Effect of Image Pair Temporal Gap on Spatiotemporal Fusion

Abstract. Despite significant advances in remote sensing and the growing availability of time-series data, challenges remain due to sensor limitations and the trade-off between spatial and temporal resolution. These constraints hinder the acquisition of datasets with both high spatial and temporal resolution. Spatiotemporal fusion (STF) algorithms have emerged as an effective solution, but the complexity of land surface processes can reduce their accuracy. Land surface temperature (LST) is a key parameter for agricultural monitoring, directly affecting crop growth, evapotranspiration, and plant stress detection. This study examines the impact of temporal intervals between input image pairs on STF accuracy using nine pairs of LST images from MODIS and Landsat sensors over agricultural lands, covering intervals of 32, 48, and 96 days. Two widely used STF algorithms, Random Forest (RF) and Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), were applied for comparison. Results show a consistent decline in fusion accuracy with increasing temporal gaps. For RF, the Root Mean Square Error (RMSE) increased from 1.77 to 1.81 ˚K, while for ESTARFM, RMSE ranged from 1.69 to 2.63 ˚K. Similarly, the Peak Signal-to-Noise Ratio (PSNR) decreased from 23.20 to 21.17 in RF, and from 23.59 to 17.06 in ESTARFM. These findings highlight the importance of temporal interval as a critical factor in STF and reveal differing algorithm sensitivities to varying gaps. Overall, this study demonstrates the potential of LST-based spatiotemporal fusion for improving agricultural mapping and monitoring, emphasizing the need to consider temporal intervals when generating high-resolution temperature datasets over agricultural lands.