The project “Enabling Centralized Access to Land Cover Data for Climate Change Integrated Assessment
Modeling” is making selected MODIS land cover data available as global mosaics at resolutions suitable
for use with global models. This project is part of NASA’s Advancing Collaborative Connections for Earth
System Science (ACCESS) program. The Joint Global Change Research Institute, a joint activity of the Pacific
Northwest National Laboratory and the University of Maryland, and the Global Land Cover Facility of the
University of Maryland collaborate in this activity.
Land use is an essential human activity to meet demands for food, fiber, shelter, and natural resources.
Additionally, biomass grown specifically to meet energy demands is increasingly important as an alternative
energy source, and maintaining ecosystems with higher carbon stocks is an option for mitigating climate
change. But land cover changes to meet human demands for resources and land use in themselves cause greenhouse
gas emissions and affect the climate through biophysical interactions. Thus, a variety of global, integrated
assessment and Earth system models represent not only the commitments of land and terrestrial natural resources
to meeting human demands, but also the corresponding changes in terrestrial carbon pools and carbon exchanges
with the atmosphere, other greenhouse gas emissions due to land use, and changes in land surface characteristics
that affect climate.
Global Model Requirements
In terms of temporal stability and consistent spatial coverage, the standard land cover data products derived
from Moderate Resolution Imaging Spectroradiometer (MODIS) observations by NASA’s Terra and Aqua spacecraft
provide one of the highest quality representations of global land cover over a sufficient period to investigate
But a number of factors complicate the use of standard MODIS data products for representing land attributes
within global models:
- Retrieving an attribute from a set of tiles or iterating over tiles in order to represent a land
attribute globally is inefficient and prone to error;
- Within global models, locations typically are specified in terms of latitude and longitude making
spatial reference to MODIS land product tiles in a sinusoidal projection incompatible with model coordinates;
- The storage required to represent a land attribute globally at the native spatial resolutions of MODIS
standard data products is prohibitive in both complex and more simplified models; and
- Many Earth system models are computationally demanding and require efficient interfaces.
While moderate spatial resolution land cover data are useful in studies at local to regional scales,
to be manageable within Earth system models and applications, global geospatial land cover data need
to be at coarser resolutions than are nominal for standard remote sensing data products (e.g., 500 m
MODIS land cover data).
For example, a single byte attribute of the standard MODIS land cover type data product at the 500 m
resolution requires approximately 3 Gbytes of storage for global coverage. As a result, within integrated
assessment models and more complex Earth system models, land attributes typically are represented at
resolutions of 5' latitude × 5' longitude or 2.5' × 2.5'. Simplified models may represent land attributes
at even coarser resolutions such as 0.5° latitude × 0.5° longitude.
Storage requirements for 1 Byte attributes at typical resolutions.
|500 m Pixels
|5' x 5'
|0.5° x 0.5°
The complexity of data structures is also an issue. The data structures required
to manipulate a collection of tiles are significantly more complicated than a
single array of uniform pixels covering the Earth’s surface. While in many cases
operations involving a single array can take advantage of matrix representations
and efficient algorithms for array manipulation and processing, iterating over a
set of tiles while maintaining their geographic context can be inefficient, complicated
to implement, and error prone. For example, a simple operation on a single array
such as extracting a two-dimensional slice of data can be complicated with a set
of separate tiles where the slice spans parts of multiple tiles.
Global Land Cover Mosaics
Global mosaics of the standard MODIS data product MCD12Q1, Version 5.1, in the IGBP classification of land cover
types were derived from the standard MODIS data product MCD12Q1 Version 5.1 collection. The project processed
data for each year in the period 2001-2012.
The standard MODIS land cover data product is produced at 500 m spatial resolution in a sinusoidal projection.
In order to provide data in convenient file sizes, the Earth’s surface is divided into a grid of non-overlapping
tiles, of which 460 contain land cover data. At the Equator, each tile covers approximately 10° latitude x 10°
longitude (Figure 1).
Figure 1. Grid of non-overlapping tiles that organize the standard MODIS land cover type data product in a
In order to facilitate the use of MODIS land cover data with other observations, especially imagery
collected by Landsat satellites, we reprojected MODIS tiles containing data into geographic coordinates
of latitude and longitude and created mosaics of adjacent tiles sufficient to cover areas in geographic
coordinates that align well with Landsat tracks (Figure 2).
