Source: UNIV OF WISCONSIN submitted to
COLDSNAP: AN ONLINE BUD COLD HARDINESS PREDICTION TOOL TO ASSIST GRAPEVINE MANAGEMENT DECISIONS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
ACTIVE
Funding Source
Reporting Frequency
Annual
Accession No.
1029870
Grant No.
2023-68008-39274
Cumulative Award Amt.
$300,000.00
Proposal No.
2022-10055
Multistate No.
(N/A)
Project Start Date
Feb 28, 2023
Project End Date
Feb 27, 2026
Grant Year
2023
Program Code
[A1701]- Critical Agricultural Research and Extension: CARE
Recipient Organization
UNIV OF WISCONSIN
21 N PARK ST STE 6401
MADISON,WI 53715-1218
Performing Department
(N/A)
Non Technical Summary
Extreme and erratic weather events, consequence of climate change, results in substantial cold damage to grapevines and extensive economic losses to rural grape growing areas in the Midwest and Northeastern U.S. Current methods for monitoring fluctuations in bud cold hardiness are a time-consuming and require specialized and expensive equipment. The goal of this project is to develop an online tool for grape growers that uses recently developed models to estimate bud cold hardiness throughout the dormant period from temperature data available through a wide weather network. This information will assist growers in cultivar and site selection for vineyard establishment, as well as making time sensitive decisions to protect grapevines from winter injury in established vineyards, thus minimizing crop and vine losses and increasing the long-term sustainability of the grape and wine industries in the Midwestern and eastern states. This project addresses AFRI's program priorities "Plant health and production and plant products" and "Agriculture Systems and Technology". This project proposes to enhance the economic sustainability of the Midwest and Northeast U.S. grape industry by minimizing crop and vine losses due to cold damage through: 1) validating and expanding two newly developed bud cold hardiness prediction models through analysis of regionally diverse data collected in the New York, Wisconsin, South Dakota, Minnesota, Michigan, Ohio, Pennsylvania, Iowa, and Quebec, 2) implementing model programming and development of the "ColdSnap Grape Bud Hardiness Model", producing a stakeholder accessibility-minded online tool that leverages customized climate data hosted by the Network for Environment and Weather Applications, and 3) developing educational resources for growers and provide extension-led outreach efforts to facilitate information dissemination and implementation.
Animal Health Component
70%
Research Effort Categories
Basic
(N/A)
Applied
70%
Developmental
30%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
20311311020100%
Goals / Objectives
The goal of this project is to develop an online tool for grape growers that uses recently developed models to estimate bud cold hardiness throughout the dormant period from temperature data available through a wide weather network. This information will assist growers in cultivar and site selection for vineyard establishment, as well as making time sensitive decisions to protect grapevines from winter injury in established vineyards, thus minimizing crop and vine losses and increasing the long-term sustainability of the grape and wine industries in the Midwestern and eastern states.This project proposes to enhance the economic sustainability of the Midwest and Northeast U.S. grape industry by minimizing crop and vine losses due to cold damage through: 1) validating and expanding two newly developed bud cold hardiness prediction models through analysis of regionally diverse data collected in the New York, Wisconsin, South Dakota, Minnesota, Michigan, Ohio, Pennsylvania, Iowa, and Quebec, 2) implementing model programming and development of the "ColdSnap Grape Bud Hardiness Model", producing a stakeholder accessibility-minded online tool that leverages customized climate data hosted by the Network for Environment and Weather Applications, and 3)developing educational resources for growers and provide extension-led outreach efforts to facilitate information dissemination and implementation.
