Comparing statistical and semantic approaches for identifying change from land cover datasets

Area of Science:

• Geospatial science
• Data integration
• Environmental modeling

Background:

• Spatial data often exhibit incompatibility due to differing classification systems, hindering integration despite similar data types or themes.
• This issue is prevalent in natural resources surveys, stemming from changes in survey methods, resource understanding, and policy.

Purpose of the Study:

• To develop and evaluate methods for integrating discordant spatial datasets.
• To address the widespread problem of integrating comparable but incompatible spatial data in natural resources surveys.

Main Methods:

• A statistical approach utilizing discriminant analysis.
• An expert-based approach leveraging domain knowledge.
• Application to land cover mapping in Great Britain to compare methods.

Main Results:

• The expert-based approach demonstrated superior identification of locations with incompatible classifications.
• The expert-based method achieved a significantly higher rate of change recognition between datasets.
• Statistical methods showed limitations in handling semantic discordance.

Conclusions:

• Expert-based integration is highly effective for discordant spatial data, particularly for change detection.
• Enhanced metadata reporting is essential, including data conceptualizations, semantics, ontologies, and the rationale behind feature definitions.
• Effective communication of data information is vital for realizing the full benefits of spatial data initiatives like GRID, E-science, and INSPIRE.