Published November 2, 2026 | Version v0.1

HydroTuring: a conservation-law benchmark for AI hydrologic models

Description

HydroTuring tests whether an AI-based hydrologic model conserves mass, energy and momentum, rather than whether it fits a hydrograph. Forcing is generated at run time from a recorded seed, so there is nothing to memorise. Every probe must demonstrably separate a physically exact reference model from models that close the budget by construction, that hide a silent sink, or that reach closure through a degenerate partition.

Notes

If you use HydroTuring, please cite it and record the suite version. Scores are only comparable within a suite version.

Files

Flood-Lab/HydroTuring-v0.1.zip

Files (2.9 MB)

Name Size
md5:60bc3c772b1d0cdba5eaf8eb472bd987
2.9 MB Preview Download

Additional details

Related works