Open Access

ARTICLE

Evaluating the Capability of Sentinel-3 as an Alternative to MODIS for Downscaling High Spatiotemporal Resolution LST Data Using ESTARFM and XGBoost Models

Nahid Haghshenas, Ali Shamsoddini^*

Department of RS and GIS, Tarbiat Modares University, Tehran, Iran

* Corresponding Author: Ali Shamsoddini. Email:

(This article belongs to the Special Issue: Time Series Remote Sensing Data Processing and Applications)

Revue Internationale de Géomatique 2026, 35, 249-272. https://doi.org/10.32604/rig.2026.076139

Received 14 November 2025; Accepted 23 February 2026; Issue published 21 May 2026

Abstract

This study aimed to evaluate the potential of Sentinel-3 as an alternative to Moderate Resolution Imaging Spectroradiometer (MODIS) for generating high spatiotemporal resolution land surface temperature (LST) data. The Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM) and the machine-learning-based Extreme Gradient Boosting (XGBoost) algorithm were independently assessed for fusing MODIS–Landsat and Sentinel-3–Landsat data. This comparison enabled the evaluation of each model’s capability to reconstruct spatiotemporal LST variations and assess the performance of the two sensors in the fusion process. The results showed that XGBoost outperformed ESTARFM in capturing complex and heterogeneous LST patterns, particularly under strong diurnal fluctuations and phenological differences. The mean Root Mean Square Error (RMSE) values for MODIS were 1.87 Kelvin (K) and 2.62 K using XGBoost and ESTARFM, respectively, while for Sentinel-3, they were 1.73 and 2.52 K, confirming XGBoost’s superior accuracy for both sensors. Sentinel-3, owing to its higher spatial resolution, better radiometric quality, earlier overpass time closer to Landsat-8/9 acquisitions, and improved angular effect control, more accurately reconstructed spatial variations and daily temperature dynamics. In contrast, MODIS, with its broader temporal coverage and larger dynamic range, provided more stable Spatiotemporal Fusion (STF) results with slightly higher mean values. The model transferability analysis showed that training models with MODIS data and applying them to Sentinel-3 yielded higher accuracy than the reverse configuration, highlighting the importance of sensor selection and model generalizability. Overall, the findings indicate that Sentinel-3 can serve as a viable alternative to MODIS for STF of LST data under data gaps or reduced-quality time series. Moreover, integrating data from both sensors can ensure the continuity and stability of downscaled LST products and mitigate data limitations in long-term monitoring.

---

Keywords

---