Performing Department
(N/A)
Non Technical Summary
The accelerating evolution of herbicide resistance in numerous weed species threatens gains in soil health and carbon storage as farmers return to intensive tillage for weed management. Cover crops are a powerful tool for suppressing weeds in no-till soybean production and improving soil health. However, managing cover crops for weed suppression is inherently complicated by high spatial variability in cover crop performance and weed populations. The goal of this project is to aid farmers in their use of cover crops, at the production scale, as part of an integrated weed management program. We will develop sensing technology for mapping and monitoring cover crop and weed species and biomass that assists soybean farmers in planning and decision making. PlantMap3D, an integrated hardware and software solution for plant species, density, and biomass mapping, will be deployed in five major soybean producing states. We will collect data on the impact of cereal rye cover crop biomass on weed populations and elucidate how field productivity zones influence weed suppression with cover crops. To accelerate applicability of the technology, we will iteratively improve the PlantMap3D system for production use by conducting user-testing with farmer partners and additional stakeholders on their user experience. This project addresses NIFA and USB priorities by developing low-cost precision tools to provide information for optimizing cover crop management for weed suppression and soil health and is submitted in response to USB commodity topic #12. This proposal is submitted in response to a single specific commodity board topic listed above.
Animal Health Component
50%
Research Effort Categories
Basic
20%
Applied
50%
Developmental
30%
Goals / Objectives
Long-Term Goals. Our long-term project goal is to develop cover crop-based no-till production systems that conserve and build soils, mitigate and adapt to climate change, and minimize the impact of herbicide-resistant weeds. This will be accomplished through precision integrated weed management using PlantMap3D, an integrated hardware and software solution for mapping cash crops, cover crops, and weeds species and biomass on commercial sprayers.ObjectivesRegionalize PlantMap3D across soybean producing regions of the US to ensure optimal species and biomass mapping.Determine camera requirements for height, spacing, and speed of tractor to ensure they are compatible with the range of commercial sprayers in operation. Define within-field summer weed suppression potential from cereal rye cover crops in long-term soybean production systems with a novel plant mapping technology.Conduct spatial analysis to estimate the role of cereal rye biomass alone in explaining weed suppression.Quantify weed species-specific responses to cereal rye biomass.Quantify how within-field spatial variation in historical crop productivity influences present day cover crop-weed interactions.Leverage farmer partners, mapping technology user-testers, and state soybean commodity boards to refine PlantMap3D hardware and software design.Gather feedback on the user experience of installing and operating the PlantMap3D mapping system.Design a decision support tool that aids growers in making cover crop and herbicide management decisions.
Project Methods
PlantMap3D was originally developed at USDA-ARS Beltsville Agricultural Research Center (Maryland) and North Carolina State University to be used for hand-held applications and on small-scale spray applicators. The next phase of development requires adapting PlantMap3D to operate on our partners' commercial scale applicators, which can have a multitude of boom configurations, variable heights, and different vibration patterns. Objective 1 focuses on fine-tuning the software and hardware requirements of PlantMap3D to be configurable to a more diverse range of applicators while ensuring that the image quality requirements for semantic segmentation are still met. Our team has already developed a high-throughput process for assessing image quality requirements for semantic segmentation, an image quality requirement higher than for what is required of the stereo images. Our computer vision specialists will work closely with each on-farm cooperator conducting on-farm cover crop trials in five states (Objective 2) to examine how their sprayer, tractor, and standard spraying procedure interact with PlantMap3D hardware to affect the quality of the maps produced. In year one of the project, each experimental site will receive the PlantMap3D hardwarewhich will be installed on commercial grade herbicide spray booms by farmers, with assistance from local PSA and GROW members who have expert knowledge of the system. The most recent version can generate georeferenced maps of individual plant species, density, and biomass at a 6 m x 6 m pixel size. PlantMap3D generates an image (1 m2) in the centroid of the 36 m2 pixel to quantify plant species, density, and biomass. This pixel density requires one image collected per second on a tractor driving 19 km hr-1. This requirement was set based on the time it takes to run deep neural network applications and construct a disparity map on the edge (microcomputer in the OAK-D camera) and send those images to primary computers networked to each of the individual cameras (Jetson Orin). As we adapt and calibrate PlantMap3D for each commercial sprayer, we will attempt to increase image capture and processing; potentially getting to 1.5 images per second. While variable rate technology primarily is not varying across a