Processed MIKE-generated dataset for physics-informed multi-gate scheduling in an urban river network

This dataset supports the manuscript entitled “Regime-dependent benefits of physics-informed learning for multi-gate scheduling in an urban river network”. The dataset contains processed MIKE-generated hourly simulation data for an urban multi-gate river network in the Shima River basin, Dongguan, China. The uploaded files correspond to 90 synthetic hydrological scenarios across different annual exceedance probability (AEP) levels and synthetic year identifiers. Each CSV file contains processed hourly hydrological, hydraulic, boundary-condition, gate-operation, and target variables used for deep learning model development and evaluation. The dataset was used to train, validate, and test data-driven and physics-informed sequence-learning models for multi-gate scheduling. It supports the analyses of gate-opening prediction accuracy, discharge consistency, high-opening residual behaviour, zero-flow prediction, and regime-dependent model performance reported in the manuscript. The original MIKE model configuration files and raw hydrodynamic simulation outputs are not included because they are subject to software-licensing and project data-management restrictions. The processed CSV files provided here are sufficient to reproduce the main data-driven analyses reported in the manuscript.