Retrieval research of vegetation water content based on GNSS-IR and hyperspectral data fusion

Abstract Vegetation water content (VWC) is a critical biophysical parameter for monitoring vegetation growth and phenology. Normalized difference vegetation index (NDVI) is an important indicator for characterizing VWC. However, its estimation via spectral remote sensing is often limited by insufficient spatiotemporal resolution. While Global Navigation Satellite System Interferometric Reflectometry (GNSS-IR) has emerged as a promising alternative, this technique enables the retrieval of high spatiotemporal resolution NDVI through the analysis of interference signal amplitudes. However, its accuracy in retrieving the NDVI degrades under low-elevation conditions due to multipath signal superposition and mixed-polarization effects. To address this, a novel framework that integrates ground-based precise hyperspectral data with GNSS-IR amplitude parameters to estimate NDVI was developed in this study. For this purpose, both multiple linear regression (MLR) and random forest regression (RFR) models were employed for high-precision NDVI inversion with a spatiotemporal resolution of 1 d/250 m. Experimental results confirm that our data fusion strategy significantly enhances accuracy. The Pearson correlation coefficient ( R ) of the MLR model reaches 0.8497, representing an increase of 15.50% compared to using GNSS-IR amplitude alone. Similarly, the RFR model achieves an R of 0.9752, which is a 3.48% improvement over the model utilizing only the single GNSS-IR amplitude feature. In summary, the proposed method provides a robust approach for vegetation monitoring, effectively overcoming the spatiotemporal resolution constraints of single-source data and achieving superior accuracy, with potential applications in precision agriculture and ecosystem management.