BROAD-SCALE MONITORING OF CANOPY STRUCTURE WITH MULTI-SOURCE REMOTE SENSING AND ALTERNATIVE REFERENCE DATA

Abstract

In a time of shifting disturbance regimes and anthropogenic pressures, effective management of forest and woodland ecosystems depends on continuous, landscape-level monitoring of canopy structure. While remote sensing can help meet this need, its broad operational implementation is often impeded by a myriad of data constraints, necessitating careful tailoring of monitoring programs to specific management objectives across heterogeneous environments. For example, certain regions, such as African savannas, are heavily data-limited and present unique challenges for detecting subtle structural variations within sparse, low-stature vegetation. Conversely, data-rich regions introduce complexities in reconciling disparate reference datasets created with varied methodologies. Furthermore, widely available 3D remote sensing resources like digital aerial photogrammetry (DAP) remain vastly underutilized, despite directly measuring fundamental attributes like canopy height. A final, critical gap is that remote sensing maps are rarely evaluated for one of their primary intended applications, which is aiding in the estimation of population means or totals for vegetation attributes over distinct management areas. To address these gaps, this dissertation investigates the strategic integration of multi-sensor remote sensing and diverse reference datasets across forest and savanna ecosystems. Crucially, each investigation moves beyond map development to evaluate how these products improve population estimates of vegetation structure using post-stratified and small area estimators. This hybrid approach empowers land managers and forest inventory programs by pairing time-series maps that visualize continuous landscape dynamics with rigorous statistical estimates that deliver defensible, population-level metrics for adaptive decision-making. The first chapter evaluates combining remote sensing predictors from multiple sources for extending canopy measurements from spaceborne waveform lidar (GEDI) across data-limited African savannas. By combining optical time series, synthetic aperture radar, and environmental covariates, parsimonious models successfully quantified annual structural changes from herbivory and woody encroachment, though the estimated magnitude of change was muted. The second chapter examines how the selection of canopy cover reference data for building Landsat-based canopy cover maps ultimately influences the precision of population estimates of forest area and aboveground biomass in the Rocky Mountains. Comparing diverse field, aerial, and lidar-based canopy cover sources revealed that the efficiency of post-stratified estimators for aboveground biomass is driven primarily by the plot-level correlation between the reference data and the target variable, rather than marginal improvements in mapping accuracy. Finally, the third chapter demonstrates the operational viability of multi-temporal DAP for statewide forest structure estimation and post-fire canopy change assessment. While DAP struggled to capture residual standing dead stems and sparse woodlands at high resolutions, spatial aggregated canopy height models closely matched airborne lidar benchmarks and achieved nearly identical precision in statewide aboveground biomass estimation. These investigations together advance broad-scale operational use of remote sensing in diverse and complex environments to support forest inventories and effective land management.