Stratifying land use/land cover for spatial analysis of disease ecology and risk: an example using object-based...

Area of Science:

• Environmental science
• Epidemiology
• Remote sensing

Background:

• Landscape epidemiology utilizes remotely-sensed land cover data for disease dynamic studies.
• Automated classification methods are crucial for processing this data.

Purpose of the Study:

• To demonstrate automated classification for stratifying land cover in infectious disease research.
• To compare object-based versus per-pixel classification methods for accuracy.

Main Methods:

• Used a Landsat ETM+ image for land cover classification in Paraguay.
• Applied both per-pixel and object-based classification algorithms for eight land cover classes.
• Developed a classification scheme for both land cover and land use data.

Main Results:

• Object-based classification achieved 84% overall accuracy, significantly outperforming the per-pixel method (43%).
• Producer's and user's accuracies were higher for all classes with the object-based approach.
• The Kappa statistic was significantly higher for object-based classification.

Conclusions:

• Object-based image analysis is superior to per-pixel methods for epidemiological remote sensing.
• Incorporating information beyond the spectral domain enhances classification accuracy.
• Object-based techniques are vital for accurate landscape epidemiology studies.