The MODIS Land Cover ProductThe MODIS Land Cover Product

MODIS Land Cover Team

Boston University

GLC 2000 – “FIRST RESULTS” WORKSHOPJRC – Ispra, 18-22 March 2002

Terra Launch on December 18,1999Terra Launch on December 18,1999

MODIS Land BandsMODIS Land Bands

Band Spatial Resolution Wavelength, nm

1 250 m 654–664

2 250 m 860–870

3 500 m 465–475

4 500 m 550–560

5 500 m 1234–1246

6 500 m 1632–1648

7 500 m 2120–2140

MODIS Land Cover ProductMODIS Land Cover Product

• Objective:– Provide a simple land-cover categorization for

biophysical parameterization for GCM, hydrologic, and carbon cycling models

MODIS Land Cover Product: FeaturesMODIS Land Cover Product: Features

– Categorizes land cover according to life-form, cover and height of dominant vegetation type following IGBP-DIS scheme

– Uses data from spectral and temporal domains– Relies on advanced classifier technology—i.e., decision

trees—in a supervised classification mode– Network of global test sites used for algorithm

calibration and validation– At-launch 1-km database derived from AVHRR heritage– Level 3, 1-km spatial resolution, 96-day product; Climate

Modeler’s Grid (1/4°) product also available

Recent Global Land Cover ProductsRecent Global Land Cover Products

• Beta Product, released April 15, 2001– Based on 2 16-day periods of Normalized BRDF-

Adjusted Reflectance (NBARs)• Provisional Product 2001001, released June 15,

2001– Based on 9 16-day periods of NBARs within July 11–

January 15, 2001– Uses prior probabilities to help separate agriculture and

natural vegetation– Includes IGBP classification, secondary classes,

confidence measures– Draws from at-launch product when only 0–2 views are

available or when classification confidence is less than 40%

Recent Global Land Cover Products, Cont.Recent Global Land Cover Products, Cont.

• Validated Product– Due 15 April 2002– Based on MODIS data from 2001– Includes 4 sets of labels, per-pixel confidence measures,

second choices

IGBP Land Cover Units (17)IGBP Land Cover Units (17)

• Natural Vegetation (11)– Evergreen Needleleaf

Forests– Evergreen Broadleaf Forests– Deciduous Needleleaf

Forests– Deciduous Broadleaf

Forests– Mixed Forests– Closed Shrublands– Open Shrublands– Woody Savannas– Savannas– Grasslands– Permanent Wetlands

• Developed and Mosaic Lands (3)– Croplands– Urban and Built-Up Lands– Cropland/Natural Vegetation

Mosaics

• Nonvegetated Lands (3)– Snow and Ice– Barren– Water Bodies

The Land Cover Input DatabaseThe Land Cover Input Database

• Surface Reflectance– Nadir-adjusted surface

reflectance, 7 land bands

• Vegetation Index– MODIS Vegetation Index,

maximum value composite

• Spatial Texture from 250-m Band 2*

– Standard deviation-to-mean ratio in Band 2 (near-infrared), maximum value composite in 32-day period

*To be added later

• Snow Cover*– MODIS Snow Cover Product,

number of days with snow cover

• Land Surface Temperature*– MODIS Land Surface

Temperature, maximum value composite

• Directional Information*– Bidirectional reflectance model

choices from BRDF product

• Ancillary Data– DEM,* Land/Water mask

Global Composite Map of Nadir BRDF-Adjusted Reflectance (NBAR)April 7–22 2001

10 km resolution, Hammer-Aitoff projection,produced by MODIS BRDF/Albedo Team

no data

MODLAND/Strahler et al.

No data True color, MODIS Bands 2, 4, 3

MODIS Nadir BRDF-Adjusted Reflectance

May 25–June 9 2001False Color ImageNIR–Red–Green

NBAR Time TrajectoriesNBAR Time Trajectories

NDVINDVI

EVIEVI

MODIS 500 m Vegetation Indices

MOD13A1 16 dayComposite

(September 30 – October 15, 2000

MODLAND/Huete et al

Test SitesTest Sites

• IGBP-DIS Core/Confidence Sites– Random stratified sampling of classes on 1992-93 IGBP

Global Land Cover Product– 425 sites identified; 413 SPOT and TM scenes acquired;

91% migrated to WWW by BU

• BU STEP Database– 2614 training sites from 645 TM scenes (6/6/00)– About 1000 training sites in current use for supervised

classification

DISCover Core Validation SitesDISCover Core Validation Sites

Supplemental BU Training SitesSupplemental BU Training Sites

STEP DatabaseSTEP Database

• STEP:– System for Terrestrial Ecosystem Parameterization

• Key STEP Parameters– Life form, height, cover fraction, of layers– Leaf type, phenology, periodicity, physiognomy of dominants in

layers– Elevation, moisture regime, perturbation– Classifications: IGBP, BU, EDC SLCRs– Simple description of site and type (words)

