Data, Code, and Model Artifacts for "Event-based Rainfall–Runoff Modeling with Hydrology-Aware Temporal Convolutional Networks: A Single-Basin Case Study"

Abstract

This record contains the data, code, trained model artifacts, and evaluation outputs supporting the manuscript “Event-based Rainfall–Runoff Modeling with Hydrology-Aware Temporal Convolutional Networks: A Single-Basin Case Study.” The study evaluates an event-based hydrology-aware Temporal Convolutional Network (TCN) using 144 manually delineated rainfall–runoff events from the NF Newaukum River Above Bear Creek near Forest, Washington (USGS station 12024400) at 15-min temporal resolution. The archived materials include the processed event dataset, fixed temporal train–validation–test partitions, model training and evaluation code, trained weights for the five predefined ensemble members, event-level performance metrics, and outputs from the ablation experiments. The temporal partition comprises 111 training events occurring through 2018, 15 validation events from 2019–2020, and 18 independent test events from 2021–2022. The five ensemble realizations use predefined random seeds 1337, 2026, 42, 777, and 999. The underlying hydrometeorological observations are publicly available from the U.S. Geological Survey for station 12024400. A README file included in this archive describes the directory structure, software requirements, model execution, evaluation workflow, and relationship between the archived files and the results reported in the manuscript. The underlying hydrometeorological observations were obtained from the U.S. Geological Survey National Water Information System for station 12024400 and are public-domain U.S. Government data; the archived processed event files were derived from those observations.