Identifying Canopy Snow in Subalpine Forests: A Comparative Study of Methods

Abstract

The interception of snow by the canopy is an important process in the water and energy balance in cold-region coniferous forests. Direct measurements of canopy snow interception are difficult at scales larger than individual trees, requiring indirect methods such as eddy covariance, time-lapse photography, or modeling. At the Niwot Ridge Subalpine Forest AmeriFlux site in the Colorado Front Range, USA, we compared methods that estimate or simulate the presence of snow interception. Timelapse photography images were analyzed using thresholding analysis and used to train a Convolutional Neural Network (CNN) model to estimate canopy snow presence. Interception was also estimated from eddy covariance measurements above and below the canopy, as well as from model simulations. These methods were applied over January 2019, with binarized results compared to a “ground truth” of human labeled images to calculate the Balanced Accuracy Score. The highest accuracy was achieved by the CNN predictions. Based on the Balanced Accuracy Scores, select methods were extended to estimate the presence of canopy snow for the 2018/2019 winter. All methods provided insight into the process of interception in a subalpine forest but presented challenges, including differing flux footprints of the above- and below-canopy eddy covariance measurements and the inability of red-green-blue imagery to monitor snow interception at night, during sunrise, and during sunset.