This study aims to capture groundwater potential zones integrating deep neural network and groundwater influencing factors. The present work was carried out for Gopi khola watershed, mountainous terrain in Nepal Himalaya as the watershed mainly relies upon the groundwater assets; it is a need to explore groundwater potential for better management of the aquifer framework. Ten groundwater influencing factors were collected such as elevation, slope, curvature, topographic positioning index, topographic roughness index, drainage density, topographic wetness index, geology, lineament density, and land use thematic layers. Among those influencing factors, topographic roughness index was removed because of multicollinearity issue to reduce the dimension of the dataset. A spring inventory map of 145 spring locations was prepared using field survey method and an equal number of spring absence points were randomly generated. The 70% of spring and spring absence pixels were used as training dataset and remaining as test dataset. The final map was created based on predicted probabilities ranging from 0 to 1. The validation was done using the receiver operating characteristic curve, which shows that the area under the curve is 76.1% for the training dataset and 82.1% for the test dataset. The sensitivity analysis was performed using Jackknife test which shows that the lineament density is the most important factor. The experimental results demonstrated that deep neural network is highly capable to capture groundwater potential zone in mountainous terrain. The present study might be useful and preliminary work to exploit the groundwater. The consequences of the current study may be valuable to water administrators to settle on appropriate choices on the ideal utilization of groundwater assets for future arranging in the basic investigation zone.

中文翻译：

---

深度神经网络在尼泊尔喜马拉雅山山区捕获地下水潜在带中的应用。

这项研究旨在捕获结合了深度神经网络和地下水影响因素的地下水潜在地带。目前的工作是在尼泊尔喜马拉雅山的戈皮霍拉分水岭上进行的，因为该分水岭主要依靠地下水资源。有必要探索地下水潜力，以更好地管理含水层框架。收集了十个地下水影响因素，例如海拔，坡度，曲率，地形定位指数，地形粗糙度指数，排水密度，地形湿度指数，地质学，线质密度和土地利用专题层。在这些影响因素中，由于多重共线性问题而删除了地形粗糙度指数，以减小数据集的维数。使用现场调查方法准备了145个弹簧位置的弹簧库存图，并随机生成了相等数量的弹簧缺失点。春季和春季缺席像素的70％用作训练数据集，其余作为测试数据集。根据从0到1的预测概率创建最终图。使用接收器工作特性曲线进行验证，该曲线表明训练数据集和测试数据集的曲线下面积分别为76.1％和82.1％。使用Jackknife检验进行敏感性分析，结果表明线密度是最重要的因素。实验结果表明，深度神经网络具有捕获山区地形中地下水潜在区域的强大能力。本研究可能是有用的，并且是开发地下水的初步工作。当前研究的结果可能对水管理者来说很有价值，以便他们就理想地利用地下水资产的合理选择做出选择，以便将来在基础调查区进行安排。