A Physics-Informed Empirical Model for Predicting Vegetation-Related InSAR Coherence

Vegetation-induced temporal decorrelation limits the reliability of time-series InSAR measurements over densely vegetated areas. Existing empirical models generally use NDVI from a single acquisition and therefore cannot represent vegetation changes between the two SAR acquisitions. This study investigates coherence prediction for interferometric pairs with an approximately 60-day temporal baseline. Guided by the physical understanding of vegetation-related decorrelation, we use two complementary predictors: the sum of the NDVI values from the two acquisition dates, representing overall vegetation cover, and their absolute difference, representing inter-acquisition vegetation change. Separate exponential response functions are fitted and combined through a weighted additive formulation to construct a physically motivated dual-NDVI empirical coherence model (PICM). The additive formulation is an empirical predictive representation rather than a direct derivation from the multiplicative decomposition of coherence. The model is calibrated using correlation-screened samples from the Juye area, after which its coefficients are held fixed and applied to all valid pixels without correlation-based preselection. Independent spatial validation is conducted in the Gu’an area without recalibration. The proposed model achieves an error standard deviation of 0.166 and a mean error of −0.048 in Gu’an, compared with 0.190 and 0.089 for the single-NDVI model, respectively. The results demonstrate that incorporating both vegetation cover and inter-acquisition vegetation change improves coherence prediction for the approximately 60-day interferometric pairs examined. Its applicability to shorter temporal baselines and other land-cover conditions requires further investigation.