ASPRS

PE&RS September 1999

VOLUME 65, NUMBER 9
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
JOURNAL OF THE AMERICAN SOCIETY FOR PHOTOGRAMMETRY AND REMOTE SENSING

Peer Reviewed Articles

1013 The IGBP-DIS Global 1-Km Land-Cover Data Set DISCover:  A Project Overview
Alan S. Belward, John E. Estes, and Karen D. Kline

Abstract
Since 1992 the International Geosphere Biosphere Programmer's (IGBP) Data and Information System (DIS) has been working towards the completion of a validated global land cover data set, DISCover. This 1-km resolution data set consists of 17 cover classes identified on the basis of the science requirements of the IGBP's core projects. DISCover has been created from over 4.4 Terabytes of data from the Advanced Very High Resolution Radiometer collected from 23 receiving stations. These data were processed and assembled into a coherent set of monthly Normalized Difference Vegetation Index composites (April 1992 to April 1993) and classified using unsupervised techniques with past-classification refinement. The first global land-cover classification was completed in July 1997. The IGBP-DIS Land Cover Working Group, in turn, convened a Validation Working Group to provide and implement a validation method to provide statistical statements concerning the accuracy of the global land-cover product and to allow the estimation of the error variance in areal totals of classes globally and within regions. The validation workshop was completed in September 1998 and the analysis by March 1999. This paper describes the history of the DISCover version 1.0 implementation. 

1021 An Analysis of the IGBP Global Land-Cover Characterization Process
Thomas P. Loveland, Zhiliang Zhu, Donald 0. Ohlen, Jessyln F Brown, Bradley C. Reed, and Limin Yang

Abstract
The International Geosphere Biosphere Programme (IGBP) has called for the development of improved global land-cover data for use in increasingly sophisticated global environmental models. To meet this need, the staff of the U.S. Geological Survey and the University of Nebraska-Lincoln developed and applied a global Land-cover characterization methodology using 1992-1993 1-km resolution Advanced Very High Resolution Radiometer (AVHRR) and other spatial data. The methodology, based on unsupervised classification with extensive postclassification refinement, yielded a multi-layer database consisting of eight land-cover data sets, descriptive attributes, and source data. An independent IGBP accuracy assessment reports a global accuracy of 73.5 percent, and continental results vary from 63 percent to 83 percent. Although data quality, methodology, interpreter performance, and logistics affected the results, significant problems were associated with the relationship between AVHRR data and finescale, spectrally similar land-cover patterns in complex natural or disturbed landscapes. 

1033 Landsat Thematic Mapper Registration Accuracy and its Effects on the IGBP Validation
Gregory J. Husak, Brian C. Hadley, and Kenneth C. McGwire

Abstract
Research associated with the International Geosphere/Biosphere Programme Data and Information System Cover (DISCover) validation exercise revealed possible registration errors in the high-resolution data employed. Systematically corrected National Landsat Archive Production Systems (NLAPS) Thematic Mapper (TM) data used in the validation of the DISCover dataset was compared with precision registered Multi-Resolution Land Characteristics (MRLC) TM data available only for the conterminous United States. A consistent offset in the systematically corrected data was discovered to be approximately 2 km to the east and south. The impacts of this bias on the validation of the IGBP DISCover dataset are examined and show that possibly 20 percent of the IGBP pixels for the conterminous United States could be adversely affected by this registration error. Fractal dimension is shown to have a close relationship with the effect of the offset on a class and may be used as a predictive tool for areas outside the conterminous United States. 

