2025-11-07

Active Measurement

Motivation

• Computational limitations instead of resolution
• Distributed network of observational sensors

ML prediction to scientific insights

Bird Migration

• Weather radars to track birds. They travel at night, we cannot see them but radars are sensitive enough to see them
• Dark Ecology
  • Can we extract something from the 25 year archival data?
• Extract birds from the radar data. They were not measuring birds intentionally but they can be extracted form the data
• Plot biomass density vs direction, speed, height

Challenges

• Big data
  • Huge number of radars
  • Resolution heterogeneity
  • Remove weather contamination. It is not a problem other way around (biomass so small compared to weather that they can ignore them but it doesn’t work that way for biomass)
    • MistNet. CNN for weather radar
    • Produce mask from radar measurement data
    • Training data?
      • Dual polarization radar helped with this to label
      • Rain circular birds not
      • After 2013 the sensors does the classification do not need Mistnet
    • Spring they move north then they come back later in the year to the south

Monte Carlo Sampling

• Improve variance by importance sampling