Submission information
Submission Number: 29
Submission ID: 46
Submission UUID: 204d026e-0a8e-46df-aa49-df06742cc54a
Submission URI: /form/project
Created: Tue, 09/03/2019 - 14:13
Completed: Tue, 09/03/2019 - 14:15
Changed: Thu, 05/05/2022 - 03:56
Remote IP address: 130.215.55.243
Submitted by: Kasey Legaard
Language: English
Is draft: No
Webform: Project
| Project Title | Using Genetic Algorithms and Support Vector Machines in Forest Mapping |
|---|---|
| Program | Northeast |
| Project Image |
|
| Tags | big-data (4), compiling (212), data-management (260), machine-learning (272), matlab (2), neural-networks (435), openstack (208), parallelization (223), programming (5), workflow (365) |
| Status | Complete |
| Project Leader | Kasey Legaard |
| kasey.legaard@maine.edu | |
| Mobile Phone | |
| Work Phone | |
| Mentor(s) | Chris Wilson, Larry Whitsel |
| Student-facilitator(s) | Noah Howard |
| Mentee(s) | |
| Project Description | Satellite-derived maps of forest conditions play diverse roles in research and resource management. Maps provide a basis for planning and executing field studies, developing and calibrating models, quantifying ecosystem processes or services, and evaluating environmental change. Natural resource managers use maps to characterize resource conditions, project changes, and direct management actions. However, inferences and decisions must be made within the context of map error, and methods used to produce maps generally result in patterns of error that are potentially detrimental. The researcher has developed machine learning techniques that effectively reduce undesirable systematic error when mapping forest attributes from satellite imagery and geospatial data. The approach is based on support vector machines using a multi-objective genetic algorithm (GA) designed to simultaneously minimize both total and systematic error. Using this approach, we have mapped tree species abundance and forest disturbance across northern Maine, obtaining outcomes that compare well against other approaches previously applied either regionally or nationally. Our algorithms are, however, computationally demanding, and large-scale applications will require more effective use of computing resources. The purpose of this project is to develop software that enables statewide and regional application of our algorithms through enhanced parallelization and new approaches to coordinate and accelerate the convergence of GAs operating across multiple, geographically distributed problem sets defined by input spatial data. We will specifically develop techniques for sharing prospective solutions between GAs executing on adjacent spatial data tiles, mimicking migration of individuals between locally adaptive subpopulations. Algorithm improvements will be coupled with more efficient and more automated methods for input and output data handling. The primary project outcome will be software that supports locally adaptive mapping of forest resources and environmental conditions across large spatial scales through innovative algorithms and efficient use of available cyberinfrastructure. |
| Project Deliverables | |
| Project Deliverables | |
| Student Research Computing Facilitator Profile | An undergraduate student with familiarity with Matlab, C, and High Performance Computing |
| Mentee Research Computing Profile | |
| Student Facilitator Programming Skill Level | |
| Mentee Programming Skill Level | |
| Project Institution | University of Maine |
| Project Address | 244 Neville Hall University of Maine Orono, Maine. 04469 |
| Anchor Institution | NE-University of Maine |
| Preferred Start Date | 11/10/2017 |
| Start as soon as possible. | No |
| Project Urgency | Already behind3Start date is flexible |
| Expected Project Duration (in months) | |
| Launch Presentation | |
| Launch Presentation Date | |
| Wrap Presentation | |
| Wrap Presentation Date | |
| Project Milestones | |
| Github Contributions | |
| Planned Portal Contributions (if any) | |
| Planned Publications (if any) | |
| What will the student learn? | The student will gain familiarity with techniques including genetic algorithms, support vector machines, and Matlab and C programming. |
| What will the mentee learn? | |
| What will the Cyberteam program learn from this project? | |
| HPC resources needed to complete this project? | |
| Notes | |
| What is the impact on the development of the principal discipline(s) of the project? | This work has had significant and on-going impact in the use of satellite imagery for forest mapping. This has led to a number of funded grants and industry collaborations. |
| What is the impact on other disciplines? | This work has advanced the use of Support Vector Machines and Genetic Algorithms in the context of identifying clouds in satellite images. |
| Is there an impact physical resources that form infrastructure? | |
| Is there an impact on the development of human resources for research computing? | In addition to the CI Team student, numerous other students have contributed to this work. |
| Is there an impact on institutional resources that form infrastructure? | |
| Is there an impact on information resources that form infrastructure? | |
| Is there an impact on technology transfer? | |
| Is there an impact on society beyond science and technology? | |
| Lessons Learned | We established our CI TEAM project (“Using Genetic Algorithms and Support Vector Machines in Forest Mapping”) having developed machine learning methods to improve upon the error characteristics of satellite-derived forest maps, but lacking the expertise required to implement those methods within a high performance computing environment. CI TEAM students developed parallelized code needed to operationalize our ML methods and scale forest remote sensing applications beyond limited study areas. Students further collaborated with us to develop components of a cloud-based satellite image processing and data handling system that fundamentally shifted the trajectory of our applications through more efficient processing and more effective user interfaces. Our CI TEAM experience cemented relationships with institutional computing partners and imparted practical experience working with and mentoring undergraduate computer science students. With that experience, we have since supported a senior capstone project engaging a team of five students, as well as a high school internship. We also have pending an NSF REU request, which would contribute further to the software development initiated under our CI TEAM project. Software originating with CI TEAM student contributions has been integral to research grant awards in excess of $1.5M (NSF, NASA, and USDA programs) and a newly established cooperative agreement to produce land cover and forest carbon maps for the state of Maine. |
| Overall results |