Project : USDA ARS

ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Sustainable Agricultural Systems Laboratory » Research » Research Project #439415

Research Project: Precision Integrated Weed Management in Conventional and Organic Crop Production Systems

Location: Sustainable Agricultural Systems Laboratory

2022 Annual Report

Objectives
Objective 1: Define theoretical, mechanistic, and applied underpinnings of weed management tactics in cover crop-based agronomic crop production systems to develop precision integrated weed management technologies. [NP304, Component 2, Problem Statement 2A&C] • Sub-objective 1.A. Determine optimal interactions between weed management tactics (physical, chemical, and biological) to address herbicide-resistant weeds and weed-crop competition. • Sub-objective 1.B. Develop cover crop and weed detection, identification, and mapping tools to assess the success of IWM systems and inform autonomous robotic weed control and decision support tools using machine vision technologies coupled with artificial intelligence (AI) and machine learning (ML).

Approach
To examine the interactions between cover crops and herbicides on weed-crop competition and weed suppression, we will quantify the interactive effect of cereal rye root and shoot residues with S-metolachlor rate on four weed species in the greenhouse. This will be conducted at Beltsville, Maryland and Urbana-Champaign. Next we will conduct field experiments that determine if cover crop management timing under field conditions influences the interactive effect between cereal rye and S-metolachlor, and whether the effect is due to physical (shoots) or chemical (roots via allelochemicals) mechanisms. This will be accomplished by establishing cover crop termination timing gradients and establishing a herbicide dose-response. The implications of management tactics will be further explored within a long-term cropping systems experiment. Here we will test the impact that harvest-time weed seed control, herbicides, and cover crops have on weed population dynamics and management genotypes in soybean by monitoring emergence, growth, survivorship, and fecundity of targeted summer annual weeds. To assess the performance of our IWM management systems at the field-scale, we will quantify the interaction between climate, soil, and cover crop management on cover crop performance and resulting weed suppression in an existing on-farm network across the U.S. Such an approach requires sensing technology to capture both the spatial distribution and performance. Therefore, we will develop digital weed image libraries from the on-farm network, annotate images, train machine learning models for image recognition, and test these algorithms with autonomous robots for weed control. Further, we will provide this information to growers through decision support tools.

Progress Report
This is the second year of this project. Five of the seven milestones were fully met and two were partially met due to restrictions on greenhouse usage during the pandemic. Over the past year, seventeen peer-reviewed manuscripts were submitted. Of these manuscripts, fifteen were accepted for publication while the remaining manuscripts are in review or have been revised and resubmitted. The manuscripts presented results from projects on cover crop-based corn and soybean production, cover crop management for optimal weed and nitrogen management, and multi-tactic integrated weed management solutions that combine cover crop, herbicides, reduced-tillage, and harvest weed seed control management. Furthermore, we have developed an autonomous robotic platform for 1) high throughput plant phenotyping and 2) building a National Agronomic Plant Image Repository. We have improved the ability to estimate cover crop performance with satellite imagery. Published research from this team also includes advancement in the use of computer vision and machine learning for monitoring crop and weed identification. Critical research was published on improving a cover crop decomposition and nitrogen release model. This model improvement and the redesign of a cover crop nitrogen calculator (CC-NCALC) were released this year. For Sub-objective 1A, field trials were initiated to quantify synergistic interactions between herbicides and cover crops for weed suppression and all samples were collected. The eighth year of the Lower Chesapeake Bay-Long-Term Agricultural Research multi-tactic weed management cropping systems experiment, initiated with past Area-Wide funds, was maintained and all samples collected. Data collection on weed populations and seedbank dynamics in two long-term cropping system experiments was completed and the data analyzed. For Sub-objective 1B, the on-farm monitoring of cover crop performance and subsequent impact on weed suppression is in its second year. Several hand-held and tractor-mounted sensing technologies were developed for monitoring cover crop performance, weed species identification, and biomass estimation. Computer vision and machine learning have promise for replacing destructive sampling (biomass). The camera systems used for building a weed image repository as well as 3-dimensional reconstruction of plant biomass have been tested on a greenhouse semi-autonomous robotic system (BenchBot). BenchBot is being calibrated and cyberinfrastructure necessary for data acquisition and management is under development. The technology coupled with an OAK-D camera from Luxonis Holding Corp. took third place in a North American Open Computer Vision competition. New Area-Wide funds were secured to expand the GROW (Getting Rid of Weeds) team for an additional six years. Outreach efforts have expanded through the website and decision tools. We have also initiated a national on-farm multi-tactic weed management project examining harvest weed seed control and cover crops. The study will examine the independent and combined effects of these tactics on herbicide-resistant weed emergence, seed production, and long-term persistence.

