|
Mapping deciduous fraction in interior Alaska using seasonal composites from Landsat imagery
Richard
Massey, Northern Arizona University, rm885@nau.edu
(Presenter)
Logan
Berner, Northern Arizona University, logan.berner@nau.edu
Brendan
Morris
Rogers, Woods Hole Research Center, brogers@whrc.org
Scott
J.
Goetz, Northern Arizona University, scott.goetz@nau.edu
One of the most fundamental properties for understanding vegetation in the ABoVE domain is leaf habit, i.e. deciduous vs. evergreen. Deciduous trees such as aspen and birch have substantially different impacts on ecosystem function and land-atmosphere interactions compared to conifers, including carbon cycling, albedo, and energy partitioning. Climate warming and increased disturbances, primarily wildfire, are thought to have been increasing the fraction of deciduous vegetation, but this remains largely unquantified. Here we derive spatial estimates of deciduous fraction at 30 m for Alaska using phenology-based metrics.
We initially focused on 2010 due to the availability of reference data sets and dense 30 m Landsat 5 TM and Landsat 7 ETM+ data stacks. We trained a Random Forest classifier using maximum NDVI over three periods: 16th May – 30th June, 1st July – 15th August, and 16th August – 30th September. Training and validation samples were derived from multiple forest inventory sources, including the Cooperative Alaska Forest Inventory (CAFI) and the Bonanza Creek long term ecological research transects. Initial results indicate relatively high accuracy and are consistent with the NLCD 2011 categorical land cover map. Future steps include mapping and validating in different time periods, assessing the major drivers of change, extending our analysis into Canada, and comparing against coarse-scale AVHRR estimates of deciduous fraction.
Associated Project(s):
Poster Location ID: 98
Session Assigned: Vegetation Dynamics and Distribution
|