Progress 04/01/23 to 03/31/24
Outputs
Target Audience:The target audience in this establishment year has been our internal multi-disciplinary team. Our communications have focused on increasing the familiarity among our wide range of scientists, engineers and trainees, development of the project-wide implementation plan and facilitating points of integration among Objectives. External Advisory Board members, professional staff, postdoctorals and graduate and undergraduate students have been recruited into the project. Task 1.3 has begun engagement with forestry stakeholder professionals in a Delphi-method survey to help inform forest management decisions including planning, accounting and forecasting potential issues. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?Purdue's Institute for Digital Forestry encompasses over 50 graduate students and postdoctoral researchers. Though many are working on research directly related to PERSEUS, 4 grads and 3 undergrads are presently supported, or partially supported, by PERSEUS funding. Maine supports 5 grads/postdocs. The UGA team includes 3 non-thesis MS students supported through the UGA Center for Forest Business, and 6 graduate students (PhD and MS) supported through PERSEUS. A questionnaire on professional development activities was deployed to PERSESUS students and postdocs and the broader Purdue Digital Forestry student list in January 2023. This survey assessed topics and delivery methods students were most interested in for professional development activities. Two virtual professional development workshops were presented to PERSEUS graduate and undergrad students, bringing together students across the three institutions. The first, focused on science communication, had 9 student participants total (of which 2 were PERSEUS students). A learning module "Managing your online research presence," was developed to train PERSEUS students with online tools like Google Scholar profiles and ORCiD to track/manage their research-related activities. The module was offered in Spring 2024. It had 7 students, all PERSEUS. Plans are to increase the frequency of these workshops to approximately monthly. Undergrad students were on-boarded at UGA to assist with Objective 1 activities (LiDAR data processing and wood bark image collection) and Objective 4 (OLAF AI course development). Three non-thesis MS students (not funded by PERSEUS) are assisting in the development of south-wide GIS databases (roads, streams, soils, ownership, etc.). How have the results been disseminated to communities of interest?Preliminary progress updates have been shared only internally within the PERSEUS institutions in this establishment year. Early scientific accomplishments have been published through five refereed publications in our academic fields and 9 other presentations or events. What do you plan to do during the next reporting period to accomplish the goals?Task 1.1 iForester: We will improve algorithm robustness for DBH measurement and species identification with regional specificity, and further develop biometrics measures such as stem straightness and volume. We will combine smartphone LiDAR and RGB sensors to achieve automated measurement with high accuracy (Deliverable #1). The team also will use Real Time Kinematics (RTK) extensions of the smartphone, which are enabled by communication with permanently operating GNSS reference stations, to achieve 1 cm positional accuracy. Using such extensions is expected to improve the developed algorithm performance. These somewhat less expensive units may be evaluated for student laboratory coursework (Objective 4). Task 1.2: StemMapper: We will address multiple ongoing data analytics challenges: a) Tree-scale measurement, b) Species identification and c) Stand to landscape-scale inventories (Deliverable #3). We will continue improvement of the Backpack Mobile Mapping hardware and software systems for the integration of multi-beam spinning LiDAR, solid-state LiDAR, consumer-grade cameras, machine vision cameras and GNSS/INS units (Deliverable #2). An improved hardware design of the Backpack will facilitate a scalable system for implementation by stakeholders. The performance of F2-SLAM and S2-TEAM strategies will be enhanced to ensure the Backpack LiDAR systems' reliability against extended GNSS-signal occlusions in dense canopy under leaf-on conditions. More specifically, the ability to include tree detection, localization, and segmentation in the trajectory enhancement and point cloud generation will be expanded. Field data acquisition guidelines for track/flight configuration will be developed to allow for optimal mission planning strategies, sufficient data acquisition to ensure the geometric fidelity of derived products and successful integration of acquired data. Work will be done on data acquisition guidelines that