Feature-specific nutrient management of onion (Allium cepa) using machine learning and compositional methods.

While onion cultivars, irrigation and soil and crop management have been given much attention in Brazil to boost onion yields, nutrient management at field scale is still challenging due to large dosage uncertainty. Our objective was to develop an accurate feature-based fertilization model for onion crops. We assembled climatic, edaphic, and managerial features as well as tissue tests into a database of 1182 observations from multi-environment fertilizer trials conducted during 13 years in southern Brazil. The complexity of onion cropping systems was captured by machine learning (ML) methods. The RReliefF ranking algorithm showed that the split-N dosage and soil tests for micronutrients and S were the most relevant features to predict bulb yield. The decision-tree random forest and extreme gradient boosting models were accurate to predict bulb yield from the relevant predictors (R2 > 90%). As shown by the gain ratio, foliar nutrient standards for nutritionally balanced and high-yielding specimens producing > 50 Mg bulb ha-1 set apart by the ML classification models differed among cultivars. Cultivar × environment interactions support documenting local nutrient diagnosis. The split-N dosage was the most relevant controllable feature to run future universality tests set to assess models' ability to generalize to growers' fields.

Phytoremediation of Cu-contaminated vineyard soils in Brazil: A compendium of Brazilian pot studies.

Vineyard soils can be contaminated by copper (Cu) due to successive applications of fungicides and organic fertilizers. Soil remediation can be addressed by altering soil properties or selecting efficient Cu-extracting cover crops tolerant to Cu toxicity. Our objectives were to synthesize the Cu-extracting efficiency by plant species tested in Brazil, classify them according to Cu resistance to toxicity, and assess the effect of soil properties on attenuating Cu toxicity. We retrieved results from 41 species and cultivars, totaling 565 observations. Freshly added Cu varied between 50 and 600 mg Cu kg-1 of soil across studies. The partition of Cu removal between the above- and below-ground portions was scaled as a logistic variable to facilitate data synthesis. The data were analyzed using the Adaboost machine learning model. Model accuracy (predicted vs. actual values) reached R2  = 0.862 after relating species, cultivar, Cu addition, clay, SOM, pH, soil test P, and Cu as features to predict the logistic target variable. Tissue Cu concentration varied between 7 and 105 mg Cu kg-1 in the shoot and between 73 and 1340 mg Cu kg-1 in the roots. Among soil properties, organic matter and soil test Cu most influenced the accuracy of the model. Phaseolus vulgaris, Brassica juncea, Ricinus communis, Hordeum vulgare, Sorghum vulgare, Cajanus cajan, Solanum lycopersicum, and Crotolaria spectabilis were the most efficient Cu-extracting cover crops, as shown by positive values of the logistic variable (shoot removal > root removal). Those Cu-tolerant plants showed differential capacity to extract Cu in the long run.

Prediction of Nitrogen Dosage in 'Alicante Bouschet' Vineyards with Machine Learning Models.

Vineyard soils normally do not provide the amount of nitrogen (N) necessary for red wine production. Traditionally, the N concentration in leaves guides the N fertilization of vineyards to reach high grape yields and chemical composition under the ceteris paribus assumption. Moreover, the carryover effects of nutrients and carbohydrates stored by perennials such as grapevines are neglected. Where a well-documented database is assembled, machine learning (ML) methods can account for key site-specific features and carryover effects, impacting the performance of grapevines. The aim of this study was to predict, using ML tools, N management from local features to reach high berry yield and quality in 'Alicante Bouschet' vineyards. The 5-year (2015-2019) fertilizer trial comprised six N doses (0-20-40-60-80-100 kg N ha-1) and three regimes of irrigation. Model features included N dosage, climatic indices, foliar N application, and stem diameter of the preceding season, all of which were indices of the carryover effects. Accuracy of ML models was the highest with a yield cutoff of 14 t ha-1 and a total anthocyanin content (TAC) of 3900 mg L-1. Regression models were more accurate for total soluble solids (TSS), total titratable acidity (TTA), pH, TAC, and total phenolic content (TPC) in the marketable grape yield. The tissue N ranges differed between high marketable yield and TAC, indicating a trade-off about 24 g N kg-1 in the diagnostic leaf. The N dosage predicted varied from 0 to 40 kg N ha-1 depending on target variable, this was calculated from local features and carryover effects but excluded climatic indices. The dataset can increase in size and diversity with the collaboration of growers, which can help to cross over the numerous combinations of features found in vineyards. This research contributes to the rational use of N fertilizers, but with the guarantee that obtaining high productivity must be with adequate composition.

Garlic (Allium sativum) feature-specific nutrient dosage based on using machine learning models.

