The Field Geomorphologist in a Time of Artificial Intelligence and Machine Learning
Chris Houser, Jacob Lehner, Alex Smith
Abstract
An increasing number of papers incorporate machine learning (ML) approaches to analyze spatially and temporally rich data sets in geomorphology. These data-driven approaches have the potential to significantly improve our understanding of complex systems across a range of scales and support the development of new theories of landform and landscape development that can eventually be incorporated into predictive models. Coupled with the growing availability of remotely sensed data, geomorphology could move further toward a desk-based science and erosion of the field tradition. Using examples from coastal geomorphology, this review of ML applications argues that the development of models that are scalable and can be translated between sites is dependent on experience in the field. Although ML models are shown to be effective as a surrogate to process-based numerical models, they are only as good as our conceptual understanding of landform and landscape form and evolution. This means that ML is simply a new and powerful tool in the proverbial belt of the geomorphologist and should not come at the expense of the field tradition that informs us of whether ML results are accurate, transferable, and scalable.