|
You are at: Wagner
Home > Technologies >
Environmental Data Fusion
Environmental Data Fusion
Our Environmental Data Fusion algorithms/tools
address many of the primary challenges associated with environmental and
other geostatistical data fusion.
·
They automatically access data
from multiple, remote sources by providing a system architecture that
allows any unit at any location to look for environmental or other
important data across a general network, and to pull that data into a
common database.
·
They automatically resolve
conflicts in returned data by merging or fusing this data into a single
database and comparing values over the spatial domain, identifying regions
of agreement/disagreement and selecting the “Best” data for current needs.
·
They automatically receive and
assimilate new measurements as they relate to the existing knowledge
base. This is particularly important since a single measurement has a
known validity that is generally better than that of historical (averaged
over a long period of time) data, but the single measurement only provides
a micro-view of the space, in fact only the view at a single point.
·
They automatically recommend
optimal points at which to obtain additional measurements.
· They
automatically tie environmental and other geospatial data to operational
needs. Even perfect knowledge of geospatial factors is only useful if it
can be applied to an operationally useful problem.
Our Environmental Data Fusion Tool Set contains the
following four primary tools:
1.
Information Collection Tool
2.
Polygon Combination Tool
3.
Measurement Fusion and Optimization Tool
4.
Data Analysis Tool to Relate Geospatial Data to Operational Needs
Information Collection Tool
Figure 1. Information Collection and Data
Fusion/Analysis/Mining Architecture
The most general way to think
of this tool is to envision any application (Mine Warfare or not) requesting
data from a generic data server or broker for multiple servers. The
application need not specify the specific location of the server, but should
only ask for the kind of data it needs, the geographic location associated
with that data, and other supporting information if applicable (for example,
unique definitions of “Best”). The broker then returns to the calling
application the “Best” data available to support its request.
The second step is to provide
a general broker/agent architecture that supports the access and retrieval
of environmental data from a variety of sites. In our architecture, the
broker manager takes the request from the calling application and accesses
the Data Agent, which finds all the data from all of its source repositories
and provides this data to the broker. To minimize the bandwidth associated
with these requests, and to accommodate interruptions in service between the
servers, the local system stores a cache copy of each database. Then, when
the broker makes a request, the data agent does a query for new data across
the net. If none exists, no data is pulled. Locally collected data can
also be stored, and this data is then in turn available to be pulled from
other sites. The architecture also has a Graphical User Interface (GUI) for
the definition and management of the server connections and the sites for
data sources. It should also be noted that this architecture supports data
compression, and can support information exchange on bandwidth-limited
networks, over RF networks, and over intermittent networks.
Once all the available data of
a given type and for a given location is in the local broker’s database, the
next step is to combine all of this information into a “Best” picture. We
do this using a software tool called mcmm_attribute that performs a polygon
combination of all the available data sources and then applies rules to
those polygons to determine the “Best” available view. In our design, the
rule-set for defining “Best” can come from the application or through a
GUI. Examples of these polygon combinations are provided in the following
discussion on the Polygon Combination Tool.
This approach allows us to get
the best data available, but we still need tools and techniques for the
assimilation of new measurements into the common environmental picture as
described in the Measurement Fusion Tool section. This capability is
outside the broker, but connected to the requesting application through the
common software interface. This way, the assimilation tools are always
operating on the “Best” data for their task. Note that the data can then be
inserted into both the local and “Best” databases. Finally, this system
description can be thought of as just a single node of a distributed
system. This provides a very powerful tool to support geospatial data
fusion.
Polygon Combination Tool
Figure 2. Polygon Combination Tool Example
This
example shows how the polygon combination tool can fuse data from multiple
sources, as well as data with multiple attributes and multiple sources.