• STEP Flexibility– Allows application of many different land cover labeling schemes by

inference of label from parameters in database

MODLand Support ProductsMODLand Support Products

• Six Biomes for LAI / FPAR Algorithm– Used by Ranga Myneni’s radiative transfer model in

retrieving LAI and FPAR

• Six Biomes for Net Primary Productivity (NPP)– Used by Steve Running’s Biome-BGC model in making

the MODIS NPP product

• Fourteen Classes—University of Maryland Legend– We also provide a 14-class product using the University

of Maryland scheme– Preferred by some modelers

Provisional Land Cover Product June 01Provisional Land Cover Product June 01

Northeast Provisional Land Cover Product, Jul 00–Jan 01Northeast Provisional Land Cover Product, Jul 00–Jan 01

Agriculture

Agriculture/Natural Vegetation Mosaic

Mixed Forest

Evergreen Needleleaf Forest

Deciduous Broadleaf Forest

Urban

Classification Confidence Classification Confidence MapMap

Second Most-Likely Second Most-Likely ClassClass

Land Cover ValidationLand Cover Validation

• Statistical Assessment Based on Site Data– Cross-validation provides probability estimate for errors

of omission/commission

• Two sets of site data:– DISCover Core/Confidence sites—Random stratified

sample based on DISCover Land Cover map (Loveland et al., EDC)

– Supplemental sites compiled at BU—no explicit sampling design, but large N

Validation: Accuracy AssessmentValidation: Accuracy Assessment

• Classification Accuracy from Cross-Validation of Training Sites– Hide 20 percent of training sites, classify with remaining

80 percent; repeat five times for five unique sets of all sites

– Provides “confusion matrix”– Not a stratified random sample, but a good within-class

indication of accuracy

Confusion MatrixConfusion Matrix

Global Test Site Confusion Matrix—Provisional Product

Site ClassClass Name 1 2 3 4 5 6 7 8 9 10 11 12 14 15 16 Total

1 Evgn Ndlf 85 0 6 1 30 1 1 4 0 0 0 2 0 1 0 1312 Evgn Brdlf 1 181 0 0 2 0 4 6 1 2 1 1 0 0 0 1993 Decid Ndlf 1 0 5 1 3 0 0 3 0 0 0 1 0 0 0 144 Decid Brdlf 1 3 1 22 22 0 3 5 0 0 0 3 1 0 0 615 Mixed For 12 3 1 10 80 1 1 6 1 1 0 0 0 0 2 1186 Closed Shrublnd 0 1 0 0 1 10 12 2 1 4 0 0 0 0 0 317 Open Shrublnd 1 2 1 0 0 1 51 2 0 9 0 1 0 4 5 778 Woody Savanna 5 11 0 0 6 1 8 28 5 4 0 7 0 1 0 769 Savanna 0 2 2 0 3 2 9 8 25 3 1 1 0 1 0 57

10 Grasslands 0 2 0 1 1 0 39 5 5 44 0 7 0 1 5 11011 Pmnt Wtlnd 0 2 0 1 2 0 1 0 0 0 3 0 0 0 0 912 Cropland 0 7 2 0 4 2 26 14 10 18 0 153 4 3 2 24514 Cplnd/Ntrl Vegn 0 6 0 4 5 1 12 6 8 8 0 19 16 0 1 8615 Snow+ice 0 0 0 0 0 0 3 0 0 1 0 0 0 2 0 616 Barren 0 0 0 0 2 0 20 0 2 5 0 2 0 9 73 113

Total 106 220 18 40 161 19 190 89 58 99 5 197 21 22 88 1333

Classification Outcome

Per-Pixel Confidence OutputPer-Pixel Confidence Output

• Per-pixel Confidence – Based on statistical “boosting” theory– Allows decision-tree classifier to estimate probability of

classification associated with each possible class label– Output for label with highest probability for each pixel

Validation: Other DatasetsValidation: Other Datasets

• Comparison with Community Benchmark Datasets– Global (e.g., DISCover, UMd from AVHRR)

• Comparison with independent maps derived from high resolution data, e.g.,– Humid Tropics: Landsat Pathfinder– Forest Cover: FAO Forest Resources Assessment– Western Europe: CORINE– United States: USGS/EPA MLRC– United States: California Timber Maps (McIver and

Woodcock)

Validation: ComparisonsValidation: Comparisons

• Collaborative Comparisons– BigFoot sites– MODLand Test Sites

Reprojection of BigFootUTM maps to ISIN

BigFoot ResultsBigFoot Results

ConclusionsConclusions

• MODIS Land Cover Product– Draws on AVHRR heritage – Utilizes improved data and techniques– Top-down, globally-consistent approach– Validation plan emphasizes multiple approaches to build

confidence– “Validated” product to be released about 15 April 2002

based on 2001 data