Recipient Organization
VISIMO LLC
400 MAIN ST
CORAOPOLIS,PA 151081629
Performing Department
(N/A)
Non Technical Summary
In the forestry industry, existing risk assessments are complex and time consuming, limiting the ability of forest management professionals to accurately and quickly perform risk assessments in real time. VISIMO proposes to develop a risk mitigation software application which uses a combination of Machine Learning (ML) and Subject-Matter Expertise (SME) to reduce environmental degradation during forestry operations. VISIMO's tool will predict the impact of specific forestry operations on user-specified landscapes using environmental variables related to soil, water, and residual stand quality and predict the immediate impact of land use decisions on the environment, providing mitigation strategies to reduce degradation in real time.VISIMO's proposed solution builds on our commercialized Job Safety Analysis (JSA) tool for the construction industry, which VISIMO also successfully adapted to predict risk on small and midsize farms. VISIMO's primary objective for Phase I is to prove that our established risk prediction model can be adapted to forestry while maintaining similar predictive accurary.Success in creating the proposed solution, and developing beyond Phases I and II, will allow VISIMO to supply foresters (both private and public), land managers, extension agents, and other forestry stakeholders with an efficient, intuitive tool to support tactical and strategic planning in land use decision-making. This will reduce environmental degredation through an increased ability to accurately and quickly know whether forestry operations adhere to Forestry Best Management Practices.
Animal Health Component
0%
Research Effort Categories
Basic
100%
Applied
0%
Developmental
0%
Goals / Objectives
VISIMO's tool will assess the current environmental condition of a region of interest, predict the immediate impact of land use decisions on the environment, and provide mitigation strategies to reduce degradation in real time. The solution will equip typically underserved users such as area foresters in public and small private organizations, as well as large private forestry management companies, with an intuitive, streamlined risk assessment tool that utilizes established scientific principles, ultimately reducing degradation and increasing forest productivity. VISIMO proposes to develop a risk mitigation software application which uses a combination of Machine Learning (ML) and Subject-Matter Expertise (SME) to reduce environmental degradation during forestry operations. VISIMO's tool will predict the impact of specific forestry operations on user-specified landscapes using environmental variables related to soil, water, and residual stand quality and predict the immediate impact of land use decisions on the environment, providing mitigation strategies to reduce degradation in real time.Success in creating the proposed solution, and developing beyond Phases I and II, will allow VISIMO to supply foresters (both private and public), land managers, extension agents, and other forestry stakeholders with an efficient, intuitive tool to support tactical and strategic planning in land use decision-making. This will reduce environmental degredation through an increased ability to accurately and quickly know whether forestry operations adhere to Forestry Best Management Practices.Objectives:Finalize Research and Development PlanImplement Forestry AnalysisImplemnent GIS ComponentFinalize Proof of Concept
Project Methods
VISIMO will create a scientific assessment tool to measure the current environmental condition of a particular landscape, predict the immediate impact of forestry operations on the environment, and provide risk mitigation strategies to reduce environmental degradation in real time, in accordance with BMPs. VISIMO will develop a PoC in Phase I that includes a limited number of actions, risk variables, and risk mitigations following the variable structure of our JSA tool. VISIMO's risk modeling will be guided by forestry and ecological experts at UMaine. VISIMO's decision-support tool will provide analysis at three different levels, depending upon specific data types available for input.The first level of analysis (L1) will consider readily observable factors to calculate a risk score. For example, in the case of soil compaction, users will enter variables such as equipment model, weight, tire pressure, and the level of operational traffic for the landscape, each a potential cause of soil compaction (University of Minnesota, 2018). These inputs will be combined with an SME-based algorithm to generate a risk score, similar to the qualitative risk variable setup used in VISIMO's JSA and agriculture risk tools. This level of analysis is least demanding on the user (in terms of inputs and technical expertise), but also provides the least sophisticated assessment. It is optimal for an early, wide-reaching analysis, allowing for limited resources to be concentrated on riskier areas that will require more advanced levels of analysis.The second level of analysis (L2) involves combining inputs from tools measuring physical samples of soil and water (e.g., soil density and moisture). Physical soil samples enable more detailed use of the USLE as well as the Watershed Erosion Prediction Project (WEPP), two widely used equations/models for measuring the rate of soil erosion. The USLE is a proven formula for measuring soil erosion from water based on rainfall, runoff, soil erodibility, slope length gradient, crop/vegetation management, and existing support practices (Stone & Hillborn, 2012). VISIMO will investigate which elements of the USLE can apply to our modeling specific to forestry use cases. WEPP is a simulation model for describing disturbed forest erosion conditions by calculating the probability that a given level of erosion will occur (Elliot & Hall, 2010). For Phase I, the L1 and L2 analysis levels will be validated via existing datasets, such as the Web Soil Survey, LiDAR data, and stand inventory; where data is incomplete, we will work with our SME partners to create realistic synthetic data. By analyzing the collected real-world data for trends, a synthetic dataset can be created that has statistically similar relationships, correlations, and dependencies between variables. Once sufficient real-world data is collected, the synthetic data can be removed and the ML model can be trained entirely on real-world data, optimizing its performance and accuracy. VISIMO used this combined real and synthetic data approach to successfully convert our JSA tool to the agricultural sector through a USDA SBIR Phase II to reduce risks on farms.The third level of analysis (L3) will occur after the tool has collected enough real-world data to train the GNN model to predict the risks of a given operation. This also gives the tool the ability to record and model edge cases, allowing new foresters to generate risk assessments for rare situations that previously could only be learned through experience - experience that is being lost as older foresters leave the industry. VISIMO's JSA application was able to utilize MLmodeling after collecting roughly 20,000 observations. Due to the time required to gather this volume of data, L3 analysis via ML will occur in Phase II. We will train the model on data collected during the early stages of Phase II, a core component of the Phase II work plan. A GNN will be used for its ability to analyze complex nested relationships between variables, ideal for modeling environmental data. The model will more accurately produce risk scores for identified risks and will identify at-risk zones requiring more analysis, while requiring less work-intensive inputs.Three of the main environmental concerns being tracked--soil erosion, soil compaction, and water degradation (e.g., sedimentation, runoff, etc.)--do not require evaluation at the same analysis level. This flexibility will be useful for new or inexperienced users who may not have immediate access to the tools and data needed to perform complex analyses. A user may have all the necessary inputs to perform L2 analysis on soil erosion, but only have inputs for L1 analysis on soil compaction and water erosion. Regardless of a user's starting point, the decision-support tool will suggest mitigations and will become increasingly effective as more data is collected.These analyses will be supported with a GIS data structure, where the region of interest will be automatically broken into relevant subregions, which can be manually modified further. Subregions will be prefilled with available public data such as elevation and rainfall levels and will support data updates and exports as well as GIS visualization tools. This will all be made available through the PWA which can be used when an internet connection is not available. At the completion of Phase I, VISIMO will deploy and host the server-side components of the PWA in a cloud environment, allowing end users to download the PWA to gather and analyze data. This data will be used both for software testing and model training during the Phase II development process.