The use of machine learning methods in pedotransfer functions has attracted considerable attention in recent years. These methods are fast and effective in solving complex events. The least limiting water range (LLWR) feature is very important in terms of water uptake by the plant and root development in agricultural production. In this study, the predictability of the LLWR feature was investigated with artificial neural networks, deep learning (DL) and the k-nearest neighbour (k-NN) algorithm from machine learning methods. Estimated values obtained from the model with the best estimation accuracy and observed values were evaluated through a geostatistical method from which their spatial distribution maps were created. In the present study, which was carried out on alluvial lands with different soil properties, the LLWR values of soils vary between 5.5% and 25.9%. Field capacity, bulk density, clay, organic matter, and lime content properties, which have a high correlation with the LLWR, were taken into consideration in the estimation methods. DL was determined as the best estimation method (mean absolute error [MAE]: 0.94%; root mean square error [RMSE]: 1.45%; coefficient of determination [R²]: 0.93), and the worst was k-NN (MAE: 2.00%; RMSE: 2.55%; R²: 0.77) for the LLWR. In addition, the LLWR can be estimated with high accuracy by using ReLU and softmax functions in the DL method. The study shows that distribution maps created with LLWR values obtained by observed data and the DL method have a very similar pattern.

中文翻译：

---

计算智能应用于最小限制水域，以使用 GIS 和地统计方法在冲积土地上估计土壤含水量*

近年来，机器学习方法在pedotransfer函数中的使用引起了相当大的关注。这些方法在解决复杂事件方面快速有效。最小限制水范围 (LLWR) 特征在植物吸水和农业生产中的根系发育方面非常重要。在这项研究中，使用人工神经网络、深度学习 (DL) 和机器学习方法中的 k 近邻 (k-NN) 算法研究了 LLWR 特征的可预测性。从具有最佳估计精度的模型中获得的估计值和观测值通过地统计方法进行评估，从中创建了它们的空间分布图。在本研究中，在具有不同土壤性质的冲积土地上进行，土壤的 LLWR 值在 5.5% 和 25.9% 之间变化。在估算方法中考虑了与 LLWR 具有高度相关性的田间容量、容重、粘土、有机质和石灰含量特性。DL 被确定为最佳估计方法（平均绝对误差 [MAE]：0.94%；均方根误差 [RMSE]：1.45%；确定系数 [R ² ]: 0.93)，最差的是 LLWR 的 k-NN (MAE: 2.00%; RMSE: 2.55%; R ² : 0.77)。此外，在 DL 方法中使用 ReLU 和 softmax 函数可以高精度地估计 LLWR。研究表明，使用观测数据和 DL 方法获得的 LLWR 值创建的分布图具有非常相似的模式。