Wetlands hold the highest density of belowground carbon stocks on earth, provide myriad biogeochemical and habitat functions, and are at increasing risk of degradation due to climate and land use change. Microtopographic variation is a common and functionally important feature of wetlands but is challenging to quantify, constraining estimates of the processes and functions (e.g., habitat diversity, carbon storage) that it regulates. We introduce a novel method of quantifying fine-scale microtopographic structure with Terrestrial Laser Scanning using 10 black ash (Fraxinus nigra) wetlands in northern Minnesota, USA as test cases. Our method reconstructs surface models with fine detail on the order of 1 cm. Our independent validation verifies the surface models capture hummock (local high points) and hollow (local low points) features with high precision (RMSE = 3.67 cm) and low bias (1.26 cm). A sensitivity analysis of surface model resolution showed a doubling of model error between 1 cm and 50 cm resolutions, suggesting high-resolution reconstructions most precisely capture surface variation. We also compared five classification methods at resolutions ranging from 1 cm to 1 m and determined that maximum likelihood classification at 25 cm resolution most accurately (78.7%) identifies hummock and hollow features, but a simple thresholding of surface model elevation and slope was ideal for hummock feature delineation, retaining over 91% of hummock areas. Finally, we test and validate a novel microtopographic delineation method (TopoSeg) that accurately (Bias = 0.2–11.9%, RMSE = 19.6–24.1%) estimates the height, area, volume, and perimeter of individual hummock features. For the first time, we introduce an accurate and automated approach for quantifying fine-scale microtopography through high resolution surface models, feature classification, and feature delineation, enabling geospatial statistics that can explain spatial heterogeneity of habitat structure, soil processes, and carbon storage in wetland systems.
Bibliographical noteFunding Information:
We would like to thank our funding sources: Minnesota Forest Resources Council and Virginia Tech Institute for Critical Technology and Applied Science Fellowship . Also, thank you to the three anonymous reviewers of this manuscript and the Remote Sensing of Environment editorial team.
- Machine learning
- Model resolution
- Terrestrial LiDAR