relates to pros and cons of LiDAR, multi-spectral, and RGB. We will advance the multi-system integration of both UAV and Backpack datasets in a single System Calibration and Trajectory Refinement Procedure to take advantage of the near-proximal and proximal sensing nature of these systems. Task 1.3: Data Coverage: We will initiate a comparison analysis of a suite of existing and newly developed data products for regional-scale forest carbon assessment (Deliverable #4). Starting as a pilot project focusing on Maine with plans to expand to other states and regions of the eastern U.S., we will analyze data products including both maps of forest biomass and forest carbon model outputs. The forest biomass maps include estimations based on modeling airborne (e.g., USGS 3DEP, USDA 3D NAIP, and NASA G-LiHT) and spaceborne LiDAR (e.g., NASA GEDI), as well as air photo (i.e., NAIP) point clouds, evaluated against ground-based inventory (i.e., from the US Forest Inventory and Analysis (FIA) program) plot summaries. For forest carbon model outputs, we are comparing baseline (historical) estimates and future projections of different climate and management scenarios at various spatial scales. These models include the Forest Vegetation Simulator (FVS), LANDIS-II, the Canadian Carbon Budget Model (CBM-CFS3), and the Community Land Model (CLM) as evaluated against growth and yield data derived from FIA. Tasks 2.1 Landowner optimization and 2.2 Broad simulation: We will combine geospatial and spectral (Purdue LeafSpec scanner) resolutions to develop revolutionary imaging processing algorithms for more accurate and earlier detection of disease and stress trends to optimize ecosystem services at the local level and then scale to stand and landscape level assessments. This effort will collect paired spectral and leaf reference (i.e., chemistry and physiology) measurements that can serve as input variables for scaling information to air and space borne platforms. The Maine multi-model (FVS, LANDIS-II, CBM) will be adapted for Purdue and UGA for intercomparison and initiating regional decision-making frameworks. The Broad-scale scenario development effort (feeding 2.3 Value Chain [Year 3] and 2.4 Visualization) will analyze tree changes over time at large scale and relate to tree and forest health and disease trends. Task 2.4 Data Visualization: We will focus on broadening the regional coverage and diversifying the data collection efforts while maintaining the geospatial data layers hosted on the STAC service and D2S. Task 3.1: Stakeholder Perceptions: We will complete the Delphi process with forestry professionals. Results of the process will be used to further inform the development of technology and tools to ensure that they align with the needs and interests of the forestry community. Two survey instruments will be developed in Year 2, one led by Georgia and to be adapted for Maine and Indiana targeting forest businesses and industries, and the other led by Purdue and to be adapted by Georgia and Maine targeting forest landowners. Both surveys will be designed to gather quantitative data on individual perceptions, experiences and likelihood to use digital forestry technology. Task 3.2: Scenario development: We will assess management approaches based on landowner typology and evaluate "Future Visions" for the forest sector (e.g., ASU, For/Me). A pilot survey instrument in at least one state will be developed to evaluate risk perceptions and state preference elicitation techniques. Landowner typologies will be determined by linking ownership, regional socio-economic and FIA plot data to estimate forest management and harvest decisions over the past 20 years. Task 3.4 Technology Application: We will test digital technology and assess facilitators and barriers to adoption through a series of workshops, training sessions and other venues. Task 4.1 Learning communities: The undergraduate VIP cohort will be mentored in a course-based research experience. Task 4.2 Interns and fellows: Undergraduate research cohortswill continue to be recruited in VIP and non-VIP research activities. Professional development programs for graduate students will be further refined and implemented. Graduate students will be trained in mentorship skills and serve as mentors for the undergraduate cohorts. Task 4.3 Curriculum development: An initial hands-on field technology workshop will occur at the Summer annual meeting. This will encompass forest data acquisition employing UAVs and subsequent data processing and analysis and will include PERSEUS students and potentially area stakeholders. Task 4.4 Online certificate: The UAV Forest Data Acquisition and Processing workshop will be refined for inclusion in Purdue online Master of Forestry in Digital Natural Resources degree program.
Impacts
What was accomplished under these goals?