Brazil presents large yield gaps in garlic crops partly due to nutrient mismanagement at local scale. Machine learning (ML) provides powerful tools to handle numerous combinations of yield-impacting factors that help reducing the number of assumptions about nutrient management. The aim of the current study is to customize fertilizer recommendations to reach high garlic marketable yield at local scale in a pilot study. Thus, collected 15 nitrogen (N), 24 phosphorus (P), and 27 potassium (K) field experiments conducted during the 2015 to 2017 period in Santa Catarina state, Brazil. In addition, 61 growers' observational data were collected in the same region in 2018 and 2019. The data set was split into 979 experimental and observational data for model calibration and into 45 experimental data (2016) to test ML models and compare the results to state recommendations. Random Forest (RF) was the most accurate ML to predict marketable yield after cropping system (cultivar, preceding crops), climatic indices, soil test and fertilization were included features as predictor (R2 = 0.886). Random Forest remained the most accurate ML model (R2 = 0.882) after excluding cultivar and climatic features from the prediction-making process. The model suggested the application of 200 kg N ha-1 to reach maximum marketable yield in a test site in comparison to the 300 kg N ha-1 set as state recommendation. P and K fertilization also seemed to be excessive, and it highlights the great potential to reduce production costs and environmental footprint without agronomic loss. Garlic root colonization by arbuscular mycorrhizal fungi likely contributed to P and K uptake. Well-documented data sets and machine learning models could support technology transfer, reduce costs with fertilizers and yield gaps, and sustain the Brazilian garlic production.

Esmeralda Peach (Prunus persica) Fruit Yield and Quality Response to Nitrogen Fertilization.

'Esmeralda' is an orange fleshed peach cultivar primarily used for juice extraction and secondarily used for the fresh fruit market. Fruit yield and quality depend on several local environmental and managerial factors, mainly on nitrogen, which must be balanced with other nutrients. Similar to other perennial crops, peach trees show carryover effects of carbohydrates and nutrients and of nutrients stored in their tissues. The aims of the present study are (i) to identify the major sources of seasonal variability in fruit yield and qu Fruit Tree Department of Federal University of Pelotas (UFPEL), Pelotas 96010610ality; and (ii) to establish the N dose and the internal nutrient balance to reach high fruit yield and quality. The experiment was conducted from 2014 to 2017 in Southern Brazil and it followed five N treatments (0, 40, 80, 120 and 160 kg N ha-1 year-1). Foliar compositions were centered log-ratio (clr) transformed in order to account for multiple nutrient interactions and allow computing distances between compositions. Based on the feature ranking, chilling hours, degree-days and rainfall were the most influential features. Machine learning models k-nearest neighbors (KNN) and stochastic gradient decent (SGD) performed well on yield and quality indices, and reached accuracy from 0.75 to 1.00. In 2014, fruit production did not respond to added N, and it indicated the carryover effects of previously stored carbohydrates and nutrients. The plant had a quadratic response (p < 0.05) to N addition in 2015 and 2016, which reached maximum yield of 80 kg N ha-1. In 2017, harvest was a failure due to the chilling hours (198 h) and the relatively small number of fruits per tree. Fruit yield and antioxidant content increased abruptly when foliar clrCu was >-5.410. The higher foliar P linearly decreased total titratable acidity and increased pulp firmness when clrP > 0.556. Foliar N concentration range was narrow at high fruit yield and quality. The present results have emphasized the need of accounting for carryover effects, nutrient interactions and local factors in order to predict peach yield and nutrient dosage.

Balance design for robust foliar nutrient diagnosis of "Prata" banana (Musa spp.).

The "Cavendish" and "Prata" subgroups represent respectively 47% and 24% of the world banana production. Compared to world average progressing from 10.6 to 20.6 t ha-1 between 1961 and 2016, and despite sustained domestic demand and the introduction of new cultivars, banana yield in Brazil has stagnated around 14.5 t ha-1 mainly due to nutrient and water mismanagement. "Prata" is now the dominant subgroup in N-E Brazil and is fertigated at high costs. Nutrient balances computed as isometric log-ratios (ilr) provide a comprehensive understanding of nutrient relationships in the diagnostic leaf at high yield level by combining raw concentration data. Although the most appropriate method for multivariate analysis of compositional balances may be less efficient due to non-normal data distribution and limited nutrient mobility in the plant, robustness of the nutrient balance approach could be improved using Box-Cox exponents assigned to raw foliar concentrations. Our objective was to evaluate the accuracy of nutrient balances to diagnose fertigated "Prata" orchards. The dataset comprised 609 observations on fruit yields and leaf tissue compositions collected from 2010 to 2016 in Ceará state, N-E Brazil. Raw nutrient concentration ranges were ineffective as diagnostic tool due to considerable overlapping of concentration ranges for low- and high-yielding subpopulations at cutoff yield of 40 Mg ha-1. Nutrient concentrations were combined into isometric log-ratios (ilr) and normalized by Box-Cox corrections between 0 and 1 which may also account for restricted nutrient transfer from leaf to fruit. Despite reduced ilr skewness, Box-Cox coefficients did not improve model robustness measured as the accuracy of the Cate-Nelson partition between yield and the multivariate distance across ilr values. Sensitivity was 94%, indicating that low yields are attributable primarily to nutrient imbalance. There were 148 false-positive specimens (high yield despite nutrient imbalance) likely due to suboptimal nutrition, contamination, or luxury consumption. The profitability of "Prata" orchards could be enhanced by rebalancing nutrients using ilr standards with no need for Box-Cox correction.