1041 Image Interpretation Keys for Validation of Global Land-Cover Data Sets
Melissa Kelly, John E. Estes, and Kevin A. Knight

Abstract
Photointerpretation keys originated from the need to efficiently train aerial photographic interpreters. Keys provide interpreters with a baseline of signature information that can be widely applied to images illustrating differing land covers, features. facilities, processes, and activities. Keys were produced far the International Geosphere Biosphere Programme Data Information Systems (IGBP-DIS) DISCover Validation effort. This effort had two objectives: to provide a prototype set of image interpretation keys that could be produced efficiently and cost effectively and to create a foundation of enduring reference materials that could be refined and enlarged to facilitate the validation of future global lond-cover products. The methodology of key development and production for this effort originated as an ideal, but has by necessity been shaped by pragmatic considerations. Factors including cost, time, and availability of imagery and ancillary data limited broad applicability of the keys developed. Research included determination of the most effective band combination, signature development and application, consistent implementation of a minimum mapping unit, use of signature elements in the evaluation of key use, and development of a product useful for cross applications. We continue to expand the present set of keys and develop a digital infrastructure for a global reference library of key materials. These materials will expedite future validation efforts and lead to greater confidence in the accuracy of future global land-cover products. 

1051 Thematic Validation of High-Resolution Global Land-Cover Data Sets
Joseph Scepan

Abstract
This paper describes a procedure to validate the thematic accuracy of the International Geosphere-Biosphere Programme. Data and Information System (IGBP-DIS) DISCover (Version 1.O) l kilometer Global Land-Cover Data Set. Issues of data set sampling design, image geometry and registration, and core sample interpretation procedures are addressed. Landsat Thematic Mapper and SPOT satellite image data were used to verify, 379 primary core samples selected from DISCover 1.O using a stratified random sampling procedure. The goal was to verify a minimum of 25 samples per DlSCover class; this was accomplished for 13 of the 15 verified classes. Three regional Expert Image Interpreters independently verified each sample, and a majority decision rule was used to determine sample occuracy. For the 15 DISCover classes validated, the average class accuracy was 59.4 percent with accuracies for the 15 verified DISCover classes ranging between 40.0 percent and 100 percent. The overall area-weighted accuracy of the data set was determined to be 66.9 percent. When only samples which had a majority interpretation for errors as well as for correct classification were considered, the average class accuracy of the data set was calculated to be 73.5 percent. 

1061 The IGBP DiSCover Confidence Sites and the System for Terrestrial Ecosystem Parameteriza-tion: Tools for Validating Global Land-Cover Data
Douglas Muchoney, Alan Strahler, John Hodges, and Janet LoCastro

Abstract
The IGBP Validation Confidence Site database provides a set of 379 land-cover maps, each containing an IGBP Core Validation sample. Each map is 448 km² in area and is delineated and labeled by photointerpretation of Landsat or SPOT satellite imagery at a scale of 1:125,000. Within each map, land-cover types and polygons are assigned descriptive labels and parameter codes for vegetation attributes, including life form, cover, height. and phenology for canopy and ground layers. These attributes are a subset of parameters defined by the System for Terrestrial Ecosystem Parameterization (STEP), a site model and database that characterizes land surface and vegetation for use in global algorithm training, testing, and validation of land-cover data. Because the maps are linked to the core samples, they provide a large, consistent dataset that is stratified to represent equally all of the world's major vegetation form classes and land-cover types. The confidence site database has three primary applications: (1) as a set of validation benchmarks for alternate regional or global landcover classifications emphasizing vegetation attributes, (2) as a secondary information source fur studying core sample accuracy issues, and (3) as a source of training and test sites for regional and global supervised classification of coarse-resolution satellite imagery.

1069 The Global Land-Cover Characteristics Data-base: The Users’ Perspective
Jesslyn F Brown, Thomas R. Loveland, Donald 0.Ohlen, and Zhi-liang Zhu

Abstract
A unique global land-cover characteristics database developed by the U.S. Geological Survey has been available to users since mid-1997. Access to the date is through the Internet under the EROS (Earth Resources Observation Systems) Data Center's home page (http://edtwww.cr.uags.gov/landdaac/glcc/glcc.html). Since the release of the database, the data have been incorporated into various environmental research and modeling applications, including mapping global biodiversity, mesoscale climate simulations, carbon cycle modeling, and estimating habitat destruction. Since the early stages of the project, user feedback has provided a means to understand data utility in applications, garner suggestions for data improvements, and gain insights into the technical challenges faced by users. Synthesis of user feedback provided a means to generate a user profile and derive a list of applications-critical criteria for landcover data. User suggestions have lead to revisions in the database, including label changes, alternative classification schemas and additional projections for the data. 