Project Methods
Objective 1Model optimization and validation-The PDs have identified collaborators from 8 States and one Canadian Province, stretching from the Great Plains to the East Coast: SD, WI, OH, MI (2), NY (4), PA, IA, MN, QUE. These collaborators have been pre-identified as willing to supply either cold hardiness monitoring data or have agreed to collect field material and ship them to University of Wisconsin-Madison or Cornell University for analysis have enthusiastically agreed to (i) produce and share cold hardiness data collected from local cultivars, if they have equipment to do so, or (ii) ship cuttings with buds overnight to testing locations in either Wisconsinor New York . Differential thermal analysis (DTA), the standard practice of most cold hardiness monitoring programs, will be used to measure cold hardiness. Briefly, dormant buds are collected from field grown vines, selecting buds from a consistent position along the canes (e.g., buds 3-15). Buds are excised from the cane with a razor blade, preserving the bud-cane tissue connection, and placed in thermoelectric modules. Modules are then placed inside a programmable freezer which is run through a standard freezing rate protocol. After an initial hold of temperature at 0 °C to equilibrate temperature within the sample trays, temperature is ramped downward at a steady, slow rate of -4 °C/hour until it reaches -40°C. As the freezing program runs, supercooled water within the bud releases heat when it reaches its lethal freezing point, producing a low temperature exotherm that denotes the killing temperature of buds. Exotherm data is collected from replicate buds and processed into an approximation of the cold hardiness phenotype, denoted as LT50, or lethal temperature at which 50% of buds are killed.DTA data will be obtained from collaborators. All sites have at least three cultivars that are the focus of ongoing monitoring. The combined data provided in this dataset will allow optimization of the models such that they accommodate multiple climates with the same parameters. From collaborator sites where vines must be shipped for phenotyping, a minimum of three collection data points for each of the two first years are expected, with at least ten buds measured at each date. Most of the collaborating sites conduct their own local monitoring programs for cultivars of interest with biweekly or monthly sampling. The PD and Co-PDs conduct monitoring programs during the entire dormant season (~Oct to ~April) for several cultivars and can provide higher density and longer-term datasets that are available for their regions.Model calibration and validation: Models will be optimized using stepwise iterative methods using the calibration dataset in R. Parameters will be selected based on minimization of root mean square error (RMSE) of predictions compared to the measurements in the calibration set. To run the models, weather data for each location from where data is available will be extracted from NEWA. For locations from which DTA is obtained but NEWA is currently unavailable, NOAA data will be sourced.With optimized parameters, prediction ability of the models will be tested with the validation dataset. Predictions of both models (North-Atucha and Kovaleski-Londo) will be evaluated based on RMSE, correlation coefficients, and bias (the predicted minus measured value). These statistics will be produced for each cultivar × year × location interaction. Fit will be considered poor, and a model will not be used when RMSE > 5 °C and bias > 5 °C. This will allow for each model to be evaluated based on their performance for individual cultivars, separately for V. vinifera and hybrid cultivars, as well as separate regions.Based on local, regional, and cultivar-related prediction abilities, decisions will be made on whether either one or both models (hereto referred to as "ColdSnap model") will move on for implementation within NEWA 3.0. For implementation and programming within NEWA 3.0, initial functions associated with the ColdSnap model will be identified during the first year of the project so that the tool development can move forward independent of model evaluation. After model evaluations, the optimized parameters for each cultivar and each model within ColdSnap will be updated within the ColdSnap tool if necessary.Objective 2Model implementation and site selection-Model coding: The NEWA 3.0 ColdSnap tool, hosting the ColdSnap model will be built using the React JavaScript libraryTool development process: Development of the NEWA 3.0 ColdSnap model will be divided into five phases, or milestones, to ensure this project is completed on time and within the specified budget.Phase 1 will be completed in the first six months of Y1. In phase one, PIs and key collaborator Olmstead will gather and document end user requirements, create code documentation for the web programmer, and finalize a