Achieving land capability equivalent to that which existed prior to disturbance is the primary goal of reclamation in the Athabasca Oil Sands Region of northern Alberta. To date, most reclamation has focused on the re-creation of upland forest ecosystem analogues. However, a few wetlands have also been constructed. Additionally, wetlands have appeared spontaneously on landforms reclaimed to an upland forest type. Classifying and quantifying these opportunistic wetlands is an important consideration relative to oil sands closure and reclamation planning. Here we describe an approach using topographic and spectral variables to train a machine learning model (random forest) to detect and classify wetlands as an alternative to on-screen visual delineation. The aim was to develop a model that not only predicts where wetlands occur on reclaimed landforms but that is sensitive enough to classify them as to wetland form. Two random forest models were developed that predicted wetland occurrence at two levels: (1) wetland vs. non-wetland (to generate a prediction of all wet areas on reclaimed landforms); and (2) wetland class (with specific emphasis on marsh and shallow open water wetland classes). In addition to successfully predicting wetland occurrence, the resulting models handled the variability in reclamation approach, substrate type, and soil placement depth with high accuracy. This work confirmed that ~ 18% (211 ha) of the upland-reclaimed area at Suncor Energy’s Base Plant north of Fort McMurray, Alberta develops not to upland but to unintentional wetland, consistent with earlier studies. The ability to predict wetlands on the landscape could be invaluable when considering metrics of success associated with landscape reclamation in the Athabasca Oil Sands Region and for informing future inquiries around wetland persistence, resilience, and spatial connectivity through time on reclaimed landscapes.