Task 1.1 iForester: Algorithms for the iPhone app were developed to automatically calculate DBH (with sub-inch accuracy) on individual trees given minimal user input. The prototype will advance transfer learning on region-specific bark images to retrain our lightweight Distilled-MoblieNet-V2 AI model to ensure tree identification accuracy for regions of the eastern forest. UGA interns are collecting images of tree bark for the SE. Task 1.2 StemMapper: Our custom-designed LiDAR-based and AI-assisted StemMapper platforms for automated stem- and stand-level inventory includes 4 Backpack and 2 UAV systems with integrated Global Navigation Satellite System/Inertial Navigation System (GNSS/INS), multi-beam spinning LiDAR and digital camera. The systems have been repeatedly tested through a data acquisition process that includes: 1) gathering of GNSS/INS, image and LiDAR data using proximal (Backpack and smartphone) and near-proximal (UAV) systems; 2) processing of GNSS/INS data for generation of geo-tagged imagery and point clouds; 3) data correctness and completeness quality control; 4) solicitation of airborne sensing data, e.g., Geiger-mode and 3D Elevation Program (3DEP) LiDAR, to evaluate the added benefit of proximal and near proximal sensing systems; and 5) in-field acquisition of reference data. The systems have undergone rigorous system calibration to precisely estimate the spatial and rotational offsets. Datasets have been collected from plots with extensive ground truth data for hardware integration verification: 3 plots in Purdue Martell Forest, 4 at the ME Penobscot Experimental Forest and 11 in the AL Talladega National Forest. Task 1.3: Data coverage: The main challenge in data coverage is ensuring the fidelity of collected geospatial data by integrated hardware (UAV and Backpack) and other geospatial data acquisition systems (e.g., Linear/Geiger-mode LiDAR systems on crewed aircraft). LiDAR-based Trajectory Enhancement and Mapping (TEAM) functionalities were developed to improve Backpack trajectory, which is compromised by GNSS-signal occlusions by tree canopies. Forest Feature Simultaneous Localization and Mapping (F2- SLAM) and Integrated Scan Simultaneous Trajectory Enhancement and Mapping (S2- TEAM) strategies were developed to improve Backpack data quality. This allows for integration of UAV/airborne LiDAR data, existing Digital Terrain Model and publicly available point cloud data (e.g., 3DEP) to improve the georeferencing quality of Backpack LiDAR data and ensure seamless transition between proximal, near proximal and remote aircraft data. Tasks 2.1-2.3: Leaf-scale measurement modeling and multi-model (FVS, LANDIS-II, CBM) intercomparison activities were initiated using Maine data as a pilot case. We developed an initial catalog of broad-scale simulation scenarios and also a framework for stakeholder-driven scenario development. Task 2.4 Data Visualization: A Spatio-Temporal Asset Catalog (STAC) was created and deployed at https://stac.digitalforestry.org to collect, organize and host LiDAR, imagery and GIS datasets with Purdue/IN geospatial data and geospatial data from UME and UGA. Purdue hosts UAV datasets using the Data to Science (D2S) platform (https://perseus.d2s.org) specifically designed for uploading, processing, visualizing and sharing UAV data using cloud optimized formats. In support of modeling efforts, broad-scale southern GIS databases (streams, roads, soils and some landowner parcels) are being acquired, developed and cataloged. Approximately 44TB of initial datasets have been, or are being, collected for ingestion. Task 3.1: Stakeholder Perceptions: To identify stakeholder needs, we initiated a compilation of prior surveys of forest business owners, forestry professionals and landowners on digital forestry topics. We outlined sampling frames for state-level data collection specifically concerning technology use and landowner typologies. We adopted the Delphi method for the purpose of generating in-depth insights and establishing a consensus opinion by prioritizing technology needs of forestry professionals in the Eastern U.S. Delphi involves rounds of data collection and analysis that bring together "a panel of experts, having them complete a series of questionnaires individually, and sharing these anonymized answers within the panel to allow for feedback and debate. The experts are presented with aggregated summaries of responses after each round, allowing each expert to adjust their assessment of priorities according to the group perspectives". Preliminary Round 1 analyses revealed the need for data and tools that would help inform forest management decisions including planning, accounting, and forecasting potential issues. Key factors that could influence the adoption of digital technologies