Using temporal filters to deduce the cause of forest disturbances detected with time series of Landsat data
Sean
Healey, US Forest Service, seanhealey@fs.fed.us
(Presenting)
Scott
Powell, US Forest Service, scottpowell@fs.fed.us
Randall
Morin, US Forest Service, rsmorin@fs.fed.us
Warren
Cohen, US Forest Service, wcohen@fs.fed.us
Gretchen
Moisen, US Forest Service, gmoisen@fs.fed.us
Samuel
Goward, University of Maryland, sgoward@geog.umd.edu
Jeffrey
Masek, NASA Goddard, Jeffrey.G.Masek@nasa.gov
Andrew
Lister, US Forest Service, alister@fs.fed.us
Robert
Kennedy, US Forest Service, robertkennedy@fs.fed.us
Chenquan
Huang, University of Maryland, cqhuang@umd.edu
The Forest Inventory and Analysis (FIA) unit of the Forest Service monitors the status and trends of the nation’s forests. In collaboration with NASA and the North American Carbon Program, FIA is investigating the potential use of biennial Landsat imagery to support monitoring of forest disturbance. Digital change detection methods are being coupled with FIA plot data to map the year and magnitude of forest disturbances using time series of Landsat imagery from 30 scenes across the country. Complicating change detection efforts is the fact that defoliation events (insects, wind) can create spectral signals similar to those from disturbances that permanently alter forest structure. Differentiating these disturbance types is important with respect to FIA reporting; harvests, for example, are viewed quite differently than insect activity by the forest managers and policy-makers who rely upon FIA data. Likewise, the carbon flux involved with defoliation is significantly different than the flux associated with removal of woody material. In a pilot study in western Pennsylvania, we used a post-disturbance temporal filter to discriminate between gypsy moth defoliation and harvest activity. Though these types of disturbance are typically similar both spatially and spectrally immediately following disturbance, defoliated stands (as identified with independent management records) returned to near pre-disturbance spectral values within two years while harvested stands took much longer. Post-disturbance temporal filtering was found to be an accurate and, given this project’s acquisition of relatively dense time series of imagery, expedient way to refine the change detection process.