MODIS Land Cover Product Algorithm Theoretical Basis Document (ATBD) Version 5.0 MODIS Land Cover and Land-Cover Change

This document details the structural definition, development process, and functional flow of the MODIS Land-Cover Product. The Land Cover and Land-Cover Change Parameters were proposed by the MODIS Land Team, with Team Member Alan Strahler leading the effort. The Land Cover Parameter is a 1-km product provided on a quarterly basis beginning about one year following the acquisition of a global dataset by the MODIS instrument aboard the EOS-A platform (Terra) in July, 1999. The Land-Cover Change Parameter is a post-launch, near-term parameter also planned for quarterly delivery. Both parameters rely on a 1-km gridded database composited from MODIS Level 2 and 3 products. Inputs include: (1) EOS land/water mask that restricts classification to land regions and shallow water regions; (2) Nadir BRDF-adjusted Reflectances (NBARs) derived from the MODIS BRDF/Albedo product (MOD43B4) in the MODIS Land Bands (1-7), adjusted to nadir view at the median sun angle of each 16-day period; (3) spatial texture derived from Band 1 (red, 250-meter) at 1000-m resolution MODAGTEX); (4) directional reflectance information at 1k for 16-day periods (MOD43B1); (5) MODIS Enhanced Vegetation Index (EVI) at 1km for 16-day periods (MOD13); (6) snow cover at 500m for 8-day periods (MOD10); (7) land surface temperature at 1 km for 8-day periods (MOD11); and (8) terrain elevation information (MOD03). These data are composited over a one-month time period to produce a globally-consistent, multitemporal database on a 1-km grid as input to classification and change characterization algorithms. The Land Cover Parameter recognizes 17 categories of land cover following the International Geosphere-Biosphere Program (IGBP) scheme. This set of cover types includes eleven categories of natural vegetation covers broken down by life form, three classes of developed and mosaic lands, and three classes of non-vegetated lands. Land cover classes are produced by processing the 32-day database using decision tree and artificial neural network classification algorithms to assign land cover classes based on training data. To reduce computational overhead and increase flexibility, classification proceeds by continents. The Land-Cover Change 1-km Parameter is designed to quantify subtle and progressive land-surface transformations, i.e., land cover modifications, as well as obvious and instantaneous changes, such as land cover conversions. As such, it is not a conventional change product that simply compares land cover databases at two different times and identifies changes in categorical land cover. The algorithm for the Land-Cover Change Parameter combines analyses of change in multispectral-multitemporal data vectors with models of vegetation change mechanisms to recognize both the type of change as well as its intensity. The algorithm development and validation efforts for the Land Cover Product are based on a network of test sites developed to represent major global biomes and cover types. Prelaunch efforts have focused on sites for which temporal sequences of Thematic Mapper (TM) and Advanced Very High Resolution Radiometer (AVHRR) data, coupled with fine-resolution land cover and vegetation data, are available, especially in North and South America. In the postlaunch period, a global suite of sites will be used to train the classifier and validate its output. Landsat-7 and ASTER images will be particularly useful MOD12 ATBD Version 5.0—5/1/99 for updating information about test sites and identifying local change processes. The validation procedure will characterize the accuracy of the product as well as provide information that can be used in spatial aggregation to provide land cover and land-cover change data at coarser resolutions. Land Cover Parameter products will be released about three months after the acquisition of a year of 32-day composites for a given region. Full global production of land cover will be reached as global production of input products is achieved. In the interim, a number of prototype land cover products will be produced. The Land-Cover Change Parameter, which requires two years of data, will be similarly phased in, with interim prototypes made available in the early-postlaunch period.