Martinson, T. E. | Search bioRxiv

bioRxiv2023

NYUS.2: an Automated Machine Learning Prediction Model for the Large-scale Real-time Simulation of Grapevine Freezing Tolerance in North America

O_LIAccurate and real-time monitoring of grapevine freezing tolerance is crucial for the sustainability of the grape industry in cool climate viticultural regions. However, on-site data is limited. Current prediction models underperform under diverse climate conditions, which limits the large-scale deployment of these methods. C_LIO_LIWe combined grapevine freezing tolerance data from multiple regions in North America and generated a predictive model based on hourly temperature-derived features and cultivar features using AutoGluon, an automatic machine learning engine. Feature importance was quantified by AutoGluon and SHAP value. The final model was evaluated and compared with previous models for its performance under different climate conditions. C_LIO_LIThe final model achieved an overall 1.36 {degrees}C root-mean-square error during model testing and outperformed two previous models using three test cultivars at all testing regions. Two feature importance quantification methods identified five shared essential features. Detailed analysis of the features indicates that the model might have adequately extracted some biological mechanisms during training. C_LIO_LIThe final model, named NYUS.2, was deployed along with two previous models as an R shiny-based application in the 2022-2023 dormancy season, enabling large-scale and real-time simulation of grapevine freezing tolerance in North America for the first time. C_LI

plant biology↗

bioRxiv2022

Development of a new cold hardiness prediction model for grapevine using phased integration of acclimation and deacclimation responses

Cold injury limits distribution of perennial agricultural crops, though replacement of plants and other management practices may allow for some damage tolerance. However, winter damage to crops such as grapevines (Vitis spp.) can result in losses in yield the following year if buds are damaged, but over many years when vines must be replaced and reach maturity before fruiting. Despite risks, grapevines are cultivated at the edge of permissible climate and rely on cold hardiness monitoring programs to determine when cold damage mitigation and management practices are required. These monitoring programs represent a critical, but laborious process for tracking cold hardiness. To reduce the need for continuous monitoring, a model (WAUS.2) using cold hardiness data collected over many years from Washington state, USA, growers was published in 2014. Although the WAUS.2 model works well regionally, it underperforms in other regions. Therefore, the objective of this work was to develop a new model (NYUS.1) that incorporates recent knowledge of cold hardiness dynamics for better prediction outcomes. Cold hardiness data from V. labruscana Concord, and V. vinifera Cabernet Sauvignon and Riesling from Geneva, NY, USA were used. Data were separated in calibration (~2/3) and validation (~1/3) datasets. The proposed model uses three functions to describe acclimation, and two functions to describe deacclimation, with a total of nine optimized parameters. A shared chill response between acclimation and deacclimation provides a phased integration where acclimation responses decrease over the course of winter and are overcome by deacclimation. The NYUS.1 model outperforms the WAUS.2 model, reducing RMSE by up to 37% depending on cultivar. The NYUS.1 model also tends to be more conservative in its prediction, slightly underpredicting cold hardiness, as opposed to the overprediction from the WAUS.2 model. Some optimized parameters were shared between cultivars, suggesting conserved physiology was captured by the new model. Highlights- Multi-year cold hardiness data from three grapevine cultivars were used for modeling - Cold hardiness was modeled based on daily temperature and accumulated chill - Phased acclimation and deacclimation processes result in cold hardiness predictions - The new model was compared to the currently available model for grapevines - The model proposed here outperforms the currently available model

plant biology↗