A dataset of the global development likelihood of wastewater treatment plants

Sustainable management of water and sanitation requires the constructions of wastewater treatment plants (WWTPs) to remove pollutants from wastewater. A spatially explicit potential map depicting the future development sites of WWTPs is valuable for optimizing their implementation to advance progress toward SDG 6 (Clean Water and Sanitation). Here we use a novel positive-unlabeled deep neural network to model the global development likelihood of WWTPs using fourteen socioeconomic and geographic factors, including GDP, population, land type, vegetation cover, NDVI, distance to settlement, distance to power grid, distance to river, distance to lake, distance to fault, flood hazard, elevation, topographic complexity, and slope. Labeled positive data were from the global database HydroWASTE [1], and 267,325 random sites extracted globally were used as unlabeled data. The positive and unlabeled data were combined and randomly split into a training set (70%) and a test set (30%), repeated ten times. The area under the receiver operating characteristic curve, recall, precision, and F1-score of the model are 0.9876, 0.8611, 0.7178, and 0.7822, respectively. The predicted potential area suitable for WWTPs is about 8,287,091 km2 globally, 51% of which is distributed in the United States, China, India, Russia, and Germany. Cropland, built-up area, tree, and rangeland account for 96% of the potential area. The derived global probability map of WWTPs can provide input for decision-makers to implement multi-objective optimization considering various sustainable development goals. File descriptions: The maps use World Geodetic System 1984 coordinate system with the spatial resolution of 0.0083°. pro_wwtp.tif: the global probability (suitability) map of WWTP development. Ten individual models were trained from ten random realizations of the training set, and the average ensemble model was used for global mapping. std_wwtp.tif: the standard deviation map of the development likelihood of WWTPs over ten individual models. NoData Value: -9999.水资源与环境卫生的可持续管理，需要建设污水处理厂（Wastewater Treatment Plants, WWTPs）以去除废水中的污染物。一张空间显性的污水处理厂未来选址潜力图，对于优化项目落地、推进可持续发展目标6（Clean Water and Sanitation, SDG 6）的进程具有重要价值。本研究采用一种新颖的正样本未标注（positive-unlabeled, PU）深度学习神经网络，基于14项社会经济与地理因子——包括国内生产总值（GDP）、人口、土地类型、植被覆盖、归一化差异植被指数（Normalized Difference Vegetation Index, NDVI）、距居民点距离、距电网距离、距河流距离、距湖泊距离、距断层距离、洪水灾害风险、海拔、地形复杂度与坡度——构建全球污水处理厂开发潜力预测模型。带标签的正样本数据来自全球数据库HydroWASTE[1]，同时在全球范围内抽取267325个随机点位作为未标注数据。将正样本与未标注数据合并后，随机划分为训练集（70%）与测试集（30%），该划分过程重复十次。该模型的受试者工作特征曲线下面积（Area Under the Receiver Operating Characteristic Curve, AUC）、召回率、精确率与F1值分别为0.9876、0.8611、0.7178与0.7822。全球范围内适宜建设污水处理厂的预测潜力面积约为8287091平方千米，其中51%分布于美国、中国、印度、俄罗斯与德国。耕地、建成区、林地与牧场占该潜力总面积的96%。本研究生成的全球污水处理厂开发概率图，可为决策者在兼顾多项目标可持续发展的规划决策提供输入依据。 文件说明： 本数据集采用1984世界大地测量系统（World Geodetic System 1984, WGS 84）坐标系，空间分辨率为0.0083°。 pro_wwtp.tif：全球污水处理厂开发概率（适宜性）分布图。基于训练集的十次随机划分结果训练得到十个独立模型，最终采用平均集成模型生成全球分布图。 std_wwtp.tif：十个独立模型的污水处理厂开发潜力标准差分布图。 无数据值（NoData Value）：-9999。