A Suggested Methodology for the Computation of Imagery-Based Terrain Data Reliability and Implementation into Tactical Decision Models
ENGINEER RESEARCH AND DEVELOPMENT CENTER CHAMPAIGN IL CONSTRUCTION ENGINEERING RESEARCH LAB
Pagination or Media Count:
The issue of data uncertainty and its impact on model output is discussed in this report under three topic areas. Modifiable methodologies for implementation of data reliability are provided for each topic. The first topic addresses the fact that new digital terrain data are created daily that directly supports tactical decision-making. It is common for decisions to be made without the full understanding of the contributing terrain data quality. Data are treated as spatially invariable in quality and devoid of any metric measuring the underlying certainty of feature classification. Commercial-off-the- shelf COTS image processing packages provide an opportunity to identify terrain classification reliability along with the class assignments. However, this opportunity for capturing reliability information is typically neither passed along into the final terrain class map output nor is it stored as a supplementary metadata file. Training sample contingency table scores are used to assign terrain feature class theme, or layer, reliability scores, and distance to class means computed using a Mahalanobis distance technique is used for assignment of within-terrain theme individual pixel reliability scores. For topic two, the value of reliability pixel scores being carried forward into a tactical decision-modeling environment is considered. Reliabil- ity from every pixel contributes to a final decision-model product reliability score at the pixel level. An example is provided. In topic three, terrain data layer contribution to model output for HLZ and BlV area tactical decision aids is evaluated on the basis of assessing importance or weights for individual or combined terrain layers. Kappa values designating terrain variable contribution are suggested as objective surrogate weights available for use in the HLZ and BIV modeling process.
- Administration and Management
- Computer Systems
- Military Forces and Organizations