(MC) INTEGRATION OF HIGH RESOLUTION REMOTE SENSING, FIELD OBSERVATION AND MODELING TO EXAMINE FOREST STRUCTURE, COMPOSITION AND DYNAMICS

• Sponsoring Institution: National Institute of Food and Agriculture
• Project Status: ACTIVE
• Funding Source: MCINTIRE-STENNIS
• Reporting Frequency: Annual
• Accession No.: 1024612
• Project Start Date: Nov 2, 2020
• Project End Date: Sep 30, 2025
• Program Code: [(N/A)]- (N/A)

Recipient Organization
UNIVERSITY OF FLORIDA
G022 MCCARTY HALL
GAINESVILLE,FL 32611

Performing Department
Forest Resources and Conservation

Non Technical Summary
Local forests cannot be managed or conserved based only on small-scale, local assessments of forest health and status because (1) there is limited ability to assess large tracts of land from the ground and (2) regional factors, such as the ways land around each forest is used (i.e. is it urban, agriculture, or industrial?) and global factors, such as increasing temperatures or more frequent droughts, affect forest even on local scales. Remote sensing, which involves images taken from satellites, aircraft and drones, can help improve management and conservation of forests by providing large scale assessments of forests and putting each piece of forest in context of the larger region. For example, the type of research we are conducting can potentially show where individual tree growth is declining across kilometers of forest, thus allowing interventions to improve forest health. Our research also aims to understand what factors, such as change in climate that affect certain local sites more than others, affects forest status and health.The objectives of my research are to map the locations, sizes and properties (species identity, growth rate, nutritional status, phenology) of individual trees species over large areas using remote sensing images that have high enough resolution to detect individual trees. This "high resolution" remote sensing includes drones, sensors on aircraft, and certain satellites. We will particularly take advantage the large federally-funded program called the National Ecological Observatory Network (NEON) that is collecting standardized data on large suite of ecosystem features, such as mammals, plants, mosquitoes, and CO2 flux. NEON includes a high resolution state-of-the-art airborne sensor and will provide an unprecedented opportunity to examine the health and status of forested ecosystems. In addition to assessing individual trees and forests using remote sensing, we use this information in computer models of forest function that are used to predict how forests will perform under different scenarios, such as increased temperature or more frequent drought.

Animal Health Component

30%

Research Effort Categories

Basic

70%

Applied

30%

Developmental

0%

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
123	0699	1070	100%

Knowledge Area
123 - Management and Sustainability of Forest Resources;

Subject Of Investigation
0699 - Trees, forests, and forest products, general;

Field Of Science
1070 - Ecology;

Keywords

forest health

tree growth

remote sensing

climate change

drone

hyperspectral

Goals / Objectives
1. Develop techniques and algorithms to use high resolution remote sensing (color photography, multi- and hyper-spectral images, lidar mounted on aircraft, UAS, satellites, towers and ground-based) to map landscape patterns of forest composition (e.g., species composition, plant functional traits), structure (e.g., tree density and sizes), dynamics (e.g., growth, mortality, phenology) and health (e.g., water stress, pest attack). Validate techniques and algorithms with physiological, structural, compositional, phenological and allometric field data taken on individual trees and stands.2. Understand the abiotic (e.g., soils, wind, fire and drought disturbance) and human drivers (e.g., management, land use, silviculture) generating the landscape patterns of forest composition, structure and function using mapped data from high resolution remote sensing.3. Integrate ecological data derived from high resolution remote sensing and field data into diagnostic and predictive models of forest structure, dynamics and ecosystem services.4. Use understanding gained by linking field data and high resolution remote sensing to interpret coarse scale remote sensing data with broader spatial coverage.

