Forest disturbance imagery can be used to develop forest stand age maps
Time series collections of satellite imagery
can be used to identify disturbances when the forest canopy is either partially
or completely lost through land conversion, weather or insect damage, fire, or
harvesting. With Landsat satellite imagery going back over three decades, it is
possible to create a digital map of annual forest disturbances.
Forest disturbances can be classified as stand-clearing or partial
disturbances, and as natural disturbances (due to fire, weather, or insects),
or human-caused disturbances such as timber harvesting.
A recent study in Virginia by Dr. Joby
Kauffman and Dr. Steve Prisley used several machine learning algorithms to
classify forest disturbances identified by the Vegetation Change Tracker (VCT)
product into stand-clearing or partial disturbances. By identifying the year in
which stand-clearing disturbances occur, it is possible to then develop maps
for forest stand age when these disturbances are followed by forest
The Virginia Department of Forestry (VDOF)
has maintained harvest records since 2009, making it possible to compare
spatial records of timber harvest with disturbances identified by VCT. In an
accuracy assessment using the harvest records, Kauffman and Prisley found that
the VCT product correctly detected forest disturbances 87 percent of the time. Eighty-six
percent of stand-clearing disturbances were correctly classified by the best of
the machine learning algorithms.
The map of stand-clearing disturbances,
combined with year of disturbance, represents an approximate stand age map
(except when stand-clearing disturbances are not followed immediately by forest
The stand age map for Virginia was used to
estimate a forest age-class distribution which closely matched the age-class distribution
reported by forest inventory plots from the FIA program. Furthermore, maps of
this type can be used to study harvest area (and stand size) trends over time, and
to map locations and types of land use conversions following forest clearing.
Subsequent research as part of the Dr.
Kauffman’s dissertation has distinguished partial harvests from non-harvest
disturbances using parcel boundary data from public land ownership records.
These efforts are improving the utility of time-series data from the Landsat
program to generate information useful for forest planning and analysis.
Kauffman, J.S. and S.P. Prisley. 2016.
Automated estimation of forest stand age using vegetation change tracker and
machine learning. Mathematical and
Computational Forestry & Natural Resources Sciences 8(1):4-13.