Recipient Organization
UNIVERSITY OF FLORIDA
G022 MCCARTY HALL
GAINESVILLE,FL 32611
Performing Department
(N/A)
Non Technical Summary
Growers across the southeast rely predominately on broadcast herbicide applications, plastic mulches, and fumigants for weed control. Weed management tools are utilized but rarely integrated into long-term management programs. The recent development of smart spray technology at the University of Florida can facilitate the adoption of IPM principals by providing the technology necessary for growers to efficiently monitor weed populations over time and apply pre-and post-emergence herbicides only when and where they are needed. The overall goal of this project is to develop IWM programs for vegetables using tomato as the test scenario. To achieve this goal we will utilize video imagery collected during field operations to map weed populations, monitor population change over time, focus management efforts, and evaluate IWM effectiveness and compare commercial practices with precision IWM programs that rely on smart spray technology for pest monitoring and precision herbicide applications. This approach should improve weed management over time and reduce herbicide and labor inputs. We are confident the proposed research will lead to increased adoption of IPM in plasticulture vegetables.
Animal Health Component
90%
Research Effort Categories
Basic
10%
Applied
90%
Developmental
(N/A)
Goals / Objectives
Goal1: Utilize video imagery collected during field operations to map weed populations, monitor population change over time, focus management efforts, and evaluate IPM program effectiveness.Goal2: Compare commercial weed management programs with precision integrated weed management programs that rely on smart spray technology.Goal 3: Evaluate smart spray technology for pre- and post-emergence herbicide applications.Goal 4: Utilize smart spray technology to promote the adoption of IWM programs.
Project Methods
OBJECTIVES 1 AND 2Experimental Design. An experiment will be conducted for multiple years at the Gulf Coast Research and Education Center (GCREC) to achieve objectives 1 and 2. We are requesting funding for two years, but we hope to obtain funding to run the project for subsequent years with the goal of achieving complete autonomous weed control in the precision plots. The experiment will be set up as a Randomized Complete Block Design with three treatments and five blocks. Only one experiment will be conducted due to the size of the proposed experiment, the cost associated with management of a trial of this size, the labor intensity required to manage a trial this size, and the data intensity to be collected within the trial. The treatments will beNontreated Control:Commercial Production Standard: In addition to fumigants and plastic mulches weed management within this treatment will include broadcast glyphosate applications during the fallow period, banded pre-emergence applications of metribuzin applied under the plastic mulch for broadleaf and grass control, post-emergence banded application of halosulfuron for post-transplant nutsedge control on the bed, and banded applications of flumioxazin and paraquat applications in the row middle.Precision Weed Management: In addition to fumigants and plastic mulches this treatment will include precision glyphosate applications using smart spray technology during the fallow period, precision preemergence applications of metribuzin only where holes are punched for transplant in the plastic mulch using a hole punch applicator (Figure 3), precision post-emergence applications of halosulfuron only where nutsedges occur on the beds (Figure 6), banded application of flumioxazin in the row middles followed by precision applications of paraquat, halosulfuron, or clethodim only where weeds occur in the row middles (Figure 6). The same treatments will be applied to the same plots in all years.The primary purpose of this experiment will be to compare conventional weed management programs with programs that incorporate precision technology into all management components.Tomato crop production will begin with fumigation in January. The industry standard fumigant (300 lbs of Pic-Clor 60) will be used. Beds will be shaped, fumigated, double drip tapes installed, and covered with TIF plastic mulch. Each bed will be 8 inches tall, 32 inches at the base and 5 feet between beds. If nutsedge punctures the plastic mulch it will be burned back with banded or precision applications of paraquat prior to transplant. Preemergence metribuzin will be applied to the bed-top in transplant holes immediately after crop transplant. Halosulfuron will be banded following transplant to control emerged nutsedge on the bed. Row middle herbicides will be banded or broadcast as needed. Herbicide applications in conventional and precision plots will be applied with the same equipment with the smart spray technology turned on or off depending on the treatment. Tomato transplants will be installed one month after fumigation at two feet spacing between transplants.Data Collection. Video imagery will be collected from bed tops and in row middles and used to map weed populations at every spray operation during the fallow period. Within the crop, video imagery will be collected prior to transplant (first row middle application), at transplant (second row middle application and hole punch), 4 weeks after transplant to monitor weeds emerging in the planting holes, and 8 weeks after transplant to monitor weed density. Maps will provide data on the weed density and location and will be compared within and across seasons to estimate patch persistence, changes in density, and overall weed control over time. The number of herbicide applications will be determined by weed density measurements following initial applications. GPS coordinates will be captured as NMEA strings and the ground speed will be parsed out with the time stamp and lat/long coordinates at the same time that the images were captured. In this manner, we can assure every image has a time stamp and is georeferenced. We will use a RTK type GPS for this application to map weeds with centimeter accuracy. If the distance offset from the GPS antenna and the camera position are known, then the georeferenced position of every weed identified in the images can be calculated.We will create individual maps for each data acquisition session. Although the maps should be co-registered due to using RTK GNSS in georeferencing, we will verify map alignments and fix misregistration discrepancies. Our goals are to: (1) identify existing weed patches, and 2) examine the spatiotemporal distribution of the clusters, 3) measure weed density and changes in density over time. Hotspot analysis and density-based clustering of weed occurrence will be performed at multiple spatial scales at the plot level (Ord and Getis, 1995; Muller-Warrant et al., 2008). The results should quantitatively identify the locations where weed infestations larger than the expected proportion of all field occurrences occur and test these results for statistical significance. The within- and across-season data will be analyzed temporally through time series clustering and hotspot analysis of spatiotemporal data cubes (Tolcha et al., 2019). The results of this analysis will identify several weed hotspot patterns for each weed type including, oscillating, persistent, and intensifying hotspots. The results of this analysis will also identify where and when weeds are persistent and the periods where their behavior changes over the season. Both machine learning techniques and regression analysis accounting for autocorrelation (Ahmed et al., 2017) will be used to study the effect of environmental variables on weed distribution across space and time. We expect this analysis to provide answers not only to where and when weed management techniques should be applied but also to integrate methods to control other environmental variables in the management process. Weed densities will also be verified by counting weeds by species within each geo-reference image at each herbicide application.?OBJECTIVE 3We will utilize the video imagery collected throughout the experiment to improve our deep learning models. Weed images will be extracted from the video images and weeds will be labelled using YOLO labelling software. The algorithms will be trained with the new images and evaluated with images not used in the training. We will utilize standard statistical methods to evaluate precision and recall. The overall goal will be to achieve greater than 95% precision and recall for all weed species or weed categories of interest under a wide array of conditions.OBJECTIVE 4The overall goal of this project is to apply 100% of the herbicides utilized in the precision plots with smart spray technology. To our knowledge, this has not been achieved in commercial vegetable production systems anywhere in the world. We will also be able to measure the amount of herbicide used, weed management achieved and overall estimated reduction in input costs. The research site will provide an ideal location to demonstrate the smart spray technology for growers and industry representatives and allow them to see commercial standard tomato production where herbicides are applied with smart spray technology. To achieve this goal we will organize a minimum of two field tours in year two to demonstrate the technology to growers. We will also present the results at field days and develop on-line videos to demonstrate the effectiveness of the technology for herbicide applications but also as a scouting tool. In the process of demonstrating the technology and how it can be used to manage weeds we will also be highlighting the importance of conducting year-round management programs for weed control as well as the importance of integrating multiple management programs.