In: Wagner W„ Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
In: Wagr
76
2003), the LCS grid system is a congruent, unaligned discrete
global grid system; cell data is provided by remapping of pre
classified images, that are then stored as raster format in the
Tiles archive; the target grid level for each sensor depends on
the Ground Sampling Distance (GSD) of the bands used for its
classification.
Since each Tile contains the same amount of pixels, the surface
covered by the Tile varies with the sampling rate; depending on
the sensor, precisely on its spatial resolution, that pixel size
shall be under a threshold value, smaller than half of the
original image pixel size in order to allow the pixel-based
analysis and mitigate the displacement among the overlaid
images due to systematic error into the original geo-referencing,
hence defining the best fit level for that sensor. Reference grid
resolution (Tile pixel resolution), across longitude, is computed
at equator, taking into account that, for computational and
archiving optimization reasons, the sampling rate is kept at a
power of 2.
Grid
Ref.
Pixel
Samples
Supported sensors
Leve
1
GSD
res.
per deg.
m
m
#
0
1000
434,84
256
(A)ATSR, MODIS
1
500
217,42
512
MODIS HKM
2
250
108,71
1024
MODIS QKM,
MERIS
3
125
54,36
2048
LANDSAT TM TIR
4
60
27.18
4096
LANDSAT ETM+
TIR
5
30
13,59
8192
LANDSAT
TM/ETM+ MS
6
15
6,79
16384
LANDSAT ETM+
PAN SPOT5,
AVNIR-2
7
7
3,40
32768
-
8
3,5
1,70
65536
VHR MS
9
1,75
0,84
131072
VHR MS
10
0,8
0,42
262144
VHR PAN
Table 1. Grid parameters and supported sensor for levels 0-10
Table 1 lists grid parameters for levels zero to ten with
reference GSD and name for a selected series of sensors. To
avoid data loss during sampling, pixel threshold has been set at
about half the original pixel size for medium resolution sensors
at level zero, while for higher resolution the most suitable grid
level is selected considering the power of 2, nearest to half of
the original image resolution.
Tile mapping is the process dedicated to ingest the pre
classified scenes and remap them onto Earth Fixed Grid Tiles.
The ingested scene has its original geo-reference system, thus,
before applying any remapping process, a geo-referencing pre-
process is required in order to avoid co-registration problems
among images. It is then assumed that each input image to the
ALCS system is accurately geo-referenced with accuracy below
half of pixel size.
In the Tile mapping process, the original data is filtered taking
into account the scope of LCS: tiles over sea are filtered out on
the basis of a 4 valid pixels threshold (minimum amount of
pixels detectable at supported sensor resolution) using the U. S.
Geological Survey 1 Km Land Sea Mask dataset (Eidenshink et
al., 1994) for a per pixel coordinates test to assign each tile
pixel to either the land or sea classes. Moreover all cloud and
outlier pixels classes are also removed. Tile mapping is
performed using the Nearest Neighbour remapping
methodology.
2.3 Land Cover Evolution Models
LCS evolution model matching is a form of change analysis
that, according to (Lu et al., 2004), falls in the “classification”
category and especially it is a form, or extension, of the
commonly used Post-Classification Comparison. The system
lays on the basis that the evolution of land cover classification
over time can lead to identification of land use typologies, and
also effective identification of areas of rapid land-use and land-
cover variation may allow contextual detection of major
disturbances such as fires, insects and flooding. The key to the
identification of relevant evolution patterns is the definition of a
corresponding evolution model that can be systematically used
to determine if an observed data series conforms to the model
pattern.
2.3.1 LCS evolution model: is defined as a sequence of
expected sets of land cover classes along the temporal line, each
land cover set - temporal reference pair constitutes an evolution
model element (see Figure 1 for a schematic representation).
Transitions can be represented by pairing elements which
define expected land cover configuration in given points of the
time line comprising the transition. A series of evolution model
elements defines a land cover evolution pattern that, stored in
computer readable form, can be automatically matched with
actual land cover time series to search for the modelled feature
evolution pattern (evolution model matching). Relevant
metadata associated to evolution models is: model name, model
type (feature category), area of applicability, applicability to
grid levels (accounts for typical size / resolution requirements
of the modelled feature).
■RfeftnHion
HPT CimracterisMc*
Non-permanent
• Winter: Bare soil or low Veg.
crop fields (with
annual cycle)
PERIODIC
* Spring and early summer: Increase
in vegetation
• From mid-summer to winter: Bare
soil
J Lc.V'eo I High Vf 0.
f D.re So,I I V L ™ I
Winter Spring
Summer Fall
Candidate
• Mid-to-high vegetation at TO
• Low vegetation/bare soil at T1
•T1 =T0 + few days
|xw ysjs»Soiityp9 J
Any Bare Soil Type J
TRANSITION
TO
T1
Figure 1 - Description of évolution models in the land clover /
time domain for a periodic phenomenon and for a
transition (non-periodic) phenomenon
The land cover class value defines the expected land cover type
in a given model element. Multiple land cover class values can
be defined, this is obtained by setting the class tolerance
parameter in an evolution model element. An element can also
have the “Not” flag enabled to invert the expected land cover
class values.
Each element has a temporal reference that defines a point
along the timeline, either fixed or related to the previous
element, where its land cover class is expected (the data
sampling point of the model matching); in particular, three
temporal parameters are provided: Date, Time Since Previous
element (TSP) and time tolerance. All parameters are specified
in unit of days, and thus the entire system is designed to work at
day resolution.
The Date
element 1
of an eve
model. C
different
specified
any poin
complete
matched
without t
at any s
setting th
TSP defi
days afte
sequence
which it
the time!
apply the
Time To]
interval,
timeline
allows c
element’s
gaps due
evolution
expected
entire Tir
2.3.2 I
automata
of interes
Date spei
periods (
descriptic
periods, i
each one
Taking ir
the vario
model m
only at a
performei
simply pi
pixels in
that com]
the expec
In particu
exact day
data is m
day in bt
search int
the eleme
closest tc
matching
Model m:
mapped (
visual ana
