2004
Yould
their
pical
DAR
pints.
a are
ature
ass
nuch
an is
their
icing
tion
ature
| the
iting
area
rrain
also
0ses,
same
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
The set of rules defines how the measurements are applied when
constructing a terrain representation (such as a DTM) as a result
of spatial query. These rules 1) define the role of each
measurement, whether a measurement should be added as a
mass elevation point, a break line, a replacement polygon, or a
clipping polygon, and so forth; 2) specify, for each
measurement, where the elevation information comes from,
whether the Z coordinate, the value of a particular attribute, or a
result interpolated from a given DTM (which may include the
one under construction); and 3) specify at what resolution, or
resolution range, a measurement should participate. This is
necessary in order to support multi-resolution queries and
accommodate different requirements in terrain generalization
(to be discussed in section 3.2).
The introduction of Terrain allows a geo-database to store and
manage terrain data without being bound to a particular type of
DTM. A DTM can be generated dynamically upon users'
request by applying the rules. The idea of storing measurement
data (rather than explicit DTMs) in a database is similar to the
philosophy of /mplicit TIN (Kidner et al., 2000).
2.4 The Basic Requirements
Terrain provides a meaningful and comprehensive entity
through which users manage, query, and apply their terrain
relief information. Such an entity is referred to in ArcGIS as
terrain dataset.
In order to support various applications, a number of basic
requirements have been identified that Terrain should support. It
is also assumed that a DTM is still the most favorable structure
for surface visualization and analysis in GIS. Therefore, a
spatial query on a Terrain is expected to result in a DTM, upon
which various analyses can be performed. These requirements
include:
e Support a large area extent and a large amount of data
e Support point, line, and area data
e Support update, on both measurements and rules
e Allow certain measurements to be included/excluded
in a spatial query
e Support TIN and GRID DTM output
e Support spatial query with respect to a given area of
interest and vertical resolution. This would require
Terrain to dynamically generate DTMs of given
resolutions, anywhere within the extent of the Terrain
e Support “horizontal” multi-resolution query — a
special kind of query that specifies different vertical
resolutions for different parts of a given area. The
query will result in a multi-resolution DTM in which
vertical resolution varies across the whole area
3. KEY TECHNICAL ISSUES IN SYSTEM
DEVELOPMENT
To implement the supports listed in section 2.4, three key
technical issues need to be addressed: 1) tiling, 2) vertical
indexing and DTM generalization, and 3) data updating.
Although today's good geo-databases are capable of handling
large amounts of data, and 2D spatial indexing is basically a
built-in feature, extra arrangements are still necessary in order
to support fast DTM creation, multi-resolution queries, and
DTM generalization. These include internally arranging
measurement data into /iles according to data extent, density,
and hardware/software constraints; and introducing extra
indexing in the vertical dimension.
3.1 Tiling
Because data are potentially huge, it is not feasible to handle all
data at once. Not only is memory a problem, but performance
can be unacceptably poor. Tiling, on the other hand, can be a
good “divide and conquer” approach for handling data of large
extent, given the constraints of today's available technology. A
good tiling scheme can result in spatial coherence — data are
organized and stored according to their spatial proximity, thus
increasing the performance in spatial query and data transfer. It
also provides a powerful (2D indexing) mechanism for fast
searching; allows data to be handled in a more manageable
form; allows memory and CPU intensive tasks to be performed
locally without paralyzing the system; and is essential for DTM
generalization (to be discussed later).
Tiling divides a large geographic area into smaller, more
manageable, units (Figure 3), which can have different forms.
This approach uses a regular rectangle tile for its simplicity and
efficiency in computation. Choosing a proper tile size is a bit
more complicated. It depends on data density, CPU speed,
available memory, and other considerations. Basically, the size
must not be too big, so that a full resolution DTM of any tile
can be generated using an acceptable amount of system
resources.
- Terrain
Figure 3: Divide large terrain extent into regular tiles.
Tiles are used as the basis for reorganizing points. Points falling
into the same tile can be grouped into, and stored and handled
as, one single entity — a so called multi-point. This will reduce
storage space, increase spatial coherence and access speed, and
reduce disk I/O and network traffic. As most of the terrain data
will be mass points coming from remote sensing and laser
scanning, the benefit of this process can be significant. Line and
area features may require extra work in order to benefit from
this measure, as a single feature can cross many tiles.
3.2 Vertical Indexing and DTM Generalization
The vertical indexing is introduced to quickly identify those
data that contribute to a certain given resolution, so that no
redundant data will be retrieved and used to generate the output
DTM. This is achieved by assigning points to different “layers”
according to certain rules — a preprocessing step executed when
a terrain dataset is created.
Vertical indexing identifies points, and vertices of line and area
features, that contribute to a DTM of a specific vertical
resolution, or a specific layer in a DTM Pyramid (Floriani,
1995). A DTM Pyramid is composed of a list of pyramid layers,
with the first layer corresponding to the full resolution DTM,
283
