Balanço de potássio e modelos de predição de produção em vinhedos no sul do Brasil

Abstract

Soils cultivated with vines may not have the necessary amount of potassium (K) to meet the plants' demand. When this happens, it is necessary to apply doses of potassium fertilizers, which can modify the distribution of K fractions in soils over the years. However, the doses of K to be applied will be more precise if the inputs and outputs of K in vineyards are known. This information allows us to propose the K balance. However, the distribution of fractions and K balance are not sufficiently known in vineyard soils, cultivated with different cultivars, which produce and export different amounts of K. Added to this, in international viticulture there is no Models for predicting productivity and quality of grape must are sufficiently well known, which consider nutrients in soils and leaves, in addition to other variables, such as climate. Therefore, two studies were conducted that aimed to evaluate the distribution of K fractions and their balance, and to evaluate the effect of the input of different predictor variables and their importance in the accuracy of predictions of must quality and grape production in vineyards of the cultivars. Chardonnay and Pinot Noir with application of doses of K, in southern Brazil. Study 1 quantified the K balance and the fractions of exchangeable K, non-exchangeable K, structural K and total K in the soil, with evaluations carried out in an experiment with the cultivars Chardonnay and Pinot Noir, without application of K and with annual applications of 40 and 80 kg ha-1 K2O, for twelve years. Study 2 proposed models and their accuracy in predicting grape production and must quality parameters (total soluble solids - TSS, total anthocyanins - AT, total polyphenols - PT, total titratable acidity - ATT, pH). The models were developed using machine learning, through multivariate analysis including the predictor variables soil nutrients, leaf nutrients and climate variables. A structured database was used during six harvests. In Study 1, the K balance was negative in the soil that did not receive K application and positive in the soil that received annual applications of 40 and 80 kg ha-1 K2O. It was also possible to observe that the fraction of non-exchangeable K contributed to increasing the fraction of exchangeable K in the soil without application of K. In Study 2, the models that combine soil+leaf+climate variables show greater accuracy in predicting grape production, ATT and pH of the must. Models that combine leaf+climate data show greater accuracy for PT concentration and models calibrated only with climate data for anthocyanins. The climatic variables were the most important (≥80%) in the models with the best accuracy for all predicted variables, except PT, in which nutrients in leaves (K, Ca and Mg) were more important. Therefore, the results of our study present the most approximate dose (40 kg ha-1 K2O per year) of potassium fertilizer to maintain a positive K balance, but without excess. As well, it allowed identifying and quantifying the importance of soil, plant and climate variables in must production and quality prediction models.