A comparison of permafrost prediction models along a section of Trail Ridge Road, Rocky Mountain National Park, Colorado, USA

Abstract

The distribution of mountain permafrost along Trail Ridge Road (TRR) in Rocky Mountain National Park, Colorado, was modeled using ‘frost numbers’ and a ‘temperature of permafrost model’ (TTOP) in order to assess the accuracy of prediction models. The TTOP model is based on regional observations of air temperature and heat transfer functions involving vegetation, soil, and snow; whereas the frost number model is based on site-specific ratios of ground temperature measurements of frozen and thawed degree-days. Thirty HOBO© temperature data loggers were installed near the surface as well as at depth (30 to 85 cm). From mid-July 2008 to 2010, the mean annual soil temperature (MAST) for all surface sites was − 1.5 °C. Frost numbers averaged 0.56; TTOP averaged − 1.8 °C. The MAST was colder on western-facing slopes at high elevations. Surface and deeper probes had similar MASTs; however, deeper probes had less daily and seasonal variation. Another model developed at the regional scale based on proxy indicators of permafrost (rock glaciers and land cover) classified 5.1 km 2 of permafrost within the study area, whereas co-kriging interpolations of frost numbers and TTOP data indicated 2.0 km 2 and 4.6 km 2 of permafrost, respectively. Only 0.8 km 2 were common among all three models. Three boreholes drilled within 2 m of TRR indicate that permafrost does not exist at these locations despite each borehole being classified as containing permafrost by at least one model. Addressing model uncertainty is important because nutrients stored within frozen or frost-affected soils can be released and impact alpine water bodies. The uncertainty also exposes two fundamental problems: empirical models designed for high latitudes are not necessarily applicable to mountain permafrost, and the presence of mountain permafrost in the alpine tundra of the Colorado Front Range has not been validated.