Intermittent rivers and ephemeral streams (IRES) are the most prevalent types of waterways worldwide, and their occurrence is expected to increase due to climate and land-use changes. Despite their importance, IRES remain relatively understudied, especially in alpine environments. Accurate mapping is crucial for understanding their unique characteristics, and remote sensing has proven to be a valuable tool for this purpose. This study presents a methodology for surface water detection using a Random Forest model applied to high-resolution aerial imagery provided by the Swiss Federal Office of Topography (swisstopo). We investigated the potential of these images for detecting surface water and evaluated the model's performance along with the contribution of multispectral and geomorphological predictors. The methodology was tested in the Valsorey catchment, an alpine basin located in Valais, Switzerland. Two models were developed, depending on the resolution of the input imagery. Both models demonstrated strong performance and interpretability, achieving accuracies of 85% and 87% and offer advantages over traditional single index-based methods. The practical applications of the models were explored through two approaches: mapping channel wetness probability onto an existing stream network and extracting the river network directly from the model predictions. The first approach was successful, allowing for the analysis of stream network dynamics and the mapping of IRES, while the second showed potential but requires further refinement. Key limitations include misclassification due to shadow and snow, presence of noise, and the limited temporal availability of the aerial images. Future research could address these challenges by improving the model or incorporating additional data types, such as hyperspectral unmanned aerial vehicles (UAVs), which would enhance temporal resolution and allow for the use of additional spectral bands.