Application of Iterative Noise-adding Procedures for Evaluation of Moment Distance Index for LiDAR Waveforms

Abstract

The new Moment Distance (MD) framework uses the backscattering profile captured in waveform LiDAR data to characterize the complicated waveform shape and highlight specific regions within the waveform extent. To assess the strength of the new metric for LiDAR application, we use the full-waveform LVIS data acquired over La Selva, Costa Rica in 1998 and 2005. We illustrate how the Moment Distance Index (MDI) responds to waveform shape changes due to variations in signal noise levels. Our results show that the MDI is robust in the face of three different types of noise—additive, uniform additive, and impulse. In effect, the correspondence of the MDI with canopy quasi-height was maintained, as quantified by the coefficient of determination, when comparing original to noise-affected waveforms. We also compare MDIs from noise-affected waveforms to MDIs from smoothed waveforms and found that windows of 1% to 3% of the total wave counts can effectively smooth irregularities on the waveform without risking of the omission of small but important peaks, especially those located in the waveform extremities. Finally, we find a stronger positive relationship of MDI with canopy quasi-height than with the conventional area under curve (AUC) metric, e.g., r2 = 0.62 vs. r2 = 0.35 for the 1998 data and r2 = 0.38 vs. r2 = 0.002 for the 2005 data.

Keywords

• Moment Distance Index
• MD framework
• waveform LiDAR
• MDI
• noise models