Abstract

To identify the main edaphic variables most correlated to banana productivity in Venezuela and explore the development of an empirical correlation model to predict this productivity based on soil characteristics. Six agricultural fields located in two of the main banana production areas of Venezuela were selected. The experimental sites were in large farms (>= 50 ha) with four productivity levels in "Gran Nain" bananas, with an area of 4 ha for each of four productive levels: High - High, High - Low, Low - High, and Low - Low. Sixty sampling points were used to characterize the soils under study. Additionally, a Productivity Index (PI) based on three different biometric data on plant productivity was proposed. Through hierarchical statistical analysis, the first 16 soil variables that best explained the PI were selected. Thus, five multiple linear regression models were estimated, using the stepwise regression method. Subsequently, a performance analysis was used to compare the prediction quality range and the error associated with the number of soil variables selected for the proposed models. The selected model included the following soil variables: Mg, penetration resistance, total microbial respiration, bulk density, and omnivorous free-living nematodes. These variables explain the PI with anR(2)of 0.55, the mean absolute error (MAE) of 0.8, and the root of the mean squared error (RMSE) of 1.0. The five selected variables are proposed to characterize the soil Productivity Index in banana and could be used in a site-specific soil management program for the banana areas of Venezuela.