Estimation of the Retention and Availability of Water in Soils of the State of Santa Catarina

ABSTRACT: Soil water retention and availability are important properties for agricultural production, which can be measured directly or estimated by pedotransfer functions. Some studies on this topic were carried out in Santa Catarina, Brazil. To improve the estimates, it is necessary to evaluate other properties, to analyze more soil types, as well as to use other analysis techniques such as artificial neural networks and regression trees. Thus, the objective of the study was to estimate the field capacity (FC), permanent wilting point (PWP), and available water (AW) in soils of Santa Catarina (SC), through multiple linear regressions (MLR), artificial neural networks (ANN), and regression trees (RT), more efficiently than the current pedotransfer functions. For this, samples of the horizons A and B of 70 profiles were collected to determine the texture, plasticity limit, FC, PWP, AW, specific surface (SS), organic carbon (OC) content, and microporosity. Pedotransfer functions were generated through MRL, ANN, and RT, considering as dependent variables the FC, PWP, and AW, and as independent variables the content of clay, silt, OC, plasticity limit, SS, and microporosity, through the test of four models, for surface and subsurface horizons. The RT estimated FC, PWP, and AW better than ANN and MRL. The best models to estimate water retention were those that used microporosity. When the database has few input variables, the model with clay, silt, and OC content is an alternative to estimate FC, PWP, and AW.

摘要：土壤持水性与有效性是农业生产的关键属性，可通过直接测定或土壤传递函数（pedotransfer functions）进行估算。巴西圣卡塔琳娜州已针对该主题开展了若干研究。为优化估算效果，有必要评估更多土壤属性、分析更多土壤类型，并采用人工神经网络、回归树等其他分析技术。据此，本研究旨在通过多元线性回归（multiple linear regressions, MLR）、人工神经网络（artificial neural networks, ANN）与回归树（regression trees, RT），相较现有土壤传递函数，更高效地估算圣卡塔琳娜州（SC）土壤的田间持水量（field capacity, FC）、永久萎蔫点（permanent wilting point, PWP）与有效水量（available water, AW）。为此，研究采集了70个土壤剖面的A层与B层样品，用于测定土壤质地、塑限、FC、PWP、AW、比表面积（specific surface, SS）、有机碳（organic carbon, OC）含量与微孔隙度。研究通过MLR、ANN与RT构建土壤传递函数，以FC、PWP与AW作为因变量，以黏粒、粉粒、OC含量、塑限、SS与微孔隙度作为自变量，针对表层与亚表层土壤层开展四组模型测试。结果表明，RT对FC、PWP与AW的估算效果优于ANN与MLR。纳入微孔隙度的模型为估算土壤持水性的最优模型。当数据集输入变量较少时，仅包含黏粒、粉粒与OC含量的模型可作为估算FC、PWP与AW的可行替代方案。