Classification of ground points is a critical step in producing digital elevation models of the Earth’s surface for studying landscape processes and geomorphic or ecological change. This paper describes a new algorithm for ground point classification and assesses the relative accuracy of the resulting ground surface in filtered light detection and ranging (lidar) and structure-from-motion (SFM) datasets. This color-enhanced multiscale curvature classification algorithm (MCCRGB) extends a popular lidar classification method (MCC) by introducing classification updates that distinguish vegetation and ground points by color. Multispectral lidar and SFM data imaging a subalpine volcanic tree kill are used to evaluate both methods. We find that color-based classification updates remove tree fall, low canopy, and brush, often requiring fewer iterations on large, ultra-high-density datasets. SFM data capture rills, small channels, and tree fall not visible in the lidar data. “Bare-earth” datasets from each method are internally consistent (mean vertical differences: -0.01 to 0.08 m) and validation at a set of 165 checkpoints shows a mean vertical difference of 0.46 m (standard deviation: 2.21 m) with the SFM ground points. The methods produce consistent topographic derivatives from each data source, including digital elevation models, slope, and profile curvature. While SFM derivatives are more variable and less continuous, the filtered products may be useful for geomorphic mapping and analysis, including mapping of microtopography or measuring landscape change in challenging, forested settings.