Quantitative mechanism-purity metric

Develop a continuous quantitative mechanism-purity metric that is independent of machine-learning correction skill, so that precipitation-regime purity can be assessed without relying on correction outcomes.

Background

The TMI framework currently classifies mechanism purity using topographic coherence and precipitation climatology, supplemented by post hoc correction diagnostics such as SHAP directional consistency and disorder. The paper notes that this classification is not yet represented by a continuous quantitative measure that is independent of model performance.

Such a metric would support a priori assessment of whether a spatiotemporal window is sufficiently mechanistically coherent for machine-learning precipitation correction, rather than identifying low-purity regimes only after a model has been fitted and evaluated.

References

The subregional purity classification is defined a priori from topographic coherence and precipitation climatology (Section 2.1), independently of correction outcomes; a continuous, quantitative purity metric that is likewise independent of correction skill remains to be developed (Section 4.5).

— Machine learning correction of satellite precipitation is governed by mechanism purity, not algorithmic complexity: a proof-of-concept study in Hunan, China, with pre-registered cross-regional validation  (2608.12988 - Xu, 13 Aug 2026) in Section 4.1, paragraph discussing the spatial boundary and Central Hunan