Soil morphological properties described in the field, such as texture, consistence or structure, provide a valuable tool for the evaluation of soil productivity potential. In this study, we developed a regression model between the soil morphological variables of banana plantations and a crop Productivity Index (PI) previously developed for the same areas in Venezuela. For this, we implemented categorical regression, an optimal scaling procedure in which the morphological variables are transformed into a numerical scale, and can thus be entered in a multiple regression analysis. The model was developed from data from six plantations growing “Gran Nain” bananas, each with two productivity levels (high and low), in two 4-ha experimental plots, one for each productivity level. Sixty-three A horizons in thirty-six soils were described using 15 field morphological variables on a nominal scale for structure type, texture and hue, and an ordinal scale for the rest (structure grade, structure size, wet and dry consistence, stickiness, plasticity, moist value, chroma, root abundance, root size, biological activity and reaction to HCl). The optimum model selected included biological activity, texture, dry consistence, reaction to HCl and structure type variables. These variables explained the PI with an R² of 0.599, an expected prediction error (EPE) of 0.645 and a standard error (SE) of 0.135 using bootstrapping, and EPE of 0.662 with a SE of 0.236 using 10-fold cross validation. Our study showed how soil quality is clearly related to productivity on commercial banana plantations, and developed a way to correlate soil quality indicators to yield by using indicators based on easily measured soil morphological parameters. The methodology used in this study might be further expanded to other banana-producing areas to help identify the soils most suitable for its cultivation, thereby enhancing its environmental sustainability and profitability.

实地描述的土壤形态特性，如质地、稠度或结构，为评估土壤生产力潜力提供了宝贵的工具。在这项研究中，我们开发了香蕉种植园土壤形态变量与之前为委内瑞拉相同地区开发的作物生产力指数 (PI) 之间的回归模型。为此，我们实施了分类回归，这是一种将形态变量转换为数值尺度的最佳缩放程序，因此可以输入多元回归分析。该模型是根据 6 个种植“Gran Nain”香蕉的种植园的数据开发的，每个种植园有两个生产力水平（高和低），在两个 4 公顷的试验田中，每个生产力水平一个。在结构类型、质地和色调的名义尺度上使用 15 个田间形态变量描述了 36 个土壤中的 63 个 A 层，其余的则使用序数尺度（结构等级、结构尺寸、干湿稠度、粘性、可塑性、湿值、色度、根丰度、根大小、生物活性和对 HCl 的反应）。选择的最佳模型包括生物活性、质地、干稠度、对 HCl 的反应和结构类型变量。这些变量用 R 解释了 PI 选择的最佳模型包括生物活性、质地、干稠度、对 HCl 的反应和结构类型变量。这些变量用 R 解释了 PI 选择的最佳模型包括生物活性、质地、干稠度、对 HCl 的反应和结构类型变量。这些变量用 R 解释了 PI² of 0.599，使用引导法的预期预测误差 (EPE) 为 0.645，标准误差 (SE) 为 0.135，使用 10 折交叉验证的 EPE 为 0.662，SE 为 0.236。我们的研究显示了土壤质量如何与商业香蕉种植园的生产力明显相关，并通过使用基于易于测量的土壤形态参数的指标开发了一种将土壤质量指标与产量相关联的方法。本研究中使用的方法可能会进一步扩展到其他香蕉产区，以帮助确定最适合其种植的土壤，从而提高其环境可持续性和盈利能力。