AVHRR, SPOT Vegetation and MODIS Vegetation Index Anomalies:


The Global Inventory Modeling and Mapping Studies (GIMMS) group at NASA Goddard Space Flight Center processes and produces multi-instrument normalized vegetation index (NDVI) products to support USDA’s Office of Global Analysis (OGA) agricultural monitoring activities.

The principal end-user products consist of three sets of vegetation index anomalies derived from:

  • Historical NOAA-AVHRR instruments [7, 9, 11, 14,16 & 17] (NDVIg product): July 1981 - present
  • Systeme Probatoire pour l’Observation de la Terre (SPOT) Végétation : May 1998 – present.
  • NOAA-AVHRR18 (only) : September 2005 – present
  • MODIS-Terra: February 2000 – present.


The anomalies from each data set are computed as follows:

    NDVIσ = [((NDVIα)/(NDVIμ)-1)*100]

where NDVIσ are the respective monthly or 10-day percent anomalies, NDVIα are respective composite period (monthly and 10-day) values and NDVIμ are long-term means for the respective composite periods (monthly and 10-day).

The following mean baseline periods are used
NDVIg: July 1981 – June 2006
SPOT Vegetation: May 1998 – April 2008
NOAA-18: 2005 – 2008 for September-December, and 2006-2009 for January-August
MODIS-Terra: 2000-2007 for February 24 - December 31, and 2001-2008 for January 1 - February 16.

The NDVIg data set was produced using empirical mode decomposition/reconstruction (EMD) to minimize effects of orbital drift, view geometry, volcanic aerosols and clouds. Data points with missing data are filled through interpolation procedures that take into account long-term mean and nearest-neighbor values. NDVIg are screened for clouds in near real time using thresholds on the channel 5 brightness temperature. These tests are applied to Africa, Australia, and South America only. No snow/ice screening is done in near real time for NDVIg since the older AVHRR instruments did not have a channel 3A (SWIR).

The SPOT Vegetation data are screened for cloud, snow/ice, and band data quality using the QA layer provided by VITO. If a pixel QA indicates cloud, cloud shadow, unknown cloud status, snow/ice, or poor quality in B0/B2/B3 then the pixel is flagged and its value is not used in calculations of means or anomalies.

NOAA-18 data was produced to provide a one-instrument only data series as a cross check on the other two data sets and to provide redundancy in the monitoring system. The dataset uses the same channel 5 thresholds as NDVIg to screen for clouds, but also additional tests based on channels 1,2 and 5 (CLAVR algorithms) are used. Also, the SWIR band (3A) is used in conjuction with channel 5 brightness temperatue to screen out snow and ice.

The MODIS-Terra dataset was produced using the MODAPS MOD09A1 Surface Reflectance 8-day 500 m products. MOD09A1 500m tiles were processed in the standard MODIS Sinusoidal grid tiling system and reprojected/mosaiced to produce this 1km Albers Conical Equal Area dataset. MOD09A1 QC and State QA flags were applied to screen for band quality, clouds, water, and snow/ice. These products are “experimental” until the real-time production is completed and operational.

All anomalies are calculated with reference to cloud-free measurements per pixel over the given reference period. Therefore the respective baseline period for each pixel will vary by the number of cloud-free/good data measurements.

Examples of these anomaly products can be found at http://www.pecad.fas.usda.gov/cropexplorer/fews_briefing/ as part of OGA’s support to the USAID FEWSNET project.