Canavera Ginevra, Magnanini Eugenio, Lanzillotta Simone, Malchiodi Claudio, Cunial Leonardo, Poni Stefano
Dipartimento di Scienze delle Produzioni Vegetali Sostenibili, Via Emilia Parmense 84, 29122, Piacenza, Italy.
Latitudo Srl, Via Modonesi 12, 29122, Piacenza, Italy.
Sci Rep. 2023 Oct 5;13(1):16818. doi: 10.1038/s41598-023-44019-4.
A web-based app was developed and tested to provide predictions of phenological stages of budburst, flowering and veraison, as well as warnings for meteorological drought. Such predictions are especially urgent under a climate change scenario where earlier phenology and water scarcity are increasingly frequent. By utilizing a calibration data set provided by 25 vineyards observed in the Emilia Romagna Region for two years (2021-2022), the above stages were predicted as per the binary event classification paradigm and selection of the best fitting algorithm based on the comparison between several metrics. The seasonal vineyard water balance was calculated by subtracting daily bare or grassed soil evapotranspiration (ET) and canopy transpiration (T) from the initial water soil reservoir. The daily canopy water use was estimated through a multiple, non-linear (quadratic) regression model employing three independent variables defined as total direct light, vapor pressure deficit and total canopy light interception, whereas ET was entered as direct readings taken with a closed-type chamber system. Regardless of the phenological stage, the eXtreme Gradient Boosting (XGBoost) model minimized the prediction error, which was determined as the root mean square error (RMSE) and found to be 5.6, 2.3 and 8.3 days for budburst, flowering and veraison, respectively. The accuracy of the drought warnings, which were categorized as mild (yellow code) or severe (red code), was assessed by comparing them to in situ readings of leaf gas exchange and water status, which were found to be correct in 9 out of a total of 14 case studies. Regardless of the geolocation of a vineyard and starting from basic in situ or online weather data and elementary vineyard and soil characteristics, the tool can provide phenology forecasts and early warnings of meteorological drought with no need for fixed, bulky and expensive sensors to measure soil or plant water status.
开发并测试了一款基于网络的应用程序，用于预测芽萌动、开花和转色期的物候阶段，并发布气象干旱预警。在气候变化情景下，物候提前和水资源短缺日益频繁，此类预测尤为迫切。利用艾米利亚-罗马涅地区25个葡萄园在两年（2021 - 2022年）间提供的校准数据集，按照二元事件分类范式并基于多个指标的比较选择最佳拟合算法，对上述阶段进行了预测。通过从初始土壤水库蓄水量中减去每日裸地或草地土壤蒸发散（ET）和冠层蒸腾量（T）来计算季节性葡萄园水分平衡。通过一个多元非线性（二次）回归模型估算每日冠层水分利用，该模型采用三个自变量，分别定义为总直射光、水汽压差和冠层总光截获量，而ET则采用封闭型箱式系统的直接读数。无论物候阶段如何，极端梯度提升（XGBoost）模型将预测误差降至最低，预测误差以均方根误差（RMSE）确定，芽萌动、开花和转色期的预测误差分别为5.6天、2.3天和8.3天。干旱预警分为轻度（黄色代码）或重度（红色代码），通过与叶片气体交换和水分状况的原位读数进行比较来评估其准确性，在总共14个案例研究中有9个案例的预警是正确无误的。无论葡萄园的地理位置如何，从基本的原位或在线气象数据以及基本的葡萄园和土壤特征出发，该工具无需使用固定、笨重且昂贵的传感器来测量土壤或植物水分状况，就能提供物候预测和气象干旱预警。
