Historical maps inform landform cognition in machine learning.

Landforms are difficult to delineate in the field or on maps because of inherently indeterminate and fiat boundaries (Smith and Mark, 2003). Research and applications of landform delineation have progressed along two paths in the geosciences. General geomorphometry is a continuous field-rooted approach that focuses on computing localized parametric values for mapping land surface shape patterns and deriving land segments or elements that can be considered homogeneous at the chosen scale of analysis for a particular application. Specific geomorphometry is an object-oriented approach that applies the geomorphometric parameters to identify, delimit, characterize, and classify individual landform objects. There is no standard list of landform types or methods for mapping individual landforms because landform cognition is influenced substantially by people's cultural, linguistic, and individual backgrounds. Thus, delimiting and classifying individual landforms requires a multidisciplinary approach by incorporating geomorphometry, geomorphology, cognitive science, cartography, remote sensing, and geographic information science knowledge. Cartography plays a unique dual role in named landform representation by providing contextual information as input to the demarcation problem while providing a medium for expressing the human cognization of landforms. Hence, this research aims to improve the automated mapping of named landforms. [ABSTRACT FROM AUTHOR]