Predictions and supporting data for network-wide 7-day ahead forecasts of water temperature in the Delaware River Basin

Daily maximum water temperature predictions in the Delaware River Basin (DRB) can inform decision makers who can use cold-water reservoir releases to maintain thermal habitat for sensitive fish species. This data release contains the forcings and outputs of 7-day ahead maximum water temperature forecasting models that makes predictions at 70 river reaches in the upper DRB. The modeling approach includes process-guided deep learning and data assimilation (Zwart et al., 2023). The model is driven by weather forecasts and observed reservoir releases and produces maximum water temperature forecasts for the issue day (day 0) and 7 days into the future (days 1-7). In combination with data provided in Oliver et al. (2022), this release contains all data used to train and validate the water temperature forecast models. This includes a process-based model pre-trainer, forecasted gridded weather data, reservoir releases, and water temperature data. Additionally, this release contains predictions from five models: a long-short term memory network (LSTM), a recurrent graph convolution network (RGCN), LSTM with data assimilation, a RGCN with data assimilation, and a persistence model. The release contains a tidy version of the model predictions with paired observations for easier reuse. The data are organized into 4 child folders: 1) waterbody information, 2) model driver data, 3) model configurations, 4) model predictions, 5) model code. � This research was funded by the USGS. Waterbody Information - One shapefile of polylines for 70 river segments in this study, and one shapefile of reservoir polygons for the Pepacton and Cannonsville reservoirs Model Driver Data - Data used to drive predictive models (daily meteorology for river reaches and reservoirs, observed reservoir diversions and releases) Model Configurations - Model parameters and metadata used to configure GLM 3.1 reservoir models Model Predictions - Temperature predictions data files, including GLM 3.1 predictions of outflow and water temperature for reservoir outflow reaches, stream temperature predictions from the distance-weighted-average lotic-lentic input network, and 7-day ahead deep learning water temperature forecasts at 5 priority sites Model Code - Model code repository used to prepare data for training, validation, testing, and evaluation of model output

特拉华河流域（Delaware River Basin, DRB）的日最高水温预测，可为决策者提供关键决策依据，使其可通过调控冷水水库下泄流量，维持敏感鱼类赖以生存的水温栖息地。
本次发布的数据包含上游特拉华河流域70个河段的7日最高水温预报模型的强迫场数据与模型输出结果。该建模方案融合了过程引导深度学习与数据同化技术（Zwart等，2023）。
模型以气象预报数据与实测水库下泄流量为驱动输入，可生成起报当日（第0日）及未来7日（第1至7日）的最高水温预报结果。
结合Oliver等（2022）发布的数据集，本次发布内容涵盖了用于训练与验证水温预报模型的全部数据，包括过程驱动模型预训练参数、网格化预报气象数据、水库下泄流量数据以及水温实测数据。
此外，本次发布还包含5种模型的预报结果：长短期记忆网络（Long-Short Term Memory Network, LSTM）、循环图卷积网络（Recurrent Graph Convolution Network, RGCN）、搭载数据同化模块的LSTM、搭载数据同化模块的RGCN，以及持续性预报模型。
本数据集还提供了整理规范的模型预报结果与对应实测观测值配对数据集，以提升数据复用性。
数据集分为5个子文件夹：1）水体信息；2）模型驱动数据；3）模型配置文件；4）模型预报结果；5）模型代码。
本研究由美国地质调查局（USGS）资助。
水体信息文件夹：包含本研究中70个河段的多段线形状文件（Shapefile），以及佩帕克通（Pepacton）与坎农斯维尔（Cannonsville）两座水库的面状形状文件（Shapefile）。
模型驱动数据文件夹：用于驱动预测模型的相关数据，包括河段与水库的逐日气象数据、实测水库引水与下泄流量数据。
模型配置文件文件夹：用于配置GLM 3.1水库模型的参数与元数据。
模型预报结果文件夹：包含各类水温预报数据文件，具体包括GLM 3.1模型对水库出库河段的出库流量与水温的预报结果、基于距离加权平均河库输入网络的河道水温预报结果，以及5个优先监测站点的7日深度学习水温预报结果。
模型代码文件夹：用于开展数据预处理、模型训练、验证、测试以及模型输出评估的代码仓库。