Accomplishments
1. Low-cost data collection system developed for large-scale on-farm soil water monitoring. Collection of on-farm research data can be expensive due to the number of data loggers required for spatially distributed locations. Scientists at USDA-ARS facilities in Beltsville, Maryland; Bushland, Texas; and North Carolina State University, Raleigh, North Carolina, used open-source microcontroller technologies to develop a low-cost gateway-node data logging system for collecting soil water data. Incorporation of cellular technology facilities transfer data in near real-time to the cloud for ready access by researchers. The low-cost and flexible design of the system resulted in interest from a soil moisture equipment manufacturer and an irrigation equipment manufacturer who are making the technology available at lower costs compared to other systems. The system has been deployed on a total of 230 farms including 89 farms in 20 states in 2021 as part of the Precision Sustainable Agriculture network on-farm research (120+ scientists in 25 states). This work is important as it lowers the cost to researchers for collecting water sensor data on-farm, thus enabling more robust data collection and potential to capture spatial soil water patterns on-farm.

2. Modern satellites improve the range of cover crop biomass estimates. Cover crops are the premier climate smart farming practice; they are important for adapting to and mitigating against climate change, building soil health, and providing weed and nitrogen management benefits to growers. However, the performance of these biological tools varies due to the spatial variability of soil fertility and drainage within a given field. Technologies that map cover crops will facilitate variable-rate technologies that apply inputs based on cover crop performance within a field, thus enabling the merging of precision and sustainable agriculture. ARS and USGS scientists in Beltsville, Maryland, conducted a three-year observational study that determined the ability of two modern satellites to predict cover crop biomass at 20-meter spatial resolution. A modern spectral index using the “red-edge” spectral region improved prediction of cover crop biomass from 1,500 kg ha-1 to 1,900 kg ha-1. This accomplishment is an important step toward increasing our ability to estimate cover crop biomass and will aid in mapping cover crops to facilitate precision application of inputs (e.g., herbicides, fertilizer) that correspond with their spatial performance in a field.

3. Satellites can be used to estimate timing of cover crop termination. Cost-share programs in Maryland and Delaware distribute incentive payments for planting cover crops. To distribute payments, state or country agents verify that termination occurred by physically visiting enrolled fields. To reduce the amount of work on these agents, ARS and USGS scientists in Beltsville, Maryland, downloaded, processed, and extracted a timeseries of the normalized difference vegetation index for each enrolled field. Using a within season termination algorithm developed by the ARS and USGS scientists, ~84% of reported cover crop termination dates can be predicted within a 1.5-week margin of error before and after termination. Accurately estimating cover crop termination date is increasingly important since Chesapeake Bay states began increasing incentives for later termination of cover crops (after May 1). Satellite imagery can provide a fast and consistent approach for generating cover crop termination maps over large areas for verification of practice to federal and private cost-share programs.

4. BenchBot: A low-cost autonomous robot for high-throughput phenotyping and building of image repositories. High-throughput phenotyping systems for greenhouses and semi-field conditions are critical for plant breeding and building image repositories for mapping cash crops, cover crops, and weeds with computer vision and artificial intelligence. However, these high-throughput phenotyping systems are costly, resulting in limited use. ARS scientists in Beltsville, Maryland, designed and built the fully autonomous robotic platform, BenchBot in collaboration with North Carolina State University researchers. BenchBot costs under $20,000, significantly less than most commercial and research grade systems which cost in the millions of dollars. Designs for BenchBot are published on GitHub and are being used for high-throughput phenotyping by ARS and university scientists at North Carolina State University and Texas A&M for building a national agronomic plant image repository. BenchBot is a low-cost, easy to use technology that is making high throughput phenotyping accessible for researchers.

U.S. DEPARTMENT OF AGRICULTURE

Sustainable Agricultural Systems Laboratory: Beltsville, MD