Franklin Sullivan

Mr. Franklin Sullivan

orcid

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
17 January 2020 in Remote Sensing
Verified
Journal Article
Open access
PDF file
JHJack Hastings
Jack Hastings
SOScott V. Ollinger
Scott V. Ollinger
AOAndrew P. Ouimette
Andrew P. Ouimette
RSRebecca Sanders-DeMott
Rebecca Sanders-DeMott
MPMichael Palace
Michael Palace
DODavid Orwig
David Orwig
The ability to automatically delineate individual tree crowns using remote sensing data opens the possibility to collect detailed tree information over large geographic regions. While individual tree crown delineation (ITCD) methods have proven successful in conifer-dominated forests using Light Detection and Ranging (LiDAR) data, it remains unclear how well these methods can be applied in deciduous broadleaf-dominated forests. We applied five automated LiDAR-based ITCD methods across fifteen plots ranging from conifer- to broadleaf-dominated forest stands at Harvard Forest in Petersham, MA, USA, and assessed accuracy against manual delineation of crowns from unmanned aerial vehicle (UAV) imagery. We then identified tree- and plot-level factors influencing the success of automated delineation techniques. There was relatively little difference in accuracy between automated crown delineation methods (51–59% aggregated plot accuracy) and, despite parameter tuning, none of the methods produced high accuracy across all plots (27—90% range in plot-level accuracy). The accuracy of all methods was significantly higher with increased plot conifer fraction, and individual conifer trees were identified with higher accuracy (mean 64%) than broadleaf trees (42%) across methods. Further, while tree-level factors (e.g., diameter at breast height, height and crown area) strongly influenced the success of crown delineations, the influence of plot-level factors varied. The most important plot-level factor was species evenness, a metric of relative species abundance that is related to both conifer fraction and the degree to which trees can fill canopy space. As species evenness decreased (e.g., high conifer fraction and less efficient filling of canopy space), the probability of successful delineation increased. Overall, our work suggests that the tested LiDAR-based ITCD methods perform equally well in a mixed temperate forest, but that delineation success is driven by forest characteristics like functional group, tree size, diversity, and crown architecture. While LiDAR-based ITCD methods are well suited for stands with distinct canopy structure, we suggest that future work explore the integration of phenology and spectral characteristics with existing LiDAR as an approach to improve crown delineation in broadleaf-dominated stands.

8 July 2024 in Journal of Geophysical Research: Biogeosciences
Verified
Journal Article
Open access
MKMcKenzie A. Kuhn
McKenzie A. Kuhn
RVRuth K. Varner
Ruth K. Varner
CMCarrie McCalley
Carrie McCalley
CPClarice R. Perryman
Clarice R. Perryman
MAMika Aurela
Mika Aurela
Mark P. WaldropMark P. Waldrop
Mark P. Waldrop
Northern peatlands are a globally significant source of methane (CH4), and emissions are projected to increase due to warming and permafrost loss. Understanding the microbial mechanisms behind patterns in CH4 production in peatlands will be key to predicting annual emissions changes, with stable carbon isotopes (δ13C‐CH4) being a powerful tool for characterizing these drivers. Given that δ13C‐CH4 is used in top‐down atmospheric inversion models to partition sources, our ability to model CH4 production pathways and associated δ13C‐CH4 values is critical. We sought to characterize the role of environmental conditions, including hydrologic and vegetation patterns associated with permafrost thaw, on δ13C‐CH4 values from high‐latitude peatlands. We measured porewater and emitted CH4 stable isotopes, pH, and vegetation composition from five boreal‐Arctic peatlands. Porewater δ13C‐CH4 was strongly associated with peatland type, with δ13C enriched values obtained from more minerotrophic fens (−61.2 ± 9.1‰) compared to permafrost‐free bogs (−74.1 ± 9.4‰) and raised permafrost bogs (−81.6 ± 11.5‰). Variation in porewater δ13C‐CH4 was best explained by sedge cover, CH4 concentration, and the interactive effect of peatland type and pH (r2 = 0.50, p < 0.001). Emitted δ13C‐CH4 varied greatly but was positively correlated with porewater δ13C‐CH4. We calculated a mixed atmospheric δ13C‐CH4 value for northern peatlands of −65.3 ± 7‰ and show that this value is more sensitive to landscape drying than wetting under permafrost thaw scenarios. Our results suggest northern peatland δ13C‐CH4 values are likely to shift in the future which has important implications for source partitioning in atmospheric inversion models.

