Traditional and Unmixing-Based semi-empirical models for BRDF correction in time series data

This research investigates two methodologies for correcting the Bidirectional Reflectance Distribution Function (BRDF) in Sentinel-2 imagery, focusing on transition zone in Israel between arid and sub-humid climatic zones across a year-long temporal span. BRDF correction is critical in remote sensing applications, especially for multi-temporal analysis, as it accounts for variations in surface reflectance due to changing illumination and viewing geometries. The study aims to compare the performance of these methods in normalizing surface reflectance and minimizing angular effects, enhancing the accuracy of time-series analysis for environmental monitoring. In particular, we test the performance of the traditional semi-empirical kernel-driven BRDF model, namely the Ross-Thick-Maignan (RTM) volumetric kernel and the Li-Transit-Reciprocal (LTR) geometric kernel and a recently proposed correction method RTM-LS-UMx, which relies on the kernel-driven model but incorporates the spectral unmixing results within the inversion process. This method was reported to be advantageous for mosaics of airborne images and laboratory data in previous work. Therefore, we aim to test this technique on satellite images influenced by seasonal changes in the sun’s position. We analyzed images of a transition zone in Israel between arid and sub-humid climatic zones to achieve this. The images were acquired in summer, mid-fall, and winter; thus, the sun’s position at this latitude affects the measured reflectance. Pre-processing steps included radiometric calibration, atmospheric correction, and cloud masking to ensure consistency across datasets. Quantitative evaluation used performance metrics, including the Normalized Difference Vegetation Index (NDVI) stability, Root Mean Square Error (RMSE) against a ground-truth dataset, and angular dependency reduction. Results indicated that both methods significantly improved reflectance consistency compared to uncorrected imagery. However, the unmixing-based model RTM-LS-UMx was advantageous concerning all examined metrics. The study further explored the impact of BRDF correction on long-term environmental monitoring applications. Time-series analysis revealed that both methods enhanced the detection of subtle surface changes previously obscured by angular variations.