, Part 7B 
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Yol. XXXVIII, Part 7B 
511 
CHARACTERISATION OF LONG-TERM VEGETATION DYNAMICS FOR A 
SEMI-ARID WETLAND USING NDVI TIME SERIES FROM NOAA-AVHRR 
R. Seiler 3 
TU Dresden, Dept. Geosciences, Helmholtzstrasse 10-12, 01062, Dresden, Germany-rseiler@rcs.urz.tu-dresden.de 
KEY WORDS: Land Cover, Statistics, Change Detection, Modelling, Multitemporal 
ABSTRACT: 
The Niger Inland Delta represents a flat area of around 40.000 km 2 , which is annually inundated by the Niger River system. As 
the flood is driven by the rainfall in the catchment areas, it is not linked to the low precipitation of the Sahelian region. Thus, 
local rainy season and inundation show a temporal delay of 3 months and the Niger Inland Delta's ecology can be described as a 
mosaic of permanent, periodical and non-periodically flooded areas. AVHRR GIMMS Data provide NDVI values over 25 years 
with 2 data / month on a 8 x 8 km grid. Dynamics in vegetation density were modelled from the temporal variability of the NDVI. 
Therefore each time series was detrended and transformed into the frequency domain. The power spectra then were decomposed 
into a long-term cyclic component by applying a FIR with a cut-off frequency slightly lower 1 cycle / year, a seasonal (annual) 
and an irregular component. For modelling the seasonal component of a time series, an algorithm is proposed that reduces the no. 
of frequencies by referring to the most significant ones, but at the same time keeps different time series comparable, as all 
frequencies are retained that were needed to preserve an a-priori defined level of information for any of the time series. 
1. INTRODUCTION 
To investigate the state and/or the amount of vegetation is one 
of the main objective in the field of land surface related 
remote sensing applications. A prerequisite for successful 
monitoring of vegetation cover is the availability of frequent 
data that are internally consistent over a sufficient period and 
that provide information on the spatial complexity as well as 
on the temporal dynamics of vegetation. Many methods and in 
particular various vegetation indexes have been introduced, to 
quantify certain vegetation parameters. All of them take into 
account that vivid green vegetation shows a specific reflection 
signal in the red and near infrared part of the electromagnetic 
spectrum. The normalised difference vegetation index (NDVI) 
has become a commonly used index that is routinely derived 
from NOAA AVHRR images since mid 1981. To reduce 
atmospheric effects and noise, present in the direct reflectance 
measurements of an individual image, considerable effort has 
gone into the generation of multi-day composites. Such 
vegetation index composites proved to be very sensitive to a 
wide range of biophysical parameters, among them 
photosynthetically active biomass (Goetz et. al., 1999) or the 
presence of green vegetation (Myeni et al., 1995). 
Numerous studies have been conducted that use AVHRR 
NDVI data to analyse vegetation parameters on a regional to 
global scale, among them the estimation of terrestrial net 
primary production (npp) (Ruimy et. al., 1994) or the analysis 
of changes in vegetation phenology (Heumann et. al., 2007). 
The long term NDVI time series from AVHRR were related to 
climate variables such as air temperature or rainfall data with 
the objective of revealing geo-biophysical linkages for 
observed changes in vegetation parameters (greenness or npp) 
(Herrmann et. al., 2005, Xiao & Moody, 2005). 
The regional focus of this paper is the Niger Inland Delta, 
situated in the western Sahel region in Africa (see. Figure 1. 
for details). Whereas precipitation is the main constraint for 
vegetation growth in the semi arid Sahel, the Inland Delta's 
biosphere relies on water that flows in the region during the 
annual flooding period. This paper aims to analyse the 
long-term dynamics of vegetation cover in the Niger 
Floodplain over a 25 year period, based on 15day-composites 
of NDVI values from NOAA-AVHRR. To detect influences 
on vegetation cover for different time scales each time series 
was decomposed into 3 components according to the 
conventional component model. For this unbundling each time 
series was transferred into the frequency domain by a Discrete 
Fourier Transformation (DFT), making use of the advantages 
of the globally addressed (in terms of “the entire time series”) 
operators of the frequency domain. 
2. GEOGRAPHIC PARAMETERS FOR THE NIGER 
INLAND DELTA 
The geographic term „Niger Inland Delta“ stands for a vast, 
extremely flat area of some 10.000 km 2 extend, which is 
annually inundated by the water of the Niger - Bani river 
system during September to December. The ecology of the 
delta can be described as a mosaic of permanently, 
periodically and episodically flooded pat-tern, which contrasts 
sharply to the semi-arid environment of the Sahel. Spatial and 
temporal extent of the flood patterns vary due to fluctuating 
water supply by the river system caused by irregular rainfall 
in the catchment areas. Thanks to a comparatively good 
availability of (surface) water, the Niger Inland Ecosystem 
serves as stop-over for many migrating birds and other 
wildlife species as well as economic base for farmer and 
pastoral people. To foster the sustainable usage of its natural 
resources and to protect this natural heritage, the entire Niger 
Inland Delta became RAMSAR site in 2004 (RAMSAR 2008). 
(see Fig. 1 for an overview of the area) 
In contrast to its semi-arid environment, the Niger Inland 
Delta’s ecology can be described by a mosaic of permanently, 
periodically and episodically flooded areas. Their extent 
varies both in scale and in time due to irregularities of amount