21 January 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
FSFranklin Sullivan
Franklin Sullivan
JHJack H. Hastings
Jack H. Hastings
SOScott V. Ollinger
Scott V. Ollinger
AOAndrew Ouimette
Andrew Ouimette
ARAndrew Richardson
Andrew Richardson
MPMichael Palace
Michael Palace
Forest canopy near-infrared reflectance and mass-based canopy nitrogen concentration (canopy %N) have been shown to be positively correlated. While the mechanisms underpinning this relationship remain unresolved, the broad range of wavelengths involved points to structural properties that influence scattering and covary with %N. Despite this, efforts that have focused on commonly measured structural properties such as leaf area index (LAI) have failed to identify a causal mechanism. Here, we sought to understand how lidar-derived canopy traits related to additional properties of foliar arrangement and structural complexity modulate the effects of leaf spectra and leaf area index (LAI) on canopy reflectance. We developed a leaf layer spectra model to explore how canopy reflectance would change if complex foliage arrangements were removed, compressing the canopy into optically dense, uniform stacked layers while maintaining the same leaf area index. Model results showed that LAI-weighted leaf reflectance saturates at a leaf area index of approximately two for needleleaf species and four for broadleaf species. When upscaled to estimate plot-level canopy reflectance in the absence of structural complexity (NIRrLAI), results showed a strong positive relationship with canopy %N (r2 = 0.86), despite a negative relationship for individual leaves or “big-leaf” canopies with an LAI of one (NIRrL, r2 = 0.78). This result implies that the relationship between canopy near-infrared reflectance and canopy %N results from the integrated effects of canopy complexity acting on differences in leaf-level optical properties. We introduced an index of relative reflectance (IRr) that shows that the relative contribution of structural complexity to canopy near-infrared reflectance (NIRrC) is related to canopy %N (r2 = 0.55), with a three-fold reduction from potential canopy near-infrared reflectance observed in stands with low %N compared to a two-fold reduction in stands with high %N. These findings support the hypothesis that the correlation between canopy %N and canopy reflectance is the result of interactions between leaf traits and canopy structural complexity.

24 October 2023 in Remote Sensing
Verified
Journal Article
Open access
PDF file
FSFranklin Sullivan
Franklin Sullivan
AHAdam George Hunsaker
Adam George Hunsaker
MPMichael Palace
Michael Palace
Jennifer JacobsJennifer Jacobs
Jennifer Jacobs
Recently, sensors deployed on unpiloted aerial systems (UAS) have provided snow depth estimates with high spatial resolution over watershed scales. While light detection and ranging (LiDAR) produces precise snow depth estimates for areas without vegetation cover, there has generally been poorer precision in forested areas. At a constant flight speed, the poorest precision within forests is observed beneath tree canopies that retain foliage into or through winter. The precision of lidar-derived elevation products is improved by increasing the sample size of ground returns but doing so reduces the spatial coverage of a mission due to limitations of battery power. We address the influence of flight speed on ground return density for baseline and snow-covered conditions and the subsequent effect on precision of snow depth estimates across a mixed landscape, while evaluating trade-offs between precision and bias. Prior to and following a snow event in December 2020, UAS flights were conducted at four different flight speeds over a region consisting of three contrasting land types: (1) open field, (2) deciduous forest, (3) conifer forest. For all cover types, we observed significant improvements in precision as flight speeds were reduced to 2 m s−1, as well as increases in the area over which a 2 cm snow depth precision was achieved. On the other hand, snow depth estimate differences were minimized at baseline flight speeds of 2 m s−1 and 4 m s−1 and snow-on flight speeds of 6 m s−1 over open fields and between 2 and 4 m s−1 over forest areas. Here, with consideration to precision and estimate bias within each cover type, we make recommendations for ideal flight speeds based on survey ground conditions and vegetation cover.

