Comparing geomorphological maps made manually and by deep learning

Geomorphological maps provide information on the relief, genesis and shape of earth's surface are widely used in sustainable spatial developments. The quality geomorphological is however rarely assessed or reported, which limits their applicability. Moreover, older often do not meet current requirements require updating. This updating time-consuming because its qualitative nature difficult to reproduce, but can be supported by novel computational methods. In this paper, we address these issues (1) quantifying uncertainty associated with manual mapping, (2) exploring use convolutional neural networks (CNNs) for semi-automated mapping (3) testing sensitivity CNNs uncertainties manually created evaluation data. We selected a test area Dutch push-moraine district pronounced relief high variety landforms. For developed five 27 automatically landform using CNNs. resulting similar regional level. could identify causes disagreement between local level, related differences experience, choices delineation different interpretations legend. Coordination efforts field validation necessary create accurate precise maps. perform well identifying landforms units, fail at correct delineation. human geomorphologist remains classification computed data that train evaluate have large effect model performance evaluation. also advocates coordinated ensure training Further development processing required before act as standalone techniques.