U.S. Department of the Interior
U.S. Geological Survey
Collin Homer, USGS
Matt Bobo, BLM
Remote Sensing Characterization and
Monitoring of Shrubland Components across
the Western United States
1989 Landsat
2006 Landsat
What our brain can see and process, is much more difficult to code in an algorithm and then repeat….again and again under changing circumstances…
The ability of remote sensing to offer measurement and
monitoring of the landscape at first glance seems obvious…
Guiding principles for USGS shrubland remote
sensing component characterization
Recognize the difficult remote sensing issues
of semi-arid shrub lands require robust approch
Create consistent data over large areas
Use a method that quantifies shrub components in a repeatable, objective way
Support work at multiple spatial scales and time periods
Provide a good foundation for monitoring
Sustainable and affordable
Fractional vegetation
How to distill remote sensing data into meaningful information?
Labeled categories
% Bare Ground fraction Land Cover classes
-More quantifiable and repeatable
-Can support many applications
-Easily generated across scales
-Easier to monitor subtle change
-Usually needs more effort to apply
-Easier to automate
-Good for very targeted applications
-Can be easier to understand
-Difficult to repeat and monitor
-Not application generic
-Only works for targeted scale
-Hard to automate
Shrub/Bare ground products started to be developed in 2004
1 Meter Frame Components include: •Bare ground percent
•Shrub percent
•Herbaceous percent
•Litter percent
•Sagebrush percent
•Big Sagebrush percent
•Wyomingensis Sagebrush percent
•Shrub height
Component proportions are field
measured as 1m frames and
extrapolated to raster cells at various
resolutions (2.4m, 30m and 56m) using
regression tree models using 1%
increments
30m Landsat Shrub prediction
Percent Bare Ground modeled @ 3 scales
2.4m 30m
56m
Products can be produced at multiple scales,
with the same approach
1$ per acre 1 cent per acre 1/5 cent per acre
Elevation Slope
Spring Summer
Fall
Regression tree models are used to “mine” large amounts
of data to create target predictions
Input Database
Shrub estimate
1 % intervals
Regression Tree
Ancillary Data
Aspect
Position Index
Training
Data
Shrub
Approach first regionally developed in Wyoming
Approach now has been further scaled up and
implemented in the western U.S. with BLM AIM and
USGS funding
New advancements include special Landsat 8
mosaics and climate zones modeling
Mask
Mask
Bare Ground Prediction
Sagebrush Prediction
WorldView2 Imagery
High resolution
training first developed
at 2 meter resolution
Clover valley ranch, Ca
Mask
Sagebrush
Prediction Bare ground
Prediction
Then applied wall to wall at 30m resolution
All Sage Cover (%)
ValueHigh : 102
Low : 0
Annual Herbaceous Cover (%)
ValueHigh : 102
Low : 0
Bare Ground (%)
ValueHigh : 102
Low : 0
0
All Sage Height (cm)
ValueHigh : 178
Low : 0
All Shrub Height (cm)
ValueHigh : 428
Low : 0
All Sage Cover (%)
ValueHigh : 102
Low : 0
Annual Herbaceous Cover (%)
ValueHigh : 102
Low : 0
Bare Ground (%)
ValueHigh : 102
Low : 0
0
All Sage Height (cm)
ValueHigh : 178
Low : 0
All Shrub Height (cm)
ValueHigh : 428
Low : 0
All Sage Cover (%)
ValueHigh : 102
Low : 0
Annual Herbaceous Cover (%)
ValueHigh : 102
Low : 0
Bare Ground (%)
ValueHigh : 102
Low : 0
0
All Sage Height (cm)
ValueHigh : 178
Low : 0
All Shrub Height (cm)
ValueHigh : 428
Low : 0
All Sage Cover (%)
ValueHigh : 102
Low : 0
Annual Herbaceous Cover (%)
ValueHigh : 102
Low : 0
Bare Ground (%)
ValueHigh : 102
Low : 0
0
All Sage Height (cm)
ValueHigh : 178
Low : 0
All Shrub Height (cm)
ValueHigh : 428
Low : 0
All Sage Cover (%)
