OL-E2STARFM: an enhanced spatiotemporal data fusion method integrating object-oriented classification strategy and flexible object-driven pixel prediction method

Spatiotemporal Data Fusion (STDF) can effectively integrate the advantages of high spatial and temporal resolutions in remotely sensed images. However, existing STDF methods still face two persistent challenges: a trade-off between accuracy and computational efficiency, and low data utilization due to the reliance on two fully cloud-free image pairs. To address these issues, this study proposes an enhanced ESTARFM by integrating an object-oriented classification strategy and a flexible object-driven pixel prediction method, namely OL-E2STARFM. OL-E2STARFM’s performance is assessed by Landsat-MODIS surface reflectance datasets, and compared with ESTARFM, OL-ESTARFM and other STDF methods. Results show that OL-E2STARFM is 147.96−162.74 times faster than ESTARFM and 0.52−0.65 times faster than OL-ESTARFM, while maintaining good prediction accuracy, as evidenced by its lowest average RMSE value of 0.0351, compared to 0.0357 for ESTARFM and 0.0360 for OL-ESTARFM. In addition, OL-E2STARFM outperforms existing OL-STDF methods by achieving a better trade-off between accuracy and efficiency. Furthermore, OL-E2STARFM significantly increases the utilization of cloud-covered image data, with an increase from 7.58% to 50.56% when the cloud cover ranges from 5% to 50% and is randomly distributed across two reference image pairs. The proposed OL-E2STARFM shows promise in monitoring long-term dynamic changes on the Earth’s surface.