include usability, accuracy, cost, organizational capacity, complexity, comparative advantage and access to the tools. Task 3.2: Scenario development: Prior work in ME, GA and IN was related to forest products and ecosystem services scenario development. We assessed the UME model and what-if scenario framework (previously developed and published) for adaptation to GA and IN. Informal discussions have begun with stakeholders (IHA, CFRU, lumber associations) to host regional or cross-regional workshops for input on scenario development. Task 4.1 Learning communities: We assessed the Learning Community infrastructure at each institution to explore the feasibility of a university-level PERSEUS-designated community. A determination was made to focus on a cross-partner hybrid Learning Community consisting of an undergraduate student cohort experience and a graduate student / postdoc professional development community. PERSEUS is leveraging existing UGA and Purdue Vertically Integrated Projects (VIP) and Purdue DataMine learning communities and internship projects. The first round will begin in Summer 2024; adopting and following best practices from other successful undergraduate student programs (e.g., NSF REU). Task 4.2 Interns and fellows: An undergraduate student experience is under development that will encompass a cohort experiential learning framework. Five undergraduate students have been supported at UGA and Purdue. Task 4.3 Curriculum development: "Enhanced Forest Inventory and Analysis" introduced 7 UME students in forestry and natural resources to the era of Big Data and its applications, with a particular focus on the scaling linkages, via state-of-the-art modeling approaches, between the detailed measurements collected in the field and proximal remote sensing to broader scale mapping tools using remote sensing technologies (airborne LiDAR and satellite multispectral imaging). The hybrid course is intended for traditional graduate students and postgraduates from the professional workforce. Purdue is developing an online Master of Forestry in Digital Natural Resources degree that will target working professionals to add skillsets focused on data acquisition, analysis and application using next-generation approaches including UAVs, environmental sensor networks, LiDAR remote sensing, multispectral imaging and photogrammetry. Industry and agency stakeholders indicate broad support and ample demand for this program. The proposal was submitted for evaluation and approval by the Purdue administration in Spring 2024. Elements of this program will be tested during the 2024 PERSEUS annual meeting through workshops providing instruction and hands-on experience to PERSEUS students, postdoctoral researchers and area stakeholders. Task 4.4 Online certificate: OLAF (Online Learning in Applied Forestry) courses on remote sensing and artificial intelligence are being developed by forestry and AI collaborators at UGA. Purdue and UME colleagues will provide feedback. Continuing education credits will be certified by the Society of American Foresters.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Li, B., Klein, J., Michels, D. L., Pirk, S., Benes, B., Palubicki, W. (2023). Rhizomorph: The Coordinated Function of Shoots and Roots. ACM Transaction on Graphics, 42(4). DOI: 10.1145/3592145
- Type:
Journal Articles
Status:
Published
Year Published:
2024
Citation:
Roy, S., Wei, X., Weiskittel, A., Hayes, D.J., Nelson, P., Contosta, A. 2024. Influence of climate zone shifts on forest ecosystems in northeastern United States and maritime Canada. Ecological Indicators 160: 111921. DOI: 10.1016/j.ecolind.2024.111921
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Shao, Jinyuan & Habib, Ayman, Fei, S. (2023). Semantic Segmentation of UAV Lidar Data for Tree Plantations. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XLVIII-1/W2-2023. 1901-1906. 10.5194/isprs-archives-XLVIII-1-W2-2023-1901-2023. DOI:10.5194/isprs-archives-XLVIII-1-W2-2023-1901-2023.
- Type:
Websites
Status:
Published
Year Published:
2023
Citation:
PERSEUS web: https://ag.purdue.edu/digital-forestry/projects/perseus/index.html
- Type:
Websites
Status:
Published
Year Published:
2023
Citation:
Spatio-Temporal Asset Catalog (STAC): https://stac.digitalforestry.org
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Cordonnier, G., Jouvet, G., Peytavie, A., Braun, J., Cani, M.-P., Benes, B., Galin, E., Gu�rin, E., Gain, J. (2023). Forming Terrains by Glacial Erosion. ACM Transaction on Graphics, 42(4). DOI: 10.1145/3592422
- Type:
Journal Articles
Status:
Published
Year Published:
2024
Citation:
Lee, Jae Joong, Li, Bosheng, Benes, Bedrich. (2024) Latent L-systems: Transformer-based Tree Generator. ACM Transactions on Graphics. 43(102): pp 116. DOI: 10.1145/3627101.
|