A new interpolation method to resolve under-sampling of UAV-lidar snow depth observations in coniferous forests

Abstract

Obtaining accurate snow depth estimates under dense canopies using airborne lidar (light detection and ranging) techniques is challenging due to the under-sampling of ground and snow surfaces. Existing interpolation techniques do not adequately address this problem and they often result in an overestimation of under-canopy snow depths. To address this issue, we introduce and evaluate a new interpolation method that incorporates intra-canopy snow depth variability to provide more accurate estimations at unsampled locations. Four interpolation methods were tested, considering systematic trends (landscape trend, canopy vs. gap trend, and intra-canopy trend) along with spatial interpolation of the residuals. Our results show that spatial interpolation methods without consideration of trends are sufficient to capture and reconstruct the small-scale variability of snow depths below a separation distance of 1 m between sampled and unsampled locations, (i.e., ground surface point density > 1 pt. m−2). However, beyond a separation distance of 2.5–3 m (point density < 0.33–0.40 pt. m−2), spatial interpolation based on proximity alone becomes unreliable because point separation becomes larger than the snow depth spatial correlation scale. Within these limiting distances, the method that incorporates trends along with spatial interpolation techniques can resolve the small-scale variability and thereby reduce the likely overestimation of snow depths under the canopy.