scope of work. Any feature or change requests made after completion of Phase one will be recorded for consideration in phase three.Phase 2 will be completed in the last six months of Y1. In phase two, PIs and key collaborator Olmstead will work directly with the software engineer at Northeast Regional Climate Center to implement the ColdSnap model iterated in the documentation created in phase one. To avoid confusion and project delays, key collaborator Olmstead will be the single channel of communication with the software engineer during this phase in order to complete a beta model for testing and validation in phase three on time.Phase 5 will be completed during Y3 and will encompass the production of predictive risk assessment for specific cultivars across the gridded climate data. Risk assessment will be determined through modeling freeze risk likelihood based on changes in the past two 30-year climate normals for each model which will allow for decision making steps for cultivar and site selection by growers based on this information. When PIs agree that all tasks related to NEWA 3.0 model development have been completed, the development portion of this project will be closed out.Objective 3. Development of educational resources for stakeholders-Develop three webinars: The webinars will be used to inform growers, consultants, and extension educators about the project results and deliverables, and provide technical instruction on how to get access and use the ColdSnap tool. Early webinar content (Y2) will cover monitoring and modeling acclimation and deacclimation in grapevines, cold damage risk periods, and available management to minimize cold damage in grapevines. Webinars in Y3 will cover how to access and use ColdSnap and how to interpret model recommendations. Webinar content will be archived for continued use and captioned to increase accessibility.Videos: PD Atucha, Key collaborator Olmstead, and postdoctoral researcher will produce two-short videos (~5-7 minutes) consisting of an online quickstart tutorial for the NEWA 3.0 ColdSnap model, to be hosted on the NEWA Help Desk at Annual meetings and conferences: Introduction to the ColdSnap tool will be presented at local and regional meetings. During winter months the teamwill present in ~6 grape grower meetings including those in WI, NY, PA, MI, IA, and MN.Extension publications: The teamwill distribute information via newsletters, such as 'Wisconsin Fruit News' from UW-Madison (750 subscribers), Veraison to Harvest from Cornell University (>1,000 subscribers), 'Fruit Times' from Penn State Extension (>1,000 subscribers), and the Minnesota Fruit and Vegetable News (900 subscribers) and other extension newsletters edited by our extension collaborators. We will also pursue publishing 2 articles in high impact trade journals such as The Good Fruit Growers and Growing Produce. We will publish 2 posts in the NEWA blog that reaches over 4,000 unique users.

Progress 02/28/23 to 02/27/24

Outputs
Target Audience:Grape growers, extension professionals, scientist. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?This project has created the opportunity for Michael North (post doctoral fellow) to (1) facilitate a multi-state collaborative research project for the first time, (2) practice communicating research procedures and outcomes to a general audience, (3) expand his experience collecting a new time-series cold hardiness dataset, and (4) further develop his experience analyzing cold hardiness data and writing code to generate process-based mathematical models. How have the results been disseminated to communities of interest?M. North presented a seminar at thePlants and Climate Symposium at Reiman Gardens (Ames, IA) in September 2023. This event was attended by a wide audience demographic, including general public, farmers, Iowa State University students, staff and faculty, and Iowa State University Extension and Outreach staff. Co-PI A. Kovaleski presented a seminartitled "Cold hardiness is a covariant in dormancy assays" at the International Plant Dormancy Symposium in Perth, WA, Australia on 09/2023. This event was attended by scientist, extension professional, and stakeholders. Co-PI A. Kovaleski presented a seminar titled Dormancy non-binary: moving away from the endo-ecodormancy dichotomy (part of workshop: "Fruit and nut tree dormancy and flowering biology in a changing environment") at the American Society for Horticultural Science Annual Conference on 08/2023 in Orlando, FL. This event was attended by scientists, students, extension professional, and stakeholders. Co-PI A. Kovaleski presented a seminar titled "Activity of woody perennials through winter" at the Allen Centennial Garden Winter Presentation Series at the University of Wisconsin Madison, Madison WI on 02/2023. This event was attended by a wide audience demographic, including general public, farmers, students, staff