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

<jats:p>Satellite precipitation products such as IMERG are widely used in hydrometeorological operations, yet their biases are jointly influenced by terrain, season, and precipitation regime, leaving the applicability boundaries of machine learning correction unclear. This study proposes the Terrain–Moisture–Intensity (TMI) framework, which centers on mechanism purity to extend the correction problem from purely algorithmic optimization to physical consistency diagnosis. Based on a proof-of-concept study in Hunan Province, IMERG V07, SRTM DEM, and ERA5 variables (tcwv, u10, v10) are employed for correction analysis. Ablation results indicate that, under the conditions of this study, terrain–moisture relationships are predominantly additive: RF-Full yields merely +0.001 R2 gain over LR-Full, while bias rises to 1.282 mm d-1; MAE decreases by approximately 14%, reflecting a trade-off between tail-fitting improvement and mean shift. SHAP diagnostics identify three categories of boundaries. Spatially, Central Hunan exhibits significant degradation (R2 = 0.133) despite strong variable activation, consistent with mechanism fragmentation induced by mixed terrain. Temporally, u10 undergoes directional reversal between summer and spring (+0.096 to −0.156), presenting "silent failure." In intensity, extreme precipitation (≥50 mm d-1) more closely approximates a mechanism saturation frontier than isolated out-ofdistribution samples, with SHAP disorder intensifying with strength and DEM showing the largest relative amplification (approximately +150%). The results demonstrate that machine learning correction performance is primarily constrained by mechanism purity. The framework can be applied to pre-operational boundary identification, adaptive model selection, and physics-constrained feature engineering. A pre-registered cross-regional test (Hunan, Guangxi, Guangdong) confirms this screening capability out of sample: a priori coherence proxies predict correction efficiency with a mean absolute error of 2.6 percentage points, while the transfer-versus-retraining contrast separates mechanismmismatch (coastal Guangdong) from portability (Guangxi), establishing the framework as a validated applicability screen.</jats:p>