1075 The DISCover Validation Image Interpretation Process
Joseph Scepan, Gunter Menz, and Matthew C. Hansen

Abstract
Thematic validation of the International Geosphere Biosphere Data and Information System (IGBP-QIS) Global 1-Kilometer Land-Cover Data Set (DISCover) was performed utilizing a "state-of-the-practice" technique by a team of Expert images Interpreters (EII) examining subscenes extracted from 379 digital Landsat (Thematic Mapper) and SPOT images. The 15 validated IGBP Land-cover classes (Snow/Ice and Water were not assessed) were not equally interpretable on the TM and SPOT imagery. Interpreter confidence was highest for Evergreen Broadleaf Forests and Urban/BuiIt-up DISCover classes while Grasslands and Permanent Wetlands were interpreted with relatively less confidence. Analysis of image interpretation in each of the 13 validation regions indicates that confidence in interpretations for North America/Canada (Region 1) and Central Asia/Japan (Region 11) are lower than average. Confidence in interpretations is significantly higher than average for North America/US (Region 2). Northern and Southern South America (Regions 4 and 5), and Southest Asia and China (Region 12). In this study, variations in interpretation confidence are also noted between regions or based upon the geographic location of samples.

This exercise demonstrates that Landsat TM and SPOT imagery can be efficiently used to validate high-resolution global land-cover products. The results suggest that the utility of and confidence that may be placed in this technique depends upon the land-cover classification scheme used and the quality of digital and ancillary data available to aid interpreters. Another important factor is the relative confidence of the interpreters to verify the land cover within their respective areas of the globe.

1083 Implications of Land-Cover Misclassification for Parameter Estimates in Global Land-Surface Models: An Example from the Simple Biosphere Model (SiB2)
R.S. DeFries and S.O. Los

Abstract
One of the primary applications of the global 1-km land-cover DISCover product is to derive biophysical and ecological parameters for a range of land-surface models, including biosphere-atmosphere, biogeochemical, and ecological models. The validation effort reported in this special issue enables a realistic assessment of the implications of misclassification errors for parameter estimates within the models. In most land-surface models, cover types are aggregated to coarser groupings than the 17 IGBP classes for estimating parameters, with aggregation schemes varying with individual models and individual parameters within each model. Misclassification errors are consequential only when they occur between cover types that are not aggregated by the model. We use examples of two biophysical parameters-leaf area index and surface roughness-as estimated for use in the Simple Biosphere Model (SiB2) and other modeling applications to quantify the effects of misclassification on parameter estimates, SiB2 relies on satellite data as well as land-cover information for estimating the biophysical parameters. Consequences of misclassification are likely to be greater for those models that do not use satellite data. Mean class accuracy based on those sites for which a majority of interpreters agreed (percentage of validation pixels classified correctly out of total number of validation pixels, averaged over all classes), adjusted by area of each cover type in the IGBP Discover product, is 78.6 when all misclassification errors are included. By excluding misclassification errors when they are inconsequential for leaf area index and surface roughness length estimates, mean class accuracies are 90.2 and 87.8, respectively The results illustrate that misclassification errors are most meaningfully viewed in the context of the application of the land-cover information. 

1089 The Way Forward
John Estes, Alan Bel ward, Thomas Loveland, Joseph Scepan, Alan Strahler, John Townshend, and Chris Justice

Abstract
This paper focuses on the lessons learned in the conduct of the International Geosphere Biosphere Program's Data and Information System (IGBP-DIS), global 1-km Land-Cover Mapping Project [DISCover).There is still considerable fundamental research to be conducted dealing with the development and validation of thematic geospatial products derived from a combination of remotely sensed and ancillary data. Issues include database and data product development, classification legend definitions, processing and analysis techniques, and sampling strategies. A significant infrastructure is required to support an effort such as DISCover. The infrastructure put in place under the auspices of the IGBP-DIS serves us a model, and must he put in place to enable replication and development of projects such as DISCover.
Top Home