Figure 2. GLCF tile scheme used to organize land cover data reprojected into geographic coordinates.
Using the Geospatial Data Abstraction Library (GDAL), continental mosaics of tiles on the native
sinusoidal grid (Figure 1) were created (gdal_merge). The gdalwarp utility was used to reproject
these continental mosaics into geographic coordinates of latitude and longitude and to crop these
reprojected continental mosaics to GLCF tiles (Figure 2). Insofar as possible, native resolution
Global mosaics of the re-projected tiles were created using the gdal_merge utility. At the native 500 m spatial
resolution of the MODIS data, each global mosaic comprised approximately 3 gigabyte of 1 byte data in GeoTIFF format.
Land cover types were assigned to coarser resolution elements by assuming the most frequently occurring type at
the higher resolution of the source data. A Python script was used to aggregate the native resolution elements
contained within global mosaics. A frequency distribution of land cover types assigned to native resolution
pixels within each coarser resolution element was derived using the bincount function of the NumPy package
for scientific computing with Python. The most frequently occurring type was assigned to the coarser resolution
Global Mosaics of Vegetation Continuous Fields
Processing of vegetation continuous fields data (e.g., the MOD44B standard MODIS product describing
fraction of tree cover), is identical to that for land cover type; however, the native resolution
of the VCF data is 250 m, and processing contends with unassigned pixels.
Each pixel of the arrays containing spatially aggregated data is assigned a value in the range
0 <= z <= 100 expressing the fraction of the area of the pixel in tree cover as [%]. The value
z = 200 indicates a pixel containing water, and the value z = 253 designates a pixel not assigned data.
Where a coarser resolution pixel (5' x 5' or 0.5° x 0.5°) contains native resolution (250 m)
pixels that are not assigned a value (z = 253), a value is estimated for the coarser resolution only
if the ration of native resolution pixels not assigned data to the total number of native resolution
pixels within the coarser resolution pixel is less than a threshold:
- Coarse resolution of 5' x 5': 5%
- Coarse resolution of 0.5° x 0.5°: 20% .
Where the fraction of pixels not assigned data is less than the threshold, percent tree cover is
estimated as the ration of tree cover area represented by pixels assigned data to the total area
of pixels assigned data.
In order to provide compatibility with land cover data, all elements assigned as water in the land
cover data were also assigned as water in the corresponding VCF data. Since the land cover data
begin in 2001, the 2001 data were used to assign water in the 2000 VCF data.
Friedl, M. A., D. Sulla-Menashe, B. Tan, A. Schneider, N. Ramankutty, A. Sibley,, and X. Huang. 2010.
MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets.
Remote Sensing of Environment 114:168–182.
García-Mora, T. J., J.-F. Mas, and E. A. Hinkley. 2012. Land cover mapping applications with MODIS: A
literature review. International Journal of Digital Earth 5:63–87.
Hansen, M. C., R. S. DeFries, J. R. G. Townshend, M. Carroll, C. DiMiceli, and R. A. Sohlberg. 2003.
Global percent tree cover at a spatial resolution of 500 meters: First results of the MODIS Vegetation
Continuous Fields algorithm. Earth Interactions 7:1-15.
Strahler, A., D. Muchoney, J. Borak, M. Friedl, S. Gopal, E. Lambin, and A. Moody. 1999. MODIS Land
Cover Product Algorithm Theoretical Basis Document (ATBD) Version 5.0: MODIS Land Cover and
Land-Cover Change. Boston University, Boston, Massachusetts.
Townshend, J., M. Hansen, M. Carroll, C. DiMiceli, R. Sohlberg, C. Huang. User Guide for the MODIS
Vegetation Continuous Fields Product Collection 5, Version 1.
Zhan, X., R. DeFries, M. Hansen, J. Townshend, C. DiMiceli, R. Sohlberg, and C. Huang. 1999. MODIS
Enhanced Land Cover and and Cover Change Product Algorithm Theoretical Basis Documents (ATBD)
Version 2.0. Department of Geography, University of Maryland, College Park, Maryland.
Geospatial Data Abstraction Library (GDAL). http://www.gdal.org/.
MODIS Land Data Product Grids. http://modis-land.gsfc.nasa.gov/MODLAND_grid.html.
NumPy Fundamental Package for Scientific Computing with Python. http://www.numpy.org/.