Project Methods
Combining different types of high resolution data (satellite, airborne hyperspectral and lidar, UAV, and webcams) at the Ordway Swisher Biological Station (OSBS), and other sites in the Southeast and across the U.S., we will investigate impacts of management, climate and natural disturbance on forest composition, structure and function, for example species distributions, cone crops, and vine dynamics. At the local scale, because of the proximity of OSBS to the UF campus, we will take frequent field measurements to validate remote sensing interpretations, including tree species mapping, tree size, leaf trait ground-based canopy and understory density measurements. To understand drivers, we will compile from all sites GIS-based data on land use history, fire frequency and other land management actions, to link remote sensing data with detailed management activities that are recorded at the field station. To map variation in functional traits within and among species, we will collect field spectra in conjunction with leaf collection to measure leaf traits, matched with soil and plant microbial analysis. This will also be linked to airborne and UAV data. This work will be done at all forested NEON sites in the US forest as well. Research at NEON will be enhanced by utilizing and developing the large (23 ha) Ordway Swisher forest dynamics plot (https://forestgeo.si.edu/sites/north-america/ordway-swisher) that was established in 2019, where every stem 1 cm DBH and above has been mapped precisely, its species identified and diameter measured. For sites outside the US, we will rely on data from NEON, and potentially additional measurements strategically collected for maximum benefit.We will work toward synthesizing RS-based ecosystem observations into models of forest structure and function. This includes continuing to develop the Perfect Plasticity Approximation (PPA) forest model, with specific focus on inclusion of modeling growth and mortality parameters from remote sensing using single high resolution hyperspectral images and multiple high-resolution optical and lidar images. In addition, we will explore how functional traits data that can be mapped from images are related to dynamic (growth and mortality) and structural (tree allometry) data, such that suites of important model variables can be parameterized from single key traits detected by remotely sensed images.Finally, we will apply results from the ecological interpretation of high resolution images to coarse scale image data that has broader availability, and spatial and temporal coverage. We will investigate analysis tools and algorithms to link high and coarse resolution images and test to what degree ecological interpretations of remote sensing images apply to both high and coarse resolution images. When beneficial, methods that are tested in international projects will be used to enhance methods and outcomes for research in the southeast US and across the continent.

Progress 11/02/20 to 09/30/21

Outputs
Target Audience:Forest scientists, scientists that use remote sensing for forest research and management Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?The project has supported a graduate student and post-doc. Gradate students and postdocs work in a collaborative group between forest ecologists and computer scientists to develop techniques to characterize individual trees from remote sensing, allowing cross-field learning between these two fields. Due to COVID, opportunities for travel, workshops and conferences have been limited. How have the results been disseminated to communities of interest?Two publicly available data sets of delieating tree crowns for hundreds of thousands of trees in 30 sites across the US have been released. These data sets can be used for analysis of forest ecosystems, as well as development and testing of remote sensing algorithms to detect trees. https://zenodo.org/record/3770410# https://zenodo.org/record/3765872 What do you plan to do during the next reporting period to accomplish the goals?Goal 1: Our group is focused this year on development of algorithms to identify tree species from remote sensing data across 30 forested sites in the US using field and remote sensing data from the National Ecological Observatory Network. One element of this is that we are writing up results from our second data science competition that worked with research groups across the world to develop algorithms that identify tree species across multiple forest ecosystem types. Goal 2: At the local large forest dynamics plot, we are examining how tree regeneration is affected by fire as well as supporting a range of nationwide and global scale studies using a network of large forest dynamic plots (forestGEO). We are also use continental scale predications of leaf traits to understand multi-scale environmental drivers of trait variation. Goal 3: We are developing a modeling framework to determine species identity and property boundaries determine rates of forest regeneration. Furthermore, we are developing a model to predict species trait variation from remote sensing data on leaf cover of individual trees.

Impacts
What was accomplished under these goals? Goal 1: Developed techniques to detect and delineate individual tree crowns from remote sensing data (Stewart et al. 2021) including deep learning algorithms (Weinstein et al. 2020). We generated a continental scale data set that can be used as baseline data for algorithm development of tree detection from remote sensing data (Weinstein et al. 2021 a,b). We also developed a technique to map tree foliar traits across the continent using remote sensing and field data (Marconi et al. 2021). Goal 2: At the local large forest dynamics plot (Davies et al. 2021) that was set up in the last 2 years, we determined canopy tree density and species identity interact to determine tree regeneration (Johnson et al. 2021). Goal 3: We developed a modeling framework to determine rates of forest regeneration using satellite remote sensing data that incorporated uncertainty due to phenology, rate of succession and noise in satellite spectral reflectance (Caughlin et al. 2021). We used local hydrologic data, remote sensing and models to determine the impact of dams on floodplains and riparian land cover (Swanson and Bohlman 2021 a,b).

