Synergistic Fusion of Aerosol Optical Depth over India from multi-sensor satellite retrievals with ground-based measurements

Abstract

Synergistic fusion of aerosol parameters from multi-sensor measurements is crucial for integrating diverse data sources and generating consistent representations of aerosol distribution for accurate climate impact assessment. In this study, satellite observations from MODIS (Moderate Resolution Imaging Spectroradiometer) and MISR (Multi-angle Imaging SpectroRadiometer) are combined with ground-based measurements from the ARFINET and AERONET to generate fused Aerosol Optical Depth (AOD) over India. The primary focus of this study is to develop a fusion framework, involving the evaluation and comparison of two approaches: geostatistical Universal Kriging (UK) and a novel hybrid Residual Kriging-Machine Learning (RK-ML). Both methods share the same geostatistical foundation (variogram-based spatial-modeling) but differ in how the mean structure of AOD is estimated. In UK, satellite-retrieved AOD serves as deterministic trend for spatial prediction and is effective when ground-based observations are well distributed, whereas RK-ML considers ML (SVR) predicted AOD as prior and applies Ordinary Kriging to interpolate residuals from real-time ground observations, maintaining a near-zero residual mean away from observations which reduces distortion under sparse and uneven data conditions. Our results highlight seasonal fused AOD maps resembling very close to ground-based AOD over India. Leave-One-Out Cross-Validation (LOOCV) is adopted as an evaluation strategy for assessing performance, showing that fused AOD from both UK and RK-ML approaches captures up to 100 % of ground observations within the 95 % confidence interval (± 2σ), indicating effectiveness in capturing regional aerosol variability. RK-ML demonstrates more stable spatial patterns and improved LOOCV performance compared to UK, particularly in regions with limited ground-based coverage.