A robust framework for estimating precipitable water vapor based on near-infrared remote sensing data and reanalysis data fusion (NRF) method

High-precision and high-resolution precipitable water vapor (PWV) are essential for enhancing the accuracy of numerical weather prediction and correcting atmospheric path delays in earth observation systems(EOS). Although existing near-infrared (NIR) remote sensing PWV products offer fine spatial resolution, their utility is limited by reduced accuracy and substantial data gaps in cloud-contaminated regions. Therefore, this study proposed a NIR remote sensing and reanalysis data fusion (NRF) framework for large-scale PWV estimation, aimed at producing seamless, high-accuracy, 1 km spatial resolution PWV. The NRF framework consists of three stages. First, a nonlinear bias calibration model (NBCM) based on machine learning was developed to effectively eliminate biases between clear-sky MODIS and ERA5 PWV. Second, an annual water vapor cycle multi-scale data fusion model (AVCM) was constructed to fill MODIS cloud-contaminated gaps by integrating the bias-corrected MODIS and ERA5 PWV. Third, a guided filtering enhancement model (GFEM) was developed to spatially enhance the fused result using spatially adjacent clear-sky MODIS observations, improving spatial continuity and preserving fine-scale heterogeneity. Validation against ground-based GNSS PWV shows strong agreement, with a Pearson correlation coefficient of 0.98, bias of 0.13 mm, root mean square error of 1.78 mm, and Kling-Gupta Efficiency of 0.93. The NRF framework preserves the native 1 km resolution of MODIS while achieving full spatial coverage under all-weather conditions. By employing pixel-level modeling, it effectively eliminating the reliance on the spatial completeness of input images, thereby offering greater adaptability and flexibility for large-scale data fusion applications. The resulting high-quality PWV datasets hold substantial value for weather prediction and atmospheric delay correction in satellite-based remote sensing.