Publications

• Type: Journal Articles Status: Published Year Published: 2021 Citation: Stewart, D., A. Zare, S. Marconi, B. G. Weinstein, E. White, S. Bohlman, and A. Singh. 2021. RandCrowns: A Quantitative metric for imprecisely labeled tree crown delineation. ISPRS Journal of Photogrammetry and Remote Sensing 14:11229-11239 doi: 10.1109/JSTARS.2021.3122345.
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Merrick, T., S. Pau, S., M. Detto, E. Broadbent, S. Bohlman, C. Still, and A. Almeyda Zambrano. 2021. Unveiling spatial and temporal heterogeneity of a tropical forest canopy using high-resolution NIRv, FCVI, and NIRvrad from UAS observations. Biogeosciences 18(22): 6077-6091 doi: 10.5194/bg-18-6077-2021
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Loubata Panzou,& S. Bohlman&. and T. Feldpausch. . 2021. Large continental variability in tropical tree crown allometry. Global Ecology and Biogeography 30 (2), 459-475. doi: 10.1111/geb.13231
• Type: Journal Articles Status: Published Year Published: 2020 Citation: Weinstein, B., S. Marconi, S. Bohlman, A. Zare and E. White. 2020. Cross-site learning in deep learning RGB tree crown detection. Ecological Informatics 56: 101061
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Mohan, M., R. Vieira Leite, E. Broadbent, W. S. W. Mohd Jaafar, S. Srinivasan, S. Bajaj, A. P. Dalla Corte, C. Hummel do Amaral, G. Gopan, C. A. Silva, S. N. M. Saad, A. M. M Kamarulzaman., G. A. Prata, E. Llewelyn, D. J. Johnson, W. Doaemo, A. M. Almeyda Zambrano, S. Bohlman, and A. Cardil. 2021. Individual tree detection using UAV-lidar and UAV-SfM data: A Stand-alone Tutorial for Beginners. Open Geosciences 13(1): 1028-1039. https://doi.org/10.1515/geo-2020-0290
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Borden, J. B., S. Bohlman and B. R. Scheffers. 2021. Niche plasticity mitigates the effects of invasion but not urbanization. Oecologia. doi: 10.1007/s00442-021-05039-x
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Swanson, A. C., D. Kaplan, D., K. B. Toh, E. Marques, and S. Bohlman. 2021. Changes in floodplain hydrology following serial damming of the Tocantins River in the eastern Amazon. Science of The Total Environment 800: 149494 doi: 10.1016/j.scitotenv.2021.149494
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Swanson, C. and S. Bohlman. 2021. Cumulative impacts of land cover change and dams on the land-water interface of the Tocantins River. Frontiers in Environmental Science 9: 120 doi: 10.3389/fenvs.2021.662904.
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Weinstein, B.G., S. Graves, S. Marconi, A. Singh, A. Zare, D. Stewart, S. Bohlman, and E. White. 2021. A benchmark dataset for canopy crown detection and delineation in co-registered airborne RGB, LiDAR and hyperspectral imagery from the National Ecological Observation Network. PLOS Computational Biology 17 (7), e1009180 doi: 10.1371/journal.pcbi.1009180
• Type: Journal Articles Status: Published Year Published: 2020 Citation: Kattge, J., B�nisch, G., Diaz, S., Lavorel, S., Prentice, I. C., Leadley, P., &.S. Bohlman... & Wirth, C. 2020. TRY plant trait database-enhanced coverage and open access. Global Change Biology 26(1), 119-188
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Weinstein, B.G., S. Marconi, S. Graves, S. Bohlman, A. Zare and E. White. 2021. A remote sensing derived dataset of 100 million individual tree crowns for the National Ecological Observatory Network. eLife 10:e62922 doi: 10.7554/eLife.62922
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Johnson, D. J., L. Magee, K. Pandit, J. Bourdon, E. N. Broadbent, K. Glenn, Y. Kaddoura, S. Machado, J. Nieves, B. Wilkinson, A. M. Almeyda Zambrano, and S. Bohlman. 2021. Canopy tree density and identity influence tree regeneration patterns and woody species diversity in a longleaf pine forest. Forest Ecology and Management 490, 119082 doi: 10.1016/j.foreco.2021.119082
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Marconi, S., S. J. Graves, B. Weinstein, S. Bohlman, and E. P. White. 2021. Estimating individual-level plant traits at scale. Ecological Applications 31(4):e02300. Doi: 10.1002/eap.2300
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Caughlin, T., C. Barber Alvarez-Buylla, G. Asner, N. Glenn, and S. Bohlman. 2021. Monitoring tropical forest succession at landscape scales despite uncertainty in Landsat time series. Ecological Applications 31(1):e02208 doi:10.1002/eap.2208
• Type: Journal Articles Status: Published Year Published: 2021 Citation: Davies, S., Abiem, I., Salim, K., Aguilar, S., Allen, D., Alonso, A., Anderson-Teixeira, K., Andrade, A., Arellano, G., Ashton, P.S. and Baker, P.J., S. Bohlman et al. 2021. ForestGEO: Understanding forest diversity and dynamics through a global observatory network. Biological Conservation, 253, p.108907. doi: 10.1016/j.biocon.2020.108907