As open source data becomes more ubiquitous, the involvement of citizen scientists has increased. The collection of large quantities of relevant data and respective labels through crowdsourcing on online platforms has yielded many exciting opportunities for machine learning applications. In geomorphology, multitemporal imagery, much of which is captured through crowdsourcing, is especially useful for training deep learning models for change detection in landscapes. This is relevant in terms of natural hazards that occur, including endogenous types like volcanoes and neotectonics, exogenous ones such as floods, karst collapses, sedimentation, erosion, tsunamis, and avalanches, as well as climate change or land use-induced hazards like permafrost and desertification. However, a challenge when harnessing crowdsourced imagery is the disorganized and “unclean” fashion in which it often presents itself. Cleaning data prior to training neural network-based computer vision models is key to success in any geomorphology change detection research. We discuss approaches such as manual techniques, image restoration and denoising, and image duplication reduction. The goal is to assimilate a diverse range of data collected from many sources to successfully train machine learning algorithms. In a broader sense, this research has the potential to save lives by detecting possibly destructive and dangerous geomorphological change, and to conserve environments that have been affected severely.