U.S. Department of the InteriorU.S. Geological Survey
USGS/EROS Data Center Global Land Cover Project – Experiences and Research Interests
GLC2000-JRCMarch 2001
U.S. Department of the InteriorU.S. Geological Survey
The USGS/IGBP Global Land Cover Database
U.S. Department of the InteriorU.S. Geological Survey
USGS/IGBP Global Land Cover Database Strategy
U.S. Department of the InteriorU.S. Geological Survey
Classification Methods
Flexible land cover database Unsupervised multi-temporal classification of
1992-1993 AVHRR NDVI data Classification implemented on a continent by
continent basis
Team interpretation to encourage consistency
External peer review of draft results
Validated IGBP land cover layer
U.S. Department of the InteriorU.S. Geological Survey
Continents to World – Combine Maps
Set rules for top and middle level classification systems
Describe land cover, vegetation seasonality, structure, and leave longevity consistently
Hold frequent project meetings to review consistency
Accuracy measured separately for each mapping area
U.S. Department of the InteriorU.S. Geological Survey
Quality of Reference Data is an Important Factor
U.S. Department of the InteriorU.S. Geological Survey
Global Forest Cover Mapping
EROS Data Center FRA2000
50% FOREST
100% AG
DARK
MODEL
BRIGHT
MODEL
Ch
ann
el 2
(N
IR)
100% FOREST
AVHRR Channel 1 (Visible)
Canopy Density Model
U.S. Department of the InteriorU.S. Geological Survey
EDC FRA2000 Project:— Estimating density of forest canopy cover …
62 78 82 68 5846 72 78 80 6454 54 78 78 8248 28 52 84 8226 60 58 60 70
Estimated percent forest
U.S. Department of the InteriorU.S. Geological Survey
A New Global Forest Cover MapImproved USGS global land cover database
U.S. Department of the InteriorU.S. Geological Survey
Current and Future R&D Interests
Continue global land cover database research using new coarse/moderate resolution sensors
Test new techniques/algorithms Integrate satellite imagery with sampling-based
field data
Sampling-based field data
Focus on attributes, themes that are useful for both science and land management
U.S. Department of the InteriorU.S. Geological Survey
Land Cover Techniques at EDC
Unsupervised classification
Decision-tree models
Spectral mixture analysis
Experimental: Co-kriging, KNN, NN
Continued emphasis on database strategy and its
improvement
Stratification before and after clustering
U.S. Department of the InteriorU.S. Geological Survey
Example of Tree Canopy Density
0 20 40 60 80 100
actual (%)
02
04
06
08
01
00
pre
dic
ted
(%
)
A comparison of predicted forest% to actual values
R2 = 0.7374
0 100%
U.S. Department of the InteriorU.S. Geological Survey
Spatial Modeling Techniques for Satellite Imagery-Field Data Integration
= pz pi
k
ipi zw ,
1,
• Spatial models such as KNN, Co-kriging are nonparametric spatial statistics
• Potential tool for extending field measurements to image data/maps
• Mapping vegetation structure measured on permanent plots
Sample1: U1, V1 at (x1, y1)
Sample2: U2, V2 at (x2, y2)
Sample3: V3at (x3, y3)
Calculate U0at (x0, y0)
U.S. Department of the InteriorU.S. Geological Survey
Key Experimental Vegetation Type and Structure Variables
Biomass Net primary productivity Canopy density Canopy height Age Size class DBH Vegetation species, types, associations
U.S. Department of the InteriorU.S. Geological Survey
Summary
EDC is committed to continuing its global land cover R&D
Working with partners is important for USGS