Predição de colheita e estimativas do tamanho e formato de parcelas experimentais para palma forrageira ‘Gigante’

Abstract

Among the forage crops, ‘Gigante’ cactus pear (Opuntia ficus-indica Mill) stands out, especially in the semiarid, for its high resilience to adverse conditions associated with high biomass production. Due to its unique importance in agriculture, several studies have been developed in search of greater understanding and agronomic assumptions about this cactus variety. Predicting harvest and defining the experimental plot size are important to safely extrapolate research results. Thus, the objective of this work was to build harvest-predicting models to estimate the size and shape of plots for phenotypic evaluation of the ‘Gigante’ forage cactus pear. The field experiment was carried out under rainfed condition in the experimental area of the Federal Institute of Bahia. A blank test, that is, with no treatments, was conducted, in which the soil was evenly prepared throughout the experimental area. The cladodes used for planting were selected and cured for five days in shade. Organic fertilizers were applied three times at a rate of 60 Mg ha-1 year-1 of sheep manure in the first year: 30 Mg ha-1 in the planting furrow and 30 Mg ha-1 as top dressing after rooting of the plants. In the second year, which corresponded to 360 days after planting (DAP), a new application was carried out with 60 Mg ha-1 between rows, repeating the same rate of 60 Mg ha-1 at 720 DAP. Other crop practices were done based on recommendations for forage crops. Therefore, the plants, each considered a basic unit (BU), were subjected to the same crop practices and planted at the same spacing of 2.0 x 0.2 m. The plants were arranged into 10 rows containing 50 plants each. Following the grid i x j, 384 plants or BUs arranged in i = 48 rows and j = 8 columns were considered for measurements, for a total area of 153.60 m2. At 930 DAP, the following morphological descriptors were measured in the third production cycle: plant height, cladode length, cladode width, cladode thickness, cladode area, number of cladodes, weight of cladodes, and total area of cladodes. For each vegetative characteristic measured in the 384 BUs, several sizes of plots were combined so that the plants would fill the whole experimental are. Thus, 15 different sizes of pre-established rectangular-shaped plots and rows were used. Harvest prediction was performed by Artificial Neural Networks (ANN) and by the models of simple, multiple, quadratic and interaction regression analysis. Plot sizes were estimated by the methods: modified maximum curvature, convenient plot size, linear and quadratic response plateau, comparison of variances and relative information. To predict the harvest, artificial neural networks with a coefficient of determination of 0.87 were fitted for validating the sample, which ensured the potential of generalization of the model. The morphological characters that had the greatest relative contribution were total area of the cladode, plant height, cladode thickness, and cladode length; however, all traits measured are important for predicting cactus pear yield. Through multiple linear regression, quadratic interaction across all variables or simple linear regression for total cladode area, it was possible to estimate the yield of forage cactus pear grown in Brazilian northeastern semiarid as nutritional and water resource for livestock. Regarding the determination of plot sizes, the modified maximum curvature model estimated plot sizes between three and nine plants in the direction of the plant row for the characteristics studied. Estimates of different plot sizes were built using the matrix methodology proposed by Hatheway, in which the investigator can make use of a series of combinations between parameters, number of treatment, number of replicates, difference to be detected and the coefficient of variation of the characteristic to select the convenient plot size. The linear and quadratic regression with plateau response methods determined plots with 10 and 17 basic units, respectively. By the method of comparison of variance, plots with 12 BUs (4.8 m2) were the optimum size since the use of larger sizes does not significantly decrease the variance. The most suitable plot size and shape was determined by the greater relative information associated with the lower coefficient of variation. Studies on ‘Gigante’ forage cactus pear have been conducted using plot sizes having 15, 32 and 36 basic units; however, based on the results of this work, the experimental area can be optimized by significantly reducing the size and shape of the plot. An experimental plot with eight basic units arranged in eight columns composed of one plant each ensures efficiency during the experimental evaluation. This combination of size and shape has the highest control of soil heterogeneity in addition to meeting all the characteristics normally evaluated in studies with forage cactus pear. Furthermore, reduction in experimental error and significant gains in accuracy are expected. From the morphological characters, one can predict the yield of forage cactus pear with high precision using ANNs and regression models.