27 September 2026 in Glacies
Verified
Journal Article
Open access
PDF file
FSFranklin Sullivan
Franklin Sullivan
EBElizabeth Burakowski
Elizabeth Burakowski
SNSarah J. Nelson
Sarah J. Nelson
JBJoshua Beneš
Joshua Beneš
SCSamantha Cheung
Samantha Cheung
LZLucas Zukiewicz
Lucas Zukiewicz
The Northeast Snow Survey Feasibility Study began in October 2023 to characterize snow monitoring in the Northeastern United States and identify data and knowledge gaps to guide the deployment of a snow monitoring network similar to the western United States’ automated Snow TELemetry (SNOTEL) network. Researchers and interest holders in the region were contacted and engaged through online surveys, interviews, and email, using snowball sampling to identify and evaluate data breadth, gaps, and needs for snow monitoring in the region. Surveys identified important data gaps, including under-sampling of high-elevation areas, temporal gaps early in winter, low-frequency (weekly to monthly) temporal resolution in manual snow survey efforts, and a paucity of automated snow monitoring infrastructure. The data gaps, combined with previously published interest holder data needs, led to the development of three primary network objectives and guided the systems engineering and spatial design. Here, an open-source workflow using Google Earth Engine and R was developed to identify potential sites in the northern New England states of Maine, New Hampshire, and Vermont. This process identified 4015 sq km of suitable land for network siting across the three states that would help overcome the identified spatial and temporal data gaps while meeting interest holder data needs. This study establishes the need for a coordinated snow monitoring network in the Northeastern United States and presents a spatial design workflow that can be implemented to meet the identified objectives.

15 January 2020 in ESS Open Archive
Verified
Preprint
Open access
KBKaitlyn Baillargeon
Kaitlyn Baillargeon
SOScott V. Ollinger
Scott V. Ollinger
AOAndrew Ouimette
Andrew Ouimette
RSRebecca Sanders-DeMott
Rebecca Sanders-DeMott
FSFranklin Sullivan
Franklin Sullivan
OBOlivia Bartlett
Olivia Bartlett
Forest biodiversity has been declining across the globe due to anthropogenic activities. Losses of biodiversity have led to reduced forest health and ecosystem services. Therefore, it has become necessary to monitor changes in biodiversity over wide geographic areas. Remote sensing has the potential to monitor biodiversity changes, but the accuracy in which it can be estimated is under debate. In this study, we tested 1) the relationship among distinct metrics of biodiversity, 2) the role topographic measures have on determining biodiversity, and 3) the ability of hyperspectral remote sensing to estimate biodiversity in temperate forests of the Northeastern United States. We characterized biodiversity according to four different metrics: species, functional, structural, and phylogenetic diversity. All four metrics were quantified using species inventory data as well as Light Detecting and Ranging (LiDAR) to calculate additional indices of structural diversity. A digital elevation model was used to obtain measures of slope, aspect, and other topographic indices such as topographic wetness. Hyperspectral imagery was used to obtain reflectance, entropy, and several vegetation indices. In our analyses, species, functional, and phylogenetic diversity were shown to be moderately correlated suggesting similarities between the metrics while correlations between structural diversity and the other metrics were weak. The calculated topographic indices also showed weak correlations with the biodiversity metrics suggesting that topography does not influence measures of biodiversity at the plot level. Depending on the biodiversity metric, relationships between the hyperspectral analyses and biodiversity were weak to moderately strong. These findings suggest that hyperspectral imagery holds some potential for estimating multiple metrics of biodiversity.