Review, Journal of 
J (2008) Seath - a 
he context of 
on Environmental 
The Pune Slum 
1 information base 
city development. 
7(4), 595-611. 
ssment - entropy 
opment with Fuzzy 
of Geoinformatics, 
i settlements: a 
laam, 
n Series No. 112 
rmance of urban 
ing systems. In: C. 
hallenges of high 
rmany. 
3). Object-oriented 
Paper presented at 
g for Developing 
lysing the spatial 
ing : Typology of 
indicators. Paper 
te Sensing. 
ct-Based Building 
esolution Satellite 
ia Conference on 
| slum, Libya and 
roup Meeting on 
Netherlands, 21- 
nline at 
Identification / 
)S Satellite Data: 
nvironment/pp/mi 
:s 2010/2011: 
ne (UN- 
"ugate, D. (2007). 
je air? Defining 
vJournal, 69(1), 9- 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
GLOBAL LAND COVER CLASSIFICATION USING MODIS 
SURFACE REFLECTANCE PRODUCTS 
Haruhisa Shimoda, Kiyonari Fukue 
* Tokai University Research and Information Center, 2-28-4 Tomigaya, Shibuya-ku, Tokyo 151, Japan - 
(smd, fku)@keyaki.cc.u-tokai.ac.jp 
Commission VIII, WG VIII/8 
KEY WORDS: Global, Land Cover, Classification, Algorithms, Multi-temporal, Multi-spectral 
ABSTRACT: 
The objective of this study is to develop high accuracy land cover classification algorithm for Global scale by using multi-temporal 
MODIS land reflectance products. In this study, time-domain co-occurrence matrix was introduced as a classification feature which 
provides time-series signature of land covers. Further, the non-parametric minimum distance classifier was introduced for time- 
domain co-occurrence matrix, which performs multi-dimensional pattern matching for time-domain co-occurrence matrices of a 
classification target pixel and each classification classes. The global land cover classification experiments have been conducted by 
applying the proposed classification method using 46 multi-temporal(in one year) SR(Surface Reflectance 8-Day L3) and 
NBAR(Nadir BRDF-Adjusted Reflectance 16-Day L3) products, respectively. IGBP 17 land cover categories were used in our 
classification experiments. As the results, SR product and NBAR product showed similar classification accuracy of 99%. 
1. INTRODUCTION 
Land cover maps of global or continental scale are basic 
information for many kinds of applications, i.e. global change 
research, modeling, resource management, etc. Several kinds of 
global land cover maps has been generated, such as IGBP 
DISCover Global Land Cover, UMD Global Land Cover, and 
MODIS Land Cover, etc. and these products have been 
distributed widely. However, accuracies of these global land 
maps were not sufficiently high. Most of these land cover maps 
were generated mainly using NDVI and its seasonal changes. 
However, NDVI data lose most of information contents which 
were originally included in many channels. 
The objective of this study is to develop high accuracy land 
cover classification algorithm for global scale by using multi- 
temporal MODIS land reflectance products. There are two 
kinds product of Surface Reflectance 8-Day L3 product and 
Nadir BRDF-Adjusted Reflectance 16-Day L3 product. Both 
are composed of 7 spectral bands (620-670nm, 841-876nm, 
459-479nm, 545-565nm, 1230-1250nm, 1628-1652nm, and 
2105-2155nm) with 500m ground resolution. The former is the 
atmospheric corrected surface reflectance, while the latter 
corrects the BRDF effects in addition to the atmospheric 
correction. In this report, these products are called SR product 
and NBAR product, respectively. 
2. STUDY AREA AND SOURCE DATA SET 
The target area set in this study covers 140 ' (from 70 north to 
-70 ' south) and 360 ' for latitude and longitude direction, 
respectively. The region is covered by about 280 sinusoidal 
projection(SIN) grids which are distribution granule of SR and 
NBAR products (as shown in Figure 1). The SR and NBAR 
products of about 280 SIN grids were mosaicked and 
transformed to geographic longitude-latitude coordinate system 
With 0.005 degree interval as shown in Figure 2. This 
processing was performed by using MODIS Reprojection Tool 
  
(MRT) which has been distributed from Land Processes DAAC. 
Because SR and NBAR products have been produced in eight- 
day period, mosaic images of 46 scenes were generated as 
classification target data set for one year of 2007. 
h > 
24 28 2829.30 31 32 3334 35 
  
SEES sun 
Figure 1. SIN grid. 
  
Figure 2. A result of mosaic and geometric transform. 
( the scene of 2007.01.01) 
3. PROPOSED CLASSIFICATION ALGORITHM 
3.1 Classification Feature 
In this study, time-domain co-occurrence matrix was introduced 
as a classification feature which provides time-series signature