Regression models based on biometric variables for the prediction of banana harvest (Musa AAA, cv. Gran Enano) in Guácimo, Limón, Costa Rica

Abstract

Introduction. Research focused on developing models to predict flowering curves and key production parameters in banana has made notable progress. However, there is still a need for specific models capable of accurately estimating crop harvest. Objective. To develop a model based on biometric variables for predicting harvest in banana (Musa AAA), cultivar Gran Enano, under humid tropical conditions in Guácimo, Limón, Costa Rica. Materials and methods. A correlational study was conducted on a commercial farm owned by EARTH University between March and July 2023. A total of 347 productive units were evaluated across 10.3 hectares. Biometric variables such as plant height, pseudostem circumference, and number of leaves were measured on both the mother plant and the following ratoon. Harvest variables assessed included bunch weight, number of hands, finger length, and finger diameter for both the second and penultimate hands. A multiple linear regression model without intercept was applied to all dependent variables, with assumption validation procedures. Results. The height and pseudostem circumference of the mother plant showed the strongest and most significant correlations with harvest variables. The most effective model, based on assumption validation, was the second model designed for bunch weight prediction (mod2_peso), which produced the following equation: Peso=0.13CircunferenciaM + 6.06AlturaM + 0.46CircunferenciaH - 9.38HojasH - 0.63AlturaH with an R² of 0.96 where M denotes the correlation between mother plant variables and various harvest-related parameters. Conclusions. The analysis revealed a significant correlation between mother plant variables and various harvest-related parameters. The results highlight the importance of considering the biometric characteristics of the follower (ratoon) plant in predictive models. Furthermore, the coefficient of determination (R²) should not be used as the sole criterion for validating multiple linear regression models.