1 of 3

Vegetation condition monitoring using

Landsat time series

Valerie Pasquarella

Boston University

2 of 3

https://valpasq.github.io/

condition_monitoring/

Obs - pred_model

RMSE_model

Condition =

https://valeriepasquarella.users.earthengine.app/view/condition-monitoring-explorer

Photo: N. Oalican

2016 → first major Lymantria dispar (defoliating insect, formerly known as gypsy moth) outbreak in Southern New England since 1989
Rapid response: modified existing harmonic change detection approach to map and monitor defoliation → Maps featured as USGS Land Remote Sensing Image of the Week, August 24, 2017
However, original approach implemented on university computing cluster, very labor-intensive process, made a lot of assumptions about baselines/model parameterization that were difficult to test
Fast-forward five years: Original approach translated to Earth Engine and iterated on several times (funding from USFS FHM and Landsat Science Team)
Study that varied spectral inputs, harmonic frequencies, baseline period, and Landsat sensors used (turns out original choices were pretty good!), developed app in collaboration with USFS and other stakeholders
Current Condition Monitoring Explorer allows users to interactively visualize condition change products
Scaled to New England/Mid Atlantic states
View both maps and time series results
Use the full Landsat record for each pixel -- original approach used a single 11-year baseline, current model uses moving window, fit a series of shorter (5-year) baseline models to generate an ensemble of predictions → more flexible, better able to characterize uncertainty across models + monitoring period
Benefits of Earth Engine → Less time downloading images, more opportunities to experimentally test modeling choices, able to scale and share results in ways never before possible
See docs page for overview of methods, link to app

1 of 3

2 of 3

3 of 3