With the accelerating glacier retreat occurring across the globe, large areas of rapidly changing proglacial terrain are emerging. High-resolution drone imagery has made it easier than ever before to produce high-accuracy geomorphological maps of these proglacial landscapes, though the abundance of data has necessitated an investigation into the best semi-automated approach to mapping with this imagery. Algorithms are required for these tasks as they are 1) capable of fully utilizing all data collected by these platforms; 2) produce maps rapidly; and 3) make data-driven decisions providing much needed objectivity and consistency to the mapping process. More objective, detailed, and frequent geomorphological maps of proglacial environments are crucial to understanding the landscape and its evolution, which in turn is imperative for both accurate interpretation of previous glacier behaviour and inferring future change. Object-based (GEOBIA) and pixel-based image analysis approaches were used to train random forest algorithms and classify resampled drone imagery of a recently deglaciated environment in Svalbard. The pixel-based approach performed well at all resampled resolutions (1-meter, 2-meter, and 5-meter pixel size), while the GEOBIA approach had a significantly lower accuracy across all resampled resolutions. Strong pixel-based performance at each resolution suggests coarser imagery may be adequate for accurate mapping, especially for larger-scale landforms. This research provides useful insights in guiding future mapping campaigns in dynamic ice-free environments with respect to modelling approach and potentially imagery collection.