Scholarlink

<p>This study presents a spatiotemporal downscaling framework for MODIS land surface temperature (LST) using Sentinel-1 and Sentinel-2 data with machine learning techniques on the Google Earth Engine (GEE) platform. Random Forest regression was applied for spatial downscaling from 1&#xa0;km to 10&#xa0;m, and linear regression was used for temporal reconstruction. The method integrates multispectral, radar, vegetation, and elevation data to generate daily 10-m LST maps for Qazvin Province, Iran, in 2022. Validation against Landsat-derived LST and 51 ground-based observations demonstrated high accuracy, with RMSE values as low as 1.25&#xa0;K. Spatial error analysis revealed higher uncertainties in croplands due to irrigation, while urban and bare areas showed more stable performance. Seasonal evaluation showed model robustness in spring and autumn, with some limitations during irrigation (summer) and snow cover (winter) periods. The fully automated pipeline offers an efficient, scalable solution for producing high-resolution LST, supporting applications in climate monitoring, agriculture, and urban heat island analysis. Despite limitations in temporal data availability, this study demonstrates the feasibility of producing accurate daily LST data in diverse landscapes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advanced spatiotemporal downscaling of MODIS land surface temperature: utilizing Sentinel-1 and Sentinel-2 data with machine learning technique in Qazvin Province, Iran

  • 期刊论文

  • 发表日期：

    2025年7月23日

查看全文
  • Zohreh Faraji,
  • Abbas Kaviani,
  • Leila Khosravi

摘要

This study presents a spatiotemporal downscaling framework for MODIS land surface temperature (LST) using Sentinel-1 and Sentinel-2 data with machine learning techniques on the Google Earth Engine (GEE) platform. Random Forest regression was applied for spatial downscaling from 1 km to 10 m, and linear regression was used for temporal reconstruction. The method integrates multispectral, radar, vegetation, and elevation data to generate daily 10-m LST maps for Qazvin Province, Iran, in 2022. Validation against Landsat-derived LST and 51 ground-based observations demonstrated high accuracy, with RMSE values as low as 1.25 K. Spatial error analysis revealed higher uncertainties in croplands due to irrigation, while urban and bare areas showed more stable performance. Seasonal evaluation showed model robustness in spring and autumn, with some limitations during irrigation (summer) and snow cover (winter) periods. The fully automated pipeline offers an efficient, scalable solution for producing high-resolution LST, supporting applications in climate monitoring, agriculture, and urban heat island analysis. Despite limitations in temporal data availability, this study demonstrates the feasibility of producing accurate daily LST data in diverse landscapes.