S, Vol. XXXVIII, Part 7B 
In: Wagner W„ Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
A 
STA 
mi, natali) @meeo.it 
an.beccati@unife.it 
le, Italy 
:aly 
:tem 
demanding in terms of 
use/land cover change 
j pre-processing of the 
megabytes in size and 
here proposed foresees 
y assigning a semantic 
ips can be expressed as 
itures - time space, that 
des fully automatic and 
ic calibration, accurate 
te characteristics of the 
d Classification System 
ional ALOS-AVNIR-2 
ssing systems shall be 
on and reduce the data 
llite data usability for 
-processing operations 
datasets homogeneous: 
steps are accurate 
; calibration; digital 
ce conversion must be 
multi-temporal images 
image registration are 
since each one has its 
rs. 
ed Land Classification 
inced applications for 
g easy-to-use tools for 
on analysis over time 
eadily usable by the 
;ers of land cover and 
ing a computer aided 
/sis tool for definition 
ind an high degree of 
ion in an effort to ease 
nd land cover change 
:h other tools to find 
uencing life on Earth 
In the following sections the methodological generic approach 
implemented on LCS is described, with specific emphasis on 
the critical subsystems. Moreover, a prototype application of 
LCS being implemented in the framework of the European 
Space Agency (ESA) Support by Pre-classification to specific 
Applications (SPA) is described; finally the preliminary results 
obtained during the SPA Project are discussed, with the clear 
aim of demonstrating the validity of the approach. 
1.1 Related work 
Many approaches and methodologies exist for land cover 
change analysis: an extensive survey is provided in (Lu et al., 
2004). Similar work for multi temporal analysis systems has 
been performed to provide targeted land cover change studies or 
develop databases of land cover (Homer et al., 2004). An 
interesting bi-temporal approach to land cover change analysis 
is provided by the Land and Ecosystem Accounting (LEAC) 
methodology whose main goal is to provide an easy and 
comprehensive access to land cover data, showing the ‘stock’ 
available for each land cover class in the different land cover 
data, and providing also the changes occurred in the periods 
between different land cover works, as land cover flows matrix 
(Haines-Young and Weber, 2006). The LCS approach however 
is not focused on a particular change pattern, land use typology 
or phenomena, it is instead proposed a generalized approach to 
land cover change analysis that, building on top of land cover 
maps stock, might serve as an interactive framework and tool 
for scientists to help quickly verify hypothesis and improve 
their research activities. Lastly, the system targets also decision 
makers to provide a practical surveying tool to systematically 
provide fast response in detection of features of interests. 
2. METHODOLOGY 
The LCS system is a generic tool for long-term time series of 
satellite data management for application in the LULCC field. 
The term “generic” refers to the wide applicability of the 
system to different type of satellite-borne sensor data, 
permitting also multi-sensor applications, and to different time 
frames; moreover it refers to its ability to fully exploit the 
multitemporal database for different LULCC phenomena 
analysis. 
LCS creates time series of homogeneous satellite data making 
use of a robust land cover classification system named SOIL 
MAPPER®: this system process satellite data coming from 
different sensors in the same way, generating land cover 
classification maps with the same semantic meaning, thus 
permitting multi-temporal and multi-sensor applications. 
The stock of classification maps, are then mapped on a common 
reference grid to allow worldwide pixel based multi temporal 
analysis of land cover to be performed in relatively small 
amounts of time since data compression, obtained through 
semantic feature extraction, delivers a map stock within 6 
Terabytes, that is an amount of data readily manageable by 
modern computer systems. 
The core change detection feature of LCS are the land cover 
evolution models and its model matching engine: in LCS, an 
evolution model is defined as a sequence of expected land cover 
classes along the temporal line; each land cover class - 
temporal reference pair constitutes an evolution model element. 
Land cover transitions can be represented by pairing elements 
which define expected land cover configuration in given points 
of the time line. A series of evolution model elements defines a 
land cover evolution pattern that can be matched with actual 
land cover time series data to determine if that data matches the 
modelled evolution pattern. 
There is almost no automation in model definition and the 
model itself is designed to let the user precisely define each 
model element, also starting by a derived model from observer 
data, with tolerance margins in both feature and time domains. 
All the knowledge for multi temporal analysis is provided by 
domain experts in the form of evolution models. 
The LCS methodology, explained hereafter in its critical 
subsystems, foresees three main elements that, chained 
together, aim at providing a consistent system for multi 
temporal land cover data analysis: original data classification, 
Earth fixed reference system and land cover evolution 
modelling and matching. Moreover the layout of user interfaces 
suitable to ease analysis, define evolution models and perform 
automated model matching are described. Following 
subsections detail each main aspect of this methodology. 
2.1 Common land cover classification system 
SOIL MAPPER® is a fully automatic software that permits to 
generate land cover classification maps through the analysis of 
multispectral satellite data in the optical domain. 
As input it requires multispectral remotely sensed (RS) images 
calibrated on Top of Atmosphere (TOA) physical values: TOA 
Reflectance values for Visible (VIS), Near Infrared (NIR), 
Short Wave Infrared (SWIR), Mid-Wave Infrared (MIR) bands 
and brightness temperature (BT) for Thermal Infrared (TIR) 
bands (Mantovani et al. 2009). 
As output, it generates a preliminary classification map where 
each pixel is associated with one label belonging to a discrete 
set of spectral categories. Spectral classes detected by SOIL 
MAPPER® have a semantic meaning belonging to the following 
main categories: Vegetation, Bare soil / Built-up, Snow / Ice, 
Clouds / Smoke plumes, Water / Shadows, Outliers. 
SOIL MAPPER® actually supports most common satellite 
optical sensors (from medium to very high resolution), like: 
MODIS, AVHRR, AATSR, MERIS, Landsat 5 TM/7 ETM+, 
ASTER, SPOT-4 HRVIR, SPOT-5 HRG, IRS 1-C/-D, IRS-P6, 
IKONOS, ALOS/AVNIR-2, QuickBird. 
Recent developments to the system (MEEO, 2010) introduced 
an uniform classification output with similar number of 
semantic classes across sensors and standardised classification 
output that makes is suitable for LCS. 
2.2 Earth Fixed Grid reference 
LCS defines a multi level global Earth fixed reference on which 
all satellite data has to be remapped to perform multi-temporal 
sample-by-sample analysis. The multi-level mesh-grid has been 
set with a variab'e uniform angle sampling rate over the 
geographic coordinates system (Lat. Lon.) with level 0 set at 
1/256 th degree. Samples (grid elements) are grouped together in 
fixed size tiles of 64 by 64 samples called Tiles. At level 0 each 
Tile covers *4 of degree in both Latitude and Longitude; each 
further level doubles the sampling rate in both dimensions (i.e. 
1/512 th degree at level 1 and so on). According to (Sahr et al., 
75