Enter Note Done
Go to previous page in this tab
Session
  • Presentation | B44B: Bridging Remote Sensing, Machine Learning, and Ecological Modeling to Address Forest Health Challenges II Oral
  • Oral
  • Bookmark Icon
  • B44B-01: Autocorrelation, not ecological context, drives tree mortality forecasts at continental scale
  • Schedule
    Notes
  • Location Icon261-262
    NOLA CC
    Set Timezone
  •  
    View Map

Generic 'disconnected' Message
Author(s):
Keenan Ganz, University of Washington Seattle Campus (First Author, Presenting Author)
Liz Van Wagtendonk, University of Washington Seattle Campus
Pratima K C, Bay Area Environmental Research Institute
L. Moskal, University of Washington Seattle Campus


Climate change is making it more likely for trees to die of drought, insect attack, or other causes. Scientists know that these causes work together to kill trees, and it often takes several years of stress before tree death occurs. Since tree death is complex scientists aren't very good at predicting it. We thought this difficulty is because of problems in how we measure tree death. Typically, observers will fly above forests in planes and manually map dead trees. Humans often make mistakes, so modern tools like satellites and computer vision might give more accurate tree death data. If we have more accurate data, then we can make better predictions about future tree death.


To test our hypothesis, we compared tree death predictions based on human observations, satellite imagery, and computer vision. Of these, we found that computer vision gave us predictions that were based on forest properties instead of simple patterns. Computer vision has only recently been used to measure tree death, so we think that this technique should be used more so that scientists can do a better job of predicting tree death.




Scientific Discipline
Suggested Itineraries
Neighborhood
Type
Main Session
Discussion