S, Vol. XXXVIII, Part 7B 
In: Wagner W„ Sz6kely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
77 
Moreover all cloud and 
)ved. Tile mapping is 
Neighbour remapping 
arm of change analysis 
s in the “classification” 
, or extension, of the 
amparison. The system 
and cover classification 
and use typologies, and 
apid land-use and land- 
al detection of major 
looding. The key to the 
ms is the definition of a 
be systematically used 
conforms to the model 
ined as a sequence of 
l the temporal line, each 
constitutes an evolution 
ematic representation), 
airing elements which 
a in given points of the 
ries of evolution model 
i pattern that, stored in 
latically matched with 
or the modelled feature 
matching). Relevant 
is: model name, model 
ibility, applicability to 
esolution requirements 
Graphic Evolution 
Winter Spring Summer Fall 
els in the land clover / 
thenomenon and for a 
emenon 
pected land cover type 
cover class values can 
g the class tolerance 
An element can also 
e expected land cover 
that defines a point 
ated to the previous 
s expected (the data 
); in particular, three 
Time Since Previous 
rameters are specified 
is designed to work at 
The Date parameter references a date in the timeline where the 
element lies; this parameter can be set only on the first element 
of an evolution model, thus indicating the starting point of the 
model. Different specification for the first element Date defines 
different kinds of models as follows: a blank Date (no date 
specified) defines a non periodic model, that can be matched at 
any point along the timeline as a sliding temporal window; a 
complete date specification defines a fixed model that can be 
matched only at its fixed temporal reference; a partial date (i.e. 
without the year) defines a seasonal model that can be matched 
at any subset of points in the temporal line identifiable by 
setting the unspecified parts of the date as matching input. 
TSP defines the sampling point of an element as the number of 
days after the previous element along the evolution model 
sequence, hence it is not applicable to the first element for 
which it is fixed to zero; Using TSP to locate elements along 
the timeline lets the user easily define non periodic models and 
apply them at any point in the timeline. 
Time Tolerance (TT): defines the radius in days of a temporal 
interval, centred on the element’s sampling point along the 
timeline in which the actual data can be validly sampled. That 
allows coping with the possibility of missing data at the 
element’s sampling point, such as cloudy acquisitions or data 
gaps due to satellite revisit time. Moreover, any element of the 
evolution model can be set to be “Persistent”, that means the 
expected land cover type must persist in actual data for the 
entire Time Tolerance of the element. 
2.3.2 Model matching algorithm: key feature of LCS is 
automated model matching that takes as its input a single area 
of interest and a variable amount of details, depending on the 
Date specification of the actual model, for one or more time 
periods over which matching is to be preformed. For this 
description we assume a single time period, in case of multiple 
periods, the process herein described is simply iterated over 
each one to deliver one result set for each period. 
Taking into account the definition of an evolution model and 
the various options for its parameters, the simplest form of 
model matching is the match of a fixed model, that is tested 
only at a fixed point along the temporal line. This matching is 
performed by testing, over each grid element (here called also 
simply pixel) covering the area of interest, the value of relevant 
pixels in the stock maps archive, according to every element 
that composes the evolution model, verifying each pixel with 
the expected value in the model. 
In particular, for each element, pixel data is first searched at the 
exact day of sampling specified by the element and, in case that 
data is missing the search interval is recursively extended one 
day in both directions to search for data within the TT. This 
search interval does not influence the marching outcome, unless 
the element is defined as persistent, that is any available data 
closest to the sampling point, within the TT, is used for 
matching with the expected set of classes for the element. 
Model matching can bring four different results for each pixel, 
mapped on a result map with different colors for immediate 
visual analysis, as follows: 
• No data (Black): for any element there is no data 
available within the classified tiles stock. 
• Match (Green): for all model elements data is 
available and, actual pixel value fits with the main 
land cover class expected value of the model. 
• Match within Tolerance (Yellow): for all model 
elements data is available and, for all observations not 
providing a Match result, the observed class is among 
the set of classes listed in the class tolerance set. 
• Not match (Red): for all model elements data is 
available and, for at least one element, actual data 
does not match neither the main class, nor any of the 
classes in the tolerance set. 
Seasonal models are matched in the same way as fixed models 
but any time range input detail for the model starting date can 
be freely specified to a full date, hence fixing the model. Each 
set of different values of the details specifying a full date (multi 
period matching), delivers its related result map. 
The most general form of evolution model matching, called 
Non Periodic Model Matching, is designed for automated 
detection of the broadest evolution patterns typologies, 
including unpredicted events like sudden deforestation / fires, 
flooding and other single or multi transition phenomena whose 
position in the temporal line cannot be pre determined by 
nature. These models are characterised by an empty Date 
specification on their fist element, hence matching these models 
require as input the full specification of star and end dates of 
each temporal range. The non periodic matching is then 
performed as a repetition of the fixed model matching above for 
each day in the time period, the model can thus be seen as 
sliding along the temporal line. The date of application (start of 
the sliding window) slides from the time period start date, to its 
end date minus the model duration. 
The match is tested for any day in the temporal range until a 
match is found; when a match is found, to avoid duplication of 
the same match that will be reported multiple, the next test is 
moved forward of an entire sliding window This non periodic 
matching can detect more than one match occurrence of the 
model in the given period if it reoccurs; to provide an 
immediate visual feedback over this reoccurrence, the first three 
occurrences of Match are marked with different tones of the 
result colour. Match result options have the same labels as the 
fixed matching but slightly different meaning as follows: 
• Match: is reported when at least one matching test 
returns a Match at any given date within the temporal 
period. Depending on the number of occurrences, this 
is marked with different tones of green. 
• Match within tolerance (Yellow): is reported when no 
Match is returned and at least one fixed match 
produces a Match with tolerance. 
• Not Match (Red): is reported when no matching 
produced a Match or a Match with Tolerance and at 
least one test produced a Not Match. 
• No Data (Black): is reported when all tests along the 
temporal interval return No Data. 
Coloured result maps are displayed by the system over a 
dynamic reference map for immediate visual analysis of the 
results but also a GeoTIFF version of the map is produced and a 
comma separated values file format has been designed for 
further results analysis with other software tools.