Comparative Analysis of Spatio-Temporal Data Fusion Algorithms

Read full text Summarise with VIRAT
Authors: Taiwo Amodu
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Spatio-temporal data fusion (STF) algorithms were developed to improve the quality of satellite images obtained from sensors with different spatial and temporal resolutions. Concretely, the goal is to take advantage of the strengths of satellites (such as Sentinel-2 & Landsat) that can have a very high spatial resolution but long monitoring cycles by combining them with images from satellites that have lower spatial resolutions but higher monitoring frequency (such as Sentinel-3 & MODIS). By integrating these datasets, STF methods generate synthetic images that are both spatially rich and temporally dense. This study focuses on three widely used STF algorithms: the Spatial and Temporal Adaptive Reflectance Model (STARFM), the Enhanced Spatial and Temporal Adaptive Reflectance Model (ESTARFM), and the Flexible Spatio-Temporal Data Fusion (FSDAF) method. We did a comparative evaluation of these algorithms using data from Sentinel-2 and Sentinel-3, because of their complementary spatial and temporal resolutions and their relevance as modern satellite sensors. We applied these algorithms across a range of landscape settings, including mountainous terrain, agricultural mosaics, and other heterogeneous environments, to assess how algorithm performance varies under different geographic and surface conditions. Our results indicate that ESTARFM is the best-performing algorithm among the three. It consistently outperforms STARFM and FSDAF across all study sites and across multiple spectral bands. On average, ESTARFM reduces root mean square error (RMSE) by approximately 40% relative to STARFM and FSDAF, while increasing structural similarity Index (SSIM) by about 20.2% and the correlation coefficient (CC) by up to 28.5%. We can attribute this superior performance to the tendency of STARFM and FSDAF to generate overly smoothed outputs, which leads to the loss of fine spatial detail in fragmented and topographically complex landscapes. In contrast, ESTARFM better preserves spatial structure under such conditions. In addition, STARFM and FSDAF make limited use of temporal correlation between observations, whereas ESTARFM explicitly incorporates imagery from three different time points to model temporal change. This additional temporal information enables a more realistic representation of surface dynamics and largely explains the improved accuracy achieved by ESTARFM. Overall, our findings show that, regardless of the algorithm used, prediction accuracy is strongly affected by landscape heterogeneity, temporal sampling gaps, and the rate of surface change. Increased heterogeneity, longer temporal intervals between input images, and more rapid intrinsic changes in land surface conditions all make accurate STF prediction more challenging.

Fields of Study & Subjects

Advanced Image Fusion Techniques Remote Sensing in Agriculture Satellite Image Processing and Photogrammetry

Publication details

Journal / source
Portuguese National Funding Agency for Science, Research and Technology (RCAAP Project by FCT)
Publication date
26 February 2026
Language
EN
Document type
Thesis