In: Wagner W., Sz6kely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010,1 APRS, Vol. XXXVIII, Part 7B 
703 
EVALUATION OF TIME-SERIES OF MODIS DATA FOR TRANSITIONAL LAND 
MAPPING IN SUPPORT OF BIOENERGY POLICY DEVELOPMENT 
F. Zhou, A. Zhang, H. Wang, G. Hong 
Canada Centre for Remote Sensing, Booth Street, Ottawa, Ontario, K1A 0Y7 Canada - 
(Fuqun.Zhou, Aining.Zhang, Huili.Wang, Gang.Hong)@nrcan.gc.ca 
KEYWORDS: Land cover, Vegetation, Change Detection, Data mining, Research, Decision Support 
ABSTRACT: 
Demanding for information on spatial distribution of biomass as feedstock supply and on land resources that could potentially be used for 
renewable bioenergy production is rising as a result of increasing government investment for bioenergy and bioeconomy development, 
and as a way of adaptation to climate warning. Lands transitioned over the past between the types of forest, grassland, forage land, and 
cropland are considered as the most promising for the production of dedicated bioenergy crops as a primary source of biomass feedstock 
for the development of the second generation biofuels, without compromising regular agriculture production. 
Aimed at the transitional land mapping at a region scale, Earth Observation data with medium spatial resolution are considered as one of 
the most effective data sources. Time series of 10 days cloud-free composite MODIS images and its derivation, NDVI and vegetation 
phenology in the vegetation-growing season, are then used to derive the required information. With these datasets, three groups of data 
combinations are explored for the identification of the best combinations for land cover identification, then for transitional land mapping, 
using a data mining tool. 
Results showed that longer time series of Earth Observation data could lead to more accurate land cover identification than that of shorter 
time series of data; Bands (1-7) only and NDVI or phenology with other bands (3-7) could yield almost the same highest accurate 
information. Results also showed that land cover identification accuracy depends on the degree of homogeneity of the landscape of the 
region under the study. 
1. INTRODUCTION 
Demanding for information on spatial distribution of biomass 
as feedstock supply is rising as a result of increasing 
government investment for bioenergy and bioeconomy 
development, and as a way of adaptation to climate warning. 
The recent global food crisis has raised an important question 
of how to develop a sustainable bioenergy and bioeconomy 
without compromising the food production for a growing world 
population. A response to this challenge involves information 
on land cover, both for lands currently used in crop production 
and for transitional lands that could be used for growth of 
dedicated energy crops. Such information will have to be 
available over large areas, usually at regional, even at national 
scale (Zhou et al„ 2009; Gang et al., 2010). 
Transitional lands refer to the lands transitioned over the past 
between types such as forest, grassland, forage land, and 
cropland due to factors such as climate or market fluctuations 
or other reasons. This category of lands is considered most 
promising for the production of dedicated bioenergy crops as a 
primary source of biomass feedstock for the development of the 
second generation biofuels, without compromising regular 
agriculture production. Identification of these lands for biomass 
growth potential is a gap in the Canadian national biomass 
inventory, which has been undertaken by Canadian Forest 
Service and Agriculture and Agri-food Canada, with funding 
from the federal, interdepartmental Program for Energy 
Research and Development (PERD) operated by Natural 
Resources Canada. The purpose of the transitional land 
mapping is to enable the application of earth observation data 
to fill this gap and to provide science-based information in 
support of bioenergy policy development. 
Aimed at the task at a regional scale, Earth Observation data 
with medium spatial resolution are considered as the most 
effective data sources. Time series of the Moderate Resolution 
Imaging Spectroradiometer (MODIS) images and its derived 
information, such as Normalized Difference Vegetation Index 
(NDVI) and vegetation phenology, are then used to derive the 
required information. Time series of data have many 
advantages compared to a single time image as the former 
captures dynamic spectral information at various vegetation 
growth stages. However, it also poses challenge of such as how 
to efficiently use the ‘high-dimensional’ and a large volume of 
data and extract needed information for the issues at hand 
(Zhou et al., 2009). In this regard, a data mining tool is applied 
to explore and identify the optimal data combinations from all 
the data available in the vegetation growing season for land 
cover identification, and then for transitional land mapping. 
Results showed that, in general, longer time series of data set 
would yield higher accurate land cover classification than that 
of a shorter time series of data, and due to the medium spatial 
resolution of MODIS data, land cover identification accuracy 
also depends on the degree of homogeneity of the landscape 
setting of the region under study. The following sections will 
describe in some details about the data used, the methodology