Source: UNIV OF MINNESOTA submitted to
REMOTE SENSING BIOLOGICAL INVASIONS: DETECTING AND MONITORING LEAFY SPURGE POPULATIONS USING HIGH RESOLUTION IMAGERY AND DEEP LEARNING
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
ACTIVE
Funding Source
Reporting Frequency
Annual
Accession No.
1030114
Grant No.
2023-67013-39894
Cumulative Award Amt.
$749,532.00
Proposal No.
2022-10123
Multistate No.
(N/A)
Project Start Date
May 1, 2023
Project End Date
Apr 30, 2026
Grant Year
2023
Program Code
[A1112]- Pests and Beneficial Species in Agricultural Production Systems
Project Director
Briscoe Runquist, R.
Recipient Organization
UNIV OF MINNESOTA
(N/A)
ST PAUL,MN 55108
Performing Department
Plant and Microbial Biology
Non Technical Summary
Invasive species are reshaping landscapes globally, causing harm to agricultural and ecological systems, and generating billions of dollars in economic losses annually. Among the many habitats threatened by invasive species, the untilled rangelands of the Northern Great Plains have been severely impacted. These habitats contribute substantially to agricultural production and also contribute valuable ecosystem services essential for healthy soil and clean water. Leafy spurge is a major weed of rangelands that cause enormous economic and environmental losses. It infests almost 2 million hectares that results in over $130 million dollars annually.The goal of our project is to leverage high resolution satellite imagery and deep learning models to detect leafy spurge populations, track population dynamics over the last decade, and assess the effects of environmental variation and biocontrol efforts on spatial patterns of invasion. This research will allow for cost-effective, rapid tracking of the spread and impact of this noxious weed. Our project will also reshape our understanding of the ecological and management factors that control invasion dynamics over large geographic scales. We will use extensive datasets of occurrences and management practices already collected by multiple government agencies over the last decade across the Northern Great Plains. This work has the ability to aid land management efforts with near real-time estimates of invasion and recommendations for likely success of biocontrol as a management tool. Further, demonstrating the use of remote sensing and deep learning in rangeland management will allow for more targeted and accurate management of leafy spurge and eventually other noxious invasive weeds
Animal Health Component
(N/A)
Research Effort Categories
Basic
100%
Applied
(N/A)
Developmental
(N/A)
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
2130799107070%
2132300208030%
Goals / Objectives
?We propose to leverage satellite imagery to detect and monitor leafy spurge populations across a large fraction of the Northern Great Plains. This research will allow for cost-effective, rapid tracking of the spread and impact of this noxious weed. Our project will also reshape our understanding of the ecological and management factors that control invasion dynamics over geographic scales. The main goals of this project are to use high-resolution satellite imagery and deep learning models to (1) detect leafy spurge populations, especially in emerging areas of invasion, (2) track population dynamics over the last decade to develop better predictive models of range expansion, and (3) to determine the causes of biocontrol success vs. failure to guide immediate management.Objectives:Objective 1: Detect leafy spurge populations across the Great Plains from satellite images using deep learning:1A. We will use existing data from agencies to assess the transferability of our Minnesota-based leafy spurge model and determine the strengths and pitfalls for transference of previously built models.1B. Develop new robust models to detect populations across geographic regions that differ in habitat characteristics.Objective 2: Track population dynamics over the last decade and predict range infilling and range expansion:2A. We will use satellite imagery to track population growth/decline over a decade across the Northern Great Plains.2B. Use time-series data of leafy spurge population growth to develop demographic species distribution models (SDMs). These SDMs will identify areas at risk of invasion or that require greater management.Objective 3: Identify environmental factors that modulate biocontrol success to inform management strategies:3A. We will determine where and under which environmental conditions biocontrol efforts have succeeded in reducing population growth across the range.3B. We will use the analyses of environmental determinants of biocontrol to provide agencies with guidance on where to best implement biocontrol versus other management strategies
Project Methods
Efforts: The project will be conducted in a number of stages:Dataset construction of high-resolution satellite imagery across multiple years.Build deep learning models that can detect leafy spurge populations and can track changes in population size over timeUse data on leafy spurge population demography to build species distribution models that determine what environmental factors influence establishment, growth, and persistence of populations.Determine environmental factors that modulate biocontrol success or failure in leafy spurge management.Evaluation: Evaluation will occur internally in the project via statistical validation of models prior to passing data to be used to accomplish the next objective in the project. Deep learning models and species distribution models should achieve leafy spurge detection rates > 80-85% in independent validation datasets to be used in subsequent objectives.Externally, validation will occur via peer-review of models, dataproducts, and results for journal publications and management recommendation documents

Progress 05/01/23 to 04/30/24

Outputs
Target Audience:Our work is of interest to professional land managers and extension agents as well as researchers also working in the field of remote sensing. In the past year of out reseach project, we have reached out to our partnership stakeholders who support our project. We have also reached out to others within the University community - including researchers and extension professors through meetings andseminars. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?In the past year, this grant has supported work by faculty, a staff scientist, a graduate student, and an undergraduate researcher. How have the results been disseminated to communities of interest?Preliminary results for the new deep learning model for leafy spurge detection and preliminary demographic data were shared in a public dissertation defense seminar for the supported graduate student. We are currently preparing a manuscript for publication about the deep learning model and demographic data for leafy spurge in Minnesota. What do you plan to do during the next reporting period to accomplish the goals?We are planning our upcoming field season to generate additional testing, training, and validation data. We are also continuing to explore new sources of imagery and deep learning architectures to build more efficient and accurate deep learning models. We will also continue to reach out to partners in the natural resources community who we have been working with on this project to gather datasets on the outcomes of biological control projects in order to further our work on the last objective of our project.

Impacts
What was accomplished under these goals? We have made substantial progress on the goals outlined in objectives 1 and 2 over the past year. In previous work, we developed a deep learning model to detect leafy spurge based on a complex landscape in Minnesota. We used data from North Dakota to assess the transferability of this previously built model and determined that it was able to detect leafy spurge but would require some tweaking to be fully implementable in the management space. We also built a new temporal-CNN deep learning model using Landsat imagery from across the Northern Great Plains. The preliminary model for the region is promising and we are continuing to refine the model and explore other architectures that may provide greater model sensitivity and precision for leafy spurge detection, especially in more complex habitat types. We also built a Landsat based model to detect leafy spurge across Minnesota. The state-based model performed well and we used our model to generate predictions about population growth in Minnesota. We are using this preliminary growth data to start developing model frameworks for demographically-based SDMs. We are in the process of planning the field season to collect more validation data from additional regions of the Northern Great Plains to assess the true accuracy of our models. We are also currently preparing a manuscript to publish our new deep learning leafy spurge detection model for Minnesota.

Publications