Classification Methodology for Land Cover Mapping Using Global Monthly 8-Km AVHRR Data
Wen Cheng-Gang and Ryutaro Tateishi
Center for Environmental Remote Sensing (CERES)
Chiba University, 1-33 Yayai-cho, Inage-ku Chiba 263 Japan
Fax :+81-43-290-3857
Email: wen@rsirc.cr.Chiba-u.ac.jp
Abstract
For the purpose of land cover classification of the global area, phenological information and temperature information were used by NDVI and Channel 4 of NOAA/NASA Pathfinder AVHRR Land Cover DATA Set. Phenological information includes onset of greenness, duration of greenness, green peak, total NDVI and a result of cluster analysis of 12 monthly NDVI. These feature images were used in order to identify ground truth data from existing maps.
Introduction
Phenological differences among vegetation types, reflected in temporal variation in the Normalized Difference Vegetation Index (NDVI) derived from satellite data have been used to classify land cover of asia. The objective of this research is to examine and illustrate the value of NOAA AVHRR data for the investigation of vegetation phenology. Specially, we are concerned with the depiction of the land cover of very large areas over time. Phenology is generally accepted as including not only the timing of recurring biological events but also their causes, especially with regard to meteorological phenomena.
Data Sources
10-days composites of NOAA/NASA Pathfinder Land Data Set with8km resolution were used in this study, and is georegistered to GOOD map projection. The size of the data are 5004 rows by 2168 columns.
Composite method
In one month, three 10-days composite data sets are included. The data sets consist of NDVI, CLAVR flag, QC flag, Scan Angle, Solar Zenith, Relative Azimuth, Ch1 Reflectance, Ch2 reflectance, Ch3 Btemp, Ch4 Btemp, Ch5 Btemp and Day of year. In order to remove noise and cloud, the monthly composite data were produced by the maximum composite method as follow:
-
If a 10-days composite NDVI value of a pixel in one month is larger than 0.656, this value is eliminated as a noise.
- If there are one or two or three 10-days composite NDVI value of a pixel in one month corresponding with no-cloud status are below than 0.656, the average of these NDVI values is assigned to a monthly NDVI value. The average of coincident 10-days composite channel 4 values is also calculated as a monthly channel 4 value.
- If there is no 10-days composite data corresponding with no-cloud status but there are one or two or three 10-days composite data with mixed cloud status a pixel in one month, the maximum NDVI value below than 0.656 is chosen as a monthly NDVI value. The coincident 10-days composite channel 4 value is also chosen as a monthly channel 4 value.
- If there is no 10-days composite data corresponding with no-cloud or mixed status of a pixel in one month, the maximum NDVI value below than 0.656 is chosen as the monthly NDVI value. but if all 10-days composite NDVI value are larger than 0.656, the minimum NDVI value is chosen. The coincident 10-days composite channel 4 is also chosen as a monthly channel 4 value.
The NDVI is defined by the equation as followed:
NDVI=(ch2 - ch1) / (ch2+ch1)
Where ch1 represents value from the visible channel (0.58-0.68
mm) and ch2 represents value from near-infrared channel (0.725-1.1
mm). The NDVI is then scaled to Binary NDVI by following equation :
Binary NDVI=(NDVI/0.008) + 128
the range of Binary NDVI value is from 3 to 253.
Image Classification
Land cover classification is carried out by analyzing monthly NDVI composite spanning the January to December 1990 period. through visual interpretation, an NDVI threshold of 0.2 was selected to separate much vegetation and less/no vegetation lands. The threshold was determined by comparison of the strata to available maps and imagery from different countries and regions in Asia and Oceania.
An unsupervised clustering algorithm was used to define 78 phenological classes using 12 monthly NDVI data sets.