To
illustrate the fusion of data from multiple sources, we have three sources
of bathymetry data with certain regions of validity associated with those
measurements, shown in the left side of Figure 2, with arrows pointing to
“Best” fused result. The least trusted, but most available data is the NIMA
digital chart. The next most trusted is the gridded data from the Precise
Underwater Mapping Analysis (PUMA) system. Finally, a single sounding from
the U.S. Naval Oceanographic Office with a certain radius of validity is the
most trusted entry. As shown in Figure 2, the polygon combination tool
fuses these data using the predefined “Best” order to produce a “Best” fused
geospatial picture.
This
technique also supports the combination of multiple attributes from multiple
sources. For example, as shown on the right side of Figure 2, we can take
Bathymetry and Bottom sediments (assuming that the “Best” values for each of
these have already been established) and then combine them. The system then
creates the fused single picture shown in the example. Note that after
fusion there are seven different regions with six different data
combinations.
Measurement
Fusion and Optimization Tool
Figure 3. Measurement Fusion and Optimization Tool Example
(Contour Plot of Absolute Value of Estimation Errors)
Our next tool is a fully automated measurement
fusio n and optimization tool for assimilating new measurements into an
existing database. This tool is based on mature statistical techniques that
work well for the bathymetry and bottom sediments that are critical to mine
warfare operations, can produce improved results over standard spatial
interpolation and prediction methods, and can be applied to any single
valued (or multi-valued) geospatial data. Also, these techniques require no
distribution assumptions, accurately estimate the variance of the prediction
error, and recommend optimal points at which to obtain additional
measurements.
When performing measurement fusion, we first
use a data set typical of the area of interest to determine how the
measurements are correlated, and to determine trends in the data. These
correlation statistics can be anisotropic, which allows us to model the fact
that water depth, for example, is usually relatively constant as one moves
parallel to a coastline, but can change quickly as one moves perpendicular
to the coastline.
After determining these correlation
statistics, a relatively small sample of data from the area of interest can
then be used directly in the measurement fusion process to estimate data
values throughout the area of interest.
This sample display demonstrates how selected measurements can be
combined with existing environmental data to predict a geographic extension
of those measurements. In the example shown in Figure 3, the 32 white
circles show the small random sample of measured data that was used in the
measurement fusion process.
For this example, we started with 3233 bathymetric soundings and
randomly chose 10% of these points. We then developed the correlation
functions using this random sample of 323 measured points. Once the
correlation functions were calculated, we then selected 32 random soundings
to represent the new data measurements. These 32 new points were combined
by the measurement fusion tool with the correlation functions to estimate
the bathymetry for the entire area. The results of this combination are
shown in Figure 3. The 32 white circles represent the randomly selected
“measurements,” and the colors in the geoplot show the difference between
the bathymetry values calculated by our measurement fusion tool and the 3233
actual soundings. As can be seen, the majority of the errors were less than
one meter (dark blue), although some were as high as 9 meters. It is also
evident that the randomly selected measurement set was clustered to the
north end of the sample, and that the majority of the larger errors were at
the southern end.
This sampling process models a real-world operation in which we would
have a few hundred measured points from an area similar to the area of
interest that are used to estimate the correlation functions, and a few tens
of points in the area of interest that are available to be used in the
actual measurement fusion process.
Data Analysis Tool to Relate
Geospatial Data to Operational Needs
 
Figure 4. Sample Results for Mine-Countermeasures (MCM)
Tactics
The fourth challenge is to tie the environment or
other geospatial data to mission/operational success. In our system, the
sensor effectiveness for a given countermine system (in this case a generic
side scan sonar) is related to the environmental factors that drive its
effectiveness. The resulting polygons are then displayed on the geographic
chart and used for mine-countermeasures planning.
In the
example above this effectiveness is shown using a colored “stoplight” scale,
with the light green showing areas in which mine-countermeasures operations
can be conducted more effectively, and the dark red showing areas in which
mine-countermeasures operations can be conducted less effectively. This
approach allows plans to be developed that place more countermine effort in
the red areas than in the green areas. This is significant since, without
this type of capability, all areas are generally treated as having the same
mine-countermeasures operational effectiveness, usually the worst, which
requires a significant amount of additional time to complete the assigned
task.
|
