Improving Multi-Sensor Spectral Harmonization: Bandpass Adjustment for GRUS-1 and Sentinel-2 Muhammad Daniel Iman bin Hussain (a*), Masahiko Nagai (a)
a) Graduate School of Sciences and Technology for Innovation, Yamaguchi University, Japan
*danieliman9897[at]gmail.com
Abstract
Sensor spectral response differences can introduce significant inconsistencies in reflectance values across satellite platforms, limiting the accuracy and interoperability of multi-source remote sensing analyses. This study presents the first reproducible framework for bandpass adjustment of the GRUS-1 satellite, aiming to harmonize its spectral reflectance with Sentinel-2 MSI, a widely used and radiometrically stable platform. Achieving this alignment enables more reliable downstream applications such as vegetation monitoring and change detection using GRUS imagery. To simulate spectral responses, we utilized three publicly available spectral libraries (USGS, ProSail, and Hyperion), which collectively represent diverse land cover types and reflectance conditions. These spectra were convolved with the relative spectral responses (RSRs) of GRUS-1 and Sentinel-2 to generate paired reflectance datasets. Initial analysis revealed a non-linear relationship between GRUS and Sentinel-2 reflectance, limiting the effectiveness of simple linear models. We therefore applied piecewise linear regression to adjust GRUS-1 reflectance values to match those of Sentinel-2. Model performance, evaluated using RMSE and R2, showed that piecewise regression substantially improved reflectance agreement across bands. To validate these findings, we used sampled reflectance data (approximately 1000 points) from GRUS and Sentinel-2 image pairs, ensuring seasonal diversity with at least one image per season. Evaluation on real satellite pairs confirmed the modeled adjustments, not only in terms of pixel-level RMSE and R2, but also through improved NDVI consistency between the two sensors. This work provides a practical spectral harmonization approach for GRUS-1, especially valuable in scenarios where spatial or temporal coverage is limited. Future work will explore advanced modeling techniques such as artificial neural networks to further improve spectral harmonization.
