Accurate mapping of streams that maintain surface flow during annual baseflow periods in mountain headwater streams is important for informing water availability for human consumption and is a fundamental determinant of in-channel conditions for stream-dwelling organisms. Yet accurate mapping that captures local spatial variability and associated local controls on surface flow presence is limited. An empirical random-forest model was developed to predict streamflow permanence (late summer surface-flow presence) for Mount Rainier National Park and the surrounding mountainous area in western Washington, USA. This model was developed to improve upon the existing multi-state, regional-scale probability of stream permanence developed for the greater Pacific Northwest Region (PROSPER_PNW). The model was trained on 544 wet/dry observations collected during the late summer, baseflow period from 2018 to 2020 using the crowd-source mobile application, FLOwPER. Final model accuracy was 0.74 with drainage area and covariates describing geology, topography, and land cover as top predictors of streamflow permanence compared to coarser resolution climatic covariates. The prevalence of static covariates over climatic covariates as top ranked important covariates highlights the importance of scale when evaluating controls on streamflow permanence. Cross validation of the model indicates that streamflow permanence probabilities from this model is an improvement over the regional-scale PROSPER_PNW model demonstrating the utility of relatively simple, crowd-sourced data to address water resource needs, and that determination of important predictors of streamflow permanence is influenced by the spatial and temporal resolution of analysis.