spray boom width and does not require such fine resolution maps, the finer pixel resolution of cover crop and weed maps will arm farmers and consultants with finer resolution knowledge on the spatial variations that exist on their farms and fields. Spatial patterns of cover crop performance (biomass quality and quantity) and weed density and biomass will be quantified in production-scale field experiments carried out in five states (MD, NC, TX, IA, and OH) spanning diverse soybean producing environments of the US (Table 1). These measurements will be made following established protocols using the PlantMap3D system. Maps will be generated for a total of 192 fields during the four years of field trials, with each field being split into strips with and without cover crops. Each cover crop and no cover crop strip in a field will be a minimum size of 0.3 hectares. All fields used for the cover crop treatment will have a minimum of five years of cover cropping history. Beyond the cover crop treatments, all crop management (i.e., cover crop termination timing, soybean planting dates/rates/row spacing, herbicide programs, and nutrient management) will be identical within a given site-year. Production practices and inputs will be allowed to vary by site and region to best reflect local prevailing practices. The no cover crop strips will serve as control plots to assess baseline existing weed populations and their spatial distribution in a field. The no cover crop controls are key for quantifying the potential for weed suppression from cover crops and how spatial variability in weed populations interacts with cover crop performance.Mounting the PlantMap3D system on spray booms will allow data collection to occur simultaneously with herbicide applications, preventing the need for additional passes of equipment through soybean fields for the purposes of mapping. Maps of cereal rye cover crop biomass accumulation will be generated at termination. Maps of weed species, density, and biomass, as well as early-season soybean growth, will be generated during post-emergence herbicide applications.Sub-objective 2.3. Cash crop yield maps will be compiled from the previous five years of crop production for each farm field included in the cover crop strip trials. These historical yield maps will be used to define regions of field productivity driven by edaphic properties. Regions of relative high and low productivity within each field will be determined by normalizing yields to the field average yield within a single crop year. The mean and standard deviation of these normalized yield data will then be calculated across all years where data are available to create maps of yield productivity and stability over time. When evaluating cover crop weed suppression outcomes, field productivity zones will serve as a proxy indicator of sub-field level variation in nutrient availability and other difficult to measure environment/management factors that are known to influence cover crop performance and weed population dynamics. After maps of field productivity are generated for each site, they will be stored in the same database as the PlantMap3D data. The data and analysis generated from this sub-objective will serve as empirical models that willbe integrated into a decision support tool that uses field productivity and cereal rye biomass distributions in a field to inform an overall integrated weed management PlantMap3D data.Objective threewill be carried out as a series of facilitated focus groups, stakeholder interviews, and surveys that will gather information at two important phases of the design and development process. First, user testing sessions will be conducted with stakeholders to understand end-user priorities for PlantMap3D features and functionality which will determine design objectives for the user-interface for the technology. Second, quality assurance testing will be conducted with end-users to determine whether design objectives have been achieved for both the hardware and software components of the system. This user testing process will follow an iterative development cycle of visioning, user testing, product development, quality testing, and refinement that will be repeated several times throughout the implementation of this project. Feedback on the user experience of installing and operating the PlantMap3D system will be gathered from stakeholders directly involved in this project that include technical professionals who will train farmers to operate the system, and the farmers themselves. These data will be collected via surveys and interviews of farmers during and after system installation and field testing (Objective 1) and use of the system for cover crop and weed mapping during the growing season (Objective 2). Feedback about user interface features and functionality will be gathered from this pool of users gaining hands-on experience with the system and from a larger pool of potential end-users including additional farmers employing cover crops and precision agriculture technologies (e.g., farmer collaborators participating in the Iowa Soybean long-term cover cropping trials), extension agents and crop consultants that support soybean growers, and members of the state-level soybean checkoff programs in the five states where field trials will be conducted. User testing plans, including focus group activities and questions and survey instruments, will be developed and carried out by members of the PSA Social Science Team. Stakeholder interviews will be conducted as needed to follow up on ideas or issues identified in focus groups or surveys.