ValueHigh : 102
Low : 0
Annual Herbaceous Cover (%)
ValueHigh : 102
Low : 0
Bare Ground (%)
ValueHigh : 102
Low : 0
0
All Sage Height (cm)
ValueHigh : 178
Low : 0
All Shrub Height (cm)
ValueHigh : 428
Low : 0
Herbaceous Cover (%)
ValueHigh : 102
Low : 0
All Shrub Height (cm)
ValueHigh : 428
Low : 0
Litter Cover (%)
ValueHigh : 102
Low : 0
All Shrub Height (cm)
ValueHigh : 428
Low : 0
Big Sage Cover (%)
ValueHigh : 102
Low : 0
All Shrub Cover (%)
ValueHigh : 102
Low : 0
0
High : 100
Low : 0
High : 100
Low : 0
High : 100
Low : 0
High : 100
Low : 0
High : 100
Low : 0
High : 100
Low : 0
High : 100
Low : 0
Shrub component
list (with
estimates in 1%
increments)
now being
produced
Mask
Mask
Mask
Shrub Prediction
Bare Ground Prediction
Shrub Absolute Error
Bare Ground Absolute Error
Mask
Validation includes independent validation, cross validation and a
spatial absolute error model prediction with products
R² = 0.4117 0
5
10
15
20
25
30
35
0 10 20 30 40)
Shrub Cover: Independent Validation
R² = 0.7202 0
20
40
60
80
100
0 20 40 60 80 100
Bare Ground: Independent Validation
R² = 0.5182 0
5
10
15
20
25
30
35
40
-15 5 25 45 65
Herbaceous: Independent Validation
R² = 0.6101 0
10
20
30
40
50
60
70
0 20 40 60 80
Litter Cover: Independent Validation
SW
Landscape Independent Validation
Shrub, grass and
bare ground
product areas
planned
through 2015 (only nominal
base year funded)
Component Tested Applications include:
Wyoming sage grouse (core area refinement and
state-wide seasonal models)
Gunnison sage grouse habitat models
Wyoming grazing assessment (clear component
differences in allotments that failed LHS)
Capturing climate-based vegetation change over time
Future forecasting of vegetation change using
precipitation scenarios
In progress: fire fuel analysis, invasive monitoring,
restoration monitoring, climate temperature analysis
WLCI Bare Ground Prediction
2006 and 2010
2006
2010
Fire event captured
with an increase in
the 2010 Bare
Ground estimate
Percent
Wyoming bare ground change analysis across time
using Landsat (30m)
Annual precipitation
and remote sensing
component change is
significantly correlated
across 4 years in
Wyoming
Research results
from field
monitoring
Mean annual
precipitation and
sagebrush component
predictions from 1984
to 2011 over the study
area , with the linear
trend line.
Research results: Can quantify climate
change over time with
models looking at
continuous field
components and
precipitation historically,
and then quantify future
component change
predictions.
Precipitation
Bare Ground Change
Difference
-4 to -31
-1 to -3
0
1 - 3
4 - 33
Mask
Shrub
Bare Ground
Research Results: Bare Ground and Shrub
2006-2050 Change Examples using
future forecasting models
A1B Climate Scenario Result Bare ground - 1.1% increase
Shrub – 4% decrease
Sage grouse
nesting habitat
prediction,
year 2050
1993 1997
2004 2009
2011
Nevada example of increasing cheatgrass abundance, 1993-2011
White – masked out areas SW of Hot Springs Range
Cheatgrass quantity
Products are under the National Land Cover
Database Umbrella……
What will the future look like…..
Once base characterization is done, then plan to
produce products back in time to 1984
Moving forward products will be repeated every 5
years under NLCD umbrella
Potential for quick annual update under high
disturbance scenarios (fires)
Likely to continue to evolve to finer time and space
scales
Become a fundamental monitoring product that
underpins a broad variety of sagebrush steppe
habitat, management and scientific analysis
NLCD is a Landsat derived 30m
cell land cover product
database covering the United
States created by 10 Federal
partners (MRLC)
www.mrlc.gov