and faculty, and Extension and Outreach staff. What do you plan to do during the next reporting period to accomplish the goals?Obj. 1:We plan expand our compiled dataset by (1) soliciting more requests for preexisting bud cold hardiness datasets, (2) repeating experiments with dormant grapevine canes in the same locations and with the same cultivars approximately monthly in winter of 2024-2025, and (3) expanding future experiments to include both new locations (coordinating with collaborators in Georgia and Ohio) and additional cultivars (considering Vitis vinifera cultivars that are available from 3 or more locations). We will also proceed with model optimization and validation using stepwise iterative methods that generate root mean square error, correlation coefficients, and bias (the predicted minus measured value). These statistics will be produced for all interaction of cultivar, year, and location. Obj. 2:A user interface (UI) for preliminary testing will be completed in Year 2. User experience and user interface (UX/UI) feedback will be collected in the coming year for improvements over the winter. NYUS1 and WIUS1 models will be finalized by June 15, 2024. These are programmed in javascript directly to the web app. NYUS2 is generated from a python-based machine learning algorithm and requires an external API call. This is within project scope but was unexpected and requires additional time for implementation because a separate workflow is required to send station data from NEWA to an external endpoint hosted by NYSIPM in the cloud, which will then return results. Once implemented, however, this will be a useful method for using new ML and AI methods in NEWA risk assessments. Northeast Regional Climate Center also developed an option for users to choose a 'gridded' data point, which will be implemented by June 1. 2024. This means a user can use a physical NEWA location or a 'virtual' location from any location in the continental United States (CONUS). This hybrid approach is the first of its kind on the NEWA platform and work completed for use in this grape cold hardiness risk tool lays the groundwork for future integrations in future models. The user will have a seamless experience when switching between models and data types, however, as the entire process will take 1 or 2 seconds at most. By October 2024, all three models will be fully integrated, allowing the user to select from an extensive list of grape cultivars, each utilizing one of the three models, predetermined by the lead investigators. Obj. 3: We will deliver 2 webinars to introduce stakeholders to the NEWA ColdSnap tool; we will present at 2 grower meetings (NY and WI); we will develop 2 blog posts in the NEWA blog. All these extension activities will target grape growers in the Midwest and Northeastern regions, extension professionals, and scientist.

Impacts
What was accomplished under these goals? Obj. 1. Validate and expand two newly developed bud cold hardiness prediction models through analysis of regionally diverse data collected in: New York, Wisconsin, South Dakota, Minnesota, Michigan, Ohio, Pennsylvania, Iowa, and Quebec. A bud cold hardiness dataset for model optimization and validation was created by compiling multiple preexisting datasets, including datasets from Iowa (3 sites, 33 cultivars, 2 seasons), Michigan (11 sites, 11 cultivars, 2 seasons), Pennsylvania (1 site, 2 cultivars, 2 seasons), Quebec (15 sites, 17 cultivars, 4 seasons), and Wisconsin (1 site, 5 cultivars, 5 seasons). In addition, new bud cold hardiness data was generated by running experiments with dormant grapevine canes from 7 states (IA, MN, 2 locations in NY, PA, SD, TX, 2 locations in WI) including 8 different cultivars (Cabernet Sauvignon, Clarion, Concord, Frontenac, Itasca, Marquette, Petite Pearl, Riesling; see list below for distribution of cultivars received according to state). These experiments were repeated in December 2023, January 2024, and February 2024. This compiled dataset has been organized and cleaned in preparation for further analysis. Locations and cultivars included in experiments conducted in winter of 2023-2024: Iowa: Concord, Marquette, Petite Pearl, Minnesota: Clarion, Frontenac, Itasca, Marquette New York: Cabernet Sauvignon (2 locations), Concord, Marquette, Riesling Pennsylvania: Concord, Marquette South Dakota: Marquette Texas: Cabernet Sauvignon, Concord, Riesling Wisconsin: Frontenac, Marquette (2 locations), Petite Pearl Objective 2. Model implementation and site selection. Phase 1/Year 1 (complete): End user requirements and technical information was gathered and compiled from project PIs, model researchers, and extension experts. Phase 2/Year 1 (90% complete):A user interface (UI) for preliminary testing has been completed. User experience and user interface (UX/UI) feedback has been collected from a small group of extension professionals.

Publications