Improving spatial resolution of Aqua MODIS and GCOM-C chlorophyll-a data for Cyprus coastal waters monitoring

Abstract

Chlorophyll-a (Chl-a), a key water quality indicator, is widely monitored via satellites to assess marine health. While satellite data provide broad coverage, their coarse spatial resolution limits detailed coastal analysis. Although downscaling methods have been explored globally, their application to European seas, particularly the Eastern Mediterranean, remains limited. This study addresses that gap by downscaling 4 km Chl-a data from Aqua MODIS and GCOM-C to a 300 m resolution near the Cyprus coast using regression-based techniques. Four images from spring-summer 2024 were selected for analysis, with Sentinel-3 spectral bands used as predictors and both multiple linear regression and random forest models applied. The results indicate that linear regression predicts higher coastal Chl-a values, while random forest smooths spatial gradients. For Aqua MODIS, both models performed similarly (R2 = 0.74, RMSE = 0.006), whereas random forest outperformed linear regression for GCOM-C (R2 = 0.603, RMSE = 0.008). Validation with in situ data showed improved correlations after downscaling, though some RMSE increases suggest model limitations or errors in ground measurements for certain dates. Overall, the downscaling method enhances the spatial resolution of satellite-derived Chl-a data, enabling more detailed monitoring of coastal waters, which could be valuable for environmental management and monitoring efforts.