Sarkar

PhD. Student, Department of CSE

University in St. Louis, St. Louis

2 of 15

Motivation

1

3 of 15

Motivation

3

2

4 of 15

Motivation

3

5 of 15

Motivation

4

6 of 15

Motivation: Traditional Approaches to Measure and Predict PFAS

  • Lab-based analysis which is costly, time-intensive, and difficult to scale for large spatial datasets.

​

6

Data gaps

$300

7 of 15

Background: FOCUS

33

FOCUS: A geospatial DL framework for PFAS contamination prediction using satellite data products.

8 of 15

FOCUS: Data

33

9 of 15

FOCUS: Training Workflow

40

10 of 15

FOCUS: Main Results

53

11 of 15

FOCUS: Main Results

53

2025?

12 of 15

Potential Directions

59

  • Active learning: Selects the most informative data points to improve sample efficiency during training in scenarios with limited prior data and costly labeling.

​

​

​

Accuracy

Fscore

Precision

Recall

AL

61%

39%

45%

33%

Prithvi

64%

48%

48%

48%

13 of 15

Potential Directions

59

  • Active learning: Selects the most informative data points to improve sample efficiency during training in scenarios with limited prior data and costly labeling.

​

​

AL typically assumes a stationary distribution of learning tasks, which limits their ability to handle evolving distributions for geospatial search.

​

​

14 of 15

Proposed Framework: Active Online-Meta Learning

59

  • This framework integrates active learning and online meta-learning to adaptively search and discover targets in evolving geospatial environments under tight resource constraints.

15 of 15

Towards Deployment

59

  • Advisory Group: Multidisciplinary advisory group (EWG, HRWC, state agencies) guiding development of community-relevant PFAS tools.

​

  • Real-World Sampling: Sampling from representative points in the Huron River identified by this framework for targeted PFAS monitoring.