A Bayesian Modeling Approach for Improved Land Use Classification
Abstract
Low-density land use is a threat to ecological connectivity and farmland security in the American mountain West. This type of land use has shown explosive growth in the last several decades but has proven difficult to map as it lacks a physically based definition and is unsuitable for classification at the pixel level. To improve understanding and quantification of low-density land use, we utilize two regionally specific typologies, and an object-oriented, remote sensing approach to classification. Implementing a two-part methodology, we 1) better explain the drivers and account for spatial variation in low-density subclasses using a spatial generalized linear model, and 2) within a Bayesian framework, assess the predictive ability of remotely sensed data in differentiating between subclasses. This work draws upon expert knowledge of the subject matter and region, a novel application of an object-oriented classification approach, and the use of a Bayesian classifier in conditions well-suited to enhance accuracy. While results are generalizable to a larger region, this modeling approach represents a potential framework for all types of land use classification, is a solid foundation for land use change detection, and is the next step in harnessing satellite-based information to create continuously updated land use maps.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMGC31N1384U
- Keywords:
-
- 6309 Decision making under uncertainty;
- POLICY SCIENCES & PUBLIC ISSUES