Graduate Defense: Angela Seibert

Thesis Information

Title: Using Terrestrial Laser Scanning to Estimate Leaf Area Index in Peatland Conifers Under a Climate Manipulation

Program: Master of Science in Geosciences

Advisor: Dr. Nancy Glenn, Geosciences

Committee Members: Dr. Anna Bergstrom, Geosciences; and Dr. Jeffrey Warren, Geosciences

Abstract

Northern peatlands cover a small portion of the Earth, but they are a major terrestrial carbon sink, storing approximately 415 150 Gt of carbon (Beaulne et al., 2021). Peatland vegetation composition impacts the ecosystem’s capacity to accumulate and store organic matter. Recent research has found that as peatlands shift from mosses to woody vegetation, their carbon storage capacity begins to decrease; this process is termed shrubification (Chasmer et al., 2020; Malhotra et al., 2020; Nelson et al., 2021). The trees that make up treed peatlands, such as a treed peatland in northern Minnesota, affect the concentration of solar insolation and evapotranspiration rates; subsequently impacting the Sphagnum sp. moss in the understory. Quantifying peatland tree canopy structure over time under manipulated temperatures and CO2 levels as part of a larger climate study known as Spruce and Peatland Responses Under Changing Environments (SPRUCE), will help elucidate what ecological processes the trees may be affecting that, in turn, impact peatland carbon storage.
Leaf area index (LAI), the total projected leaf area per unit ground area (m2/m2.), is a necessary canopy structure component in land ecosystem models representing the plant surface available for sunlight and carbon dioxide (CO2) intake. Therefore, it is a necessary component for simulating net productivity in land ecosystem models. The LAI can be estimated using either direct or indirect methods. Direct methods are laborious and involve manually measuring tree leaves. Some common indirect methods include digital hemispherical photography (DHP), airborne laser scanning (ALS), and terrestrial laser scanning (TLS). Due to the scale and experimental setup of the peatland at SPRUCE, TLS is the ideal method for estimating LAI in this study. TLS-based LAI estimates can be made using a variety of methods: gap fraction-based, volumetric pixel-based (voxel-based), and biophysical regression-based LAI estimation. Since using TLS is an indirect method of estimation there are multiple sources of possible error such as the contribution of nonphotosynthetic material (wood points), the mean projection coefficient, and leaf overlapping or clumping. We address these sources of error using a modified version of voxel-based contact frequency (VCP) model (Hosoi et al., 2006; Li et al., 2017). This study aims to 1) apply a Random Forest (RF) model and a VCP model to estimate LAI as validated by destructive harvesting and DHP LAI estimates and 2) to apply the model to predict the LAI of 12 peatland plots at the entire plot and species level under CO2 and temperature manipulations from 2015 – 2022.
Estimating the LAI of conifers under treatment conditions at a fine spatial scale has a variety of challenges. Using traditional, less expensive indirect methods such as DHP to estimate LAI is challenging at SPRUCE due to large walls required for treatment surrounding tree plots that alter the lighting of the photographs. Additionally, SPRUCE researchers had previously used destructive biophysical linear regressions to derive allometric relationships for estimating LAI, but these were pre-treatment and do not account for changes in canopy structure or LAI due to warming and CO2 treatments. This problem results in the need for a model that is robust to changing parameters that require destructively harvested validation. The first question of this study is 1) how accurately can we predict the LAI of two peatland conifers, Picea mariana (black spruce) and Larix laricina (eastern larch), using TLS data and a voxel-based method? As an active remote sensing technique, TLS is unaffected by the walls surrounding the SPRUCE plots. The modified version of the VCP model is robust to changing voxel sizes and maintains moderate accuracy at different voxel sizes. To build the model we first applied a RF model with features of reflectance and return number to remove wood points. The leaf point clouds are then applied to the VCP model where the tree or plot is voxelized and sliced into horizontal layers. For each layer, we calculated the number of contacts between laser beams and foliage and applied a mean projection coefficient and leaf angle-based correction factor. The model was validated against 8 destructively harvested LAI estimates of trees outside of the chambers and 2 unwalled plot DHP-based LAI estimates. We found that the VCP model had a coefficient of determination of 0.89 (R2 = 0.89), a RMSE of 0.98, and a nRMSE = 0.17 as validated against destructively harvested (n = 8 trees) and DHP LAI estimates (n = 2 plots) of LAI. The modified VCP model can accurately predict LAI in environments where treatment conditions affect regressive estimates of LAI.
Quantifying tree canopy structure over time is useful for monitoring how ecological processes such as transpiration and photosynthesis are affected by the environment. Estimating LAI, leaf inclination angle distributions, and leaf area densities can be laborious and destructive if done using manual methods. The second question of this study is 2) to what degree are the spruce and larch tree canopy structures within 12 SPRUCE plots changing from 2015 – 2022? To evaluate the SPRUCE plot conifer canopies we collected TLS scans of 12 SPRUCE plots in August, during the trees’ maximum growing season, from 2015 to 2022. We applied the RF and VCP model to each plot under four temperature treatments and two carbon dioxide treatments. Plot LAIs were estimated across all trees and estimated for each subset of species within a plot. Using a Mann-Kendall Test, we analyzed the LAI trendlines from 2015 – 2022 at the whole plot and species scale. The VCP model provided estimates of leaf angle distribution and leaf area densities. We analyzed the leaf angles and densities to evaluate how the vertical profile of the foliage was changing over time at the whole plot and species scale. Averaging across all trees in the elevated CO2 plots with increased temperatures or blowers, LAI values exhibited a statistically significant increasing trend (m = 0.004, p < 0.01). Within ambient CO2 plots with increased temperatures or blowers, on average, LAI values had a statistically significant decreasing trend (m = -0.001, p < 0.01). In the control plots with no walls, no elevated temperature, and no elevated CO2, LAI values resulted in a decreasing trend (m = -0.013, p < 0.01). In general, spruce tree LAI increased significantly under elevated CO2 (m = 0.006, p < 0.01) and larch tree LAI decreased significantly (m = -0.002, p < 0.01). Under control conditions, spruce tree LAI remained constant over time (p > 0.5) whereas larch tree LAI decreased significantly (m = -0.01, p < 0.01). The results indicate that the canopy structure at the SPRUCE site is generally changing over time, the spruce and larch tree dynamics may affect each other’s net resource availability, leading to the opposing growth relationships.