Classification of a stream as perennial (flowing year-round) or non-perennial (periodically going dry) has important regulatory implications with generally stronger protections for perennial steams. However, stream classifications in the US are surprisingly inaccurate, particularly for small headwater streams. The U.S. Geological Survey has developed models trained on existing and newly collected, crowd-sourced flow/no flow observations that provide updates of perennial/non-perennial classification for large, multi-state geographical extents. Important predictor variables are different depending on the geographic extent of the modeling domain. Unsurprisingly, for smaller modeling domains, local scale variables that exist at higher spatial resolution tend to be important relative to coarse resolution variables that are available over larger geographic extent. Importantly, these simple visual observations are dramatically more spatially extensive and substantially less costly than conventional streamflow data.