Downscaling land surface temperature to high resolution using Sentinel-2 and Landsat in Google Earth Engine: a case study in Da Nang City
- 1University of Science, Ho Chi Minh City, Vietnam; Vietnam National University, Ho Chi Minh City, Vietnam
Abstract
Relevance. High-resolution mapping of urban land surface temperature is essential for locating intra-urban heat hotspots, assessing heat-exposure inequities, and supporting heat-mitigation planning in rapidly growing cities such as Da Nang (Vietnam). However, satellite thermal infrared observations are typically much coarser than optical imagery (e.g., 30 m for Landsat LST), which limits neighborhood-scale analysis and introduces mixed-pixel smoothing in heterogeneous urban surfaces. Aim. To develop and evaluate a machine-learning downscaling workflow that produces 10 m urban land surface temperature maps by integrating Landsat 8/9 thermal information with Sentinel-2 multispectral predictors in Google Earth Engine for Da Nang city, Vietnam. Methods. Reference land surface temperature was retrieved from Landsat 8/9 (30 m) thermal bands using standard physical procedures. Sentinel-2 (10 m) spectral indices describing moisture, water and built-up surfaces (the normalized difference moisture index, the modified normalized difference water index using the second shortwave-infrared band, and the normalized difference built-up index) and Shuttle Radar Topography Mission topography were used as predictors. Feature screening and correlation analysis were applied to reduce redundancy. A random forest regression model was trained (70%) and tested (30%), and performance was assessed using the coefficient of determination (R2), mean absolute error, and root mean square error. Spatial patterns of 30 and 10 m land surface temperature were compared across representative surface types to examine mixed-pixel effects in heterogeneous urban areas. Results. Moisture-, water- and built-up related indices (NDMI, MNDWI2, NDBI) were the strongest predictors of land surface temperature, whereas the normalized difference vegetation index alone was insufficient in a heterogeneous urban setting. The 10 m land surface temperature product achieved good agreement with the Landsat reference (R²=0.84, MAE=1.14°C, RMSE=1.53°C) and revealed finer spatial detail over reservoirs, sports complexes and mixed land-cover zones. Independent validation with in-situ surface/air-temperature measurements is still needed to further constrain uncertainty and strengthen practical deployment.
For citation: Nguyen Tien Thanh, Lam Van Hao. Downscaling land surface temperature to high resolution using Sentinel-2 and Landsat in Google Earth Engine: a case study in Da Nang City. Bulletin of the Tomsk Polytechnic University. Geo Assets Engineering, 2026, vol. 337, no. 10, pp. 90–102. http://doi.org/10.18799/24131830/2026/10/5254
Related
- Decadal dynamics of surface water extent and quality in Fujian Province, China: An integrated optical-SAR remote sensing analysis using Landsat and Sentinel-1 time series (2015–2024)
- Steering Diffusion Priors with Sparse Observations for High-Resolution Temperature Downscaling
- Spatio-temporal precipitation downscaling at hourly resolution based on km-scale regional simulations
- Airborne observatory for high-resolution land surface temperature and spectral emissivity
- Toward fine-scale microwave monitoring of dryland vegetation: an alternative high spatial resolution vegetation optical depth retrieval algorithm from Sentinel-1 over the Sahel region
BibTeX
@article{nguyen2026downscaling,
title = {{Downscaling land surface temperature to high resolution using Sentinel-2 and Landsat in Google Earth Engine: a case study in Da Nang City}},
author = {Tien Thanh Nguyen and Lam Van Hao},
journal = {Известия Томского политехнического университета: Инжиниринг георесурсов},
year = {2026},
month = oct,
doi = {10.18799/24131830/2026/10/5254},
url = {https://doi.org/10.18799/24131830/2026/10/5254},
note = {CC BY-NC},
}
RIS
TY - JOUR
TI - Downscaling land surface temperature to high resolution using Sentinel-2 and Landsat in Google Earth Engine: a case study in Da Nang City
AU - Tien Thanh Nguyen
AU - Lam Van Hao
PY - 2026
DA - 2026/10/31
JO - Известия Томского политехнического университета: Инжиниринг георесурсов
DO - 10.18799/24131830/2026/10/5254
UR - https://doi.org/10.18799/24131830/2026/10/5254
AB - Relevance. High-resolution mapping of urban land surface temperature is essential for locating intra-urban heat hotspots, assessing heat-exposure inequities, and supporting heat-mitigation planning in rapidly growing cities such as Da Nang (Vietnam). However, satellite thermal infrared observations are typically much coarser than optical imagery (e.g., 30 m for Landsat LST), which limits neighborhood-scale analysis and introduces mixed-pixel smoothing in heterogeneous urban surfaces. Aim. To develop and evaluate a machine-learning downscaling workflow that produces 10 m urban land surface temperature maps by integrating Landsat 8/9 thermal information with Sentinel-2 multispectral predictors in Google Earth Engine for Da Nang city, Vietnam. Methods. Reference land surface temperature was retrieved from Landsat 8/9 (30 m) thermal bands using standard physical procedures. Sentinel-2 (10 m) spectral indices describing moisture, water and built-up surfaces (the normalized difference moisture index, the modified normalized difference water index using the second shortwave-infrared band, and the normalized difference built-up index) and Shuttle Radar Topography Mission topography were used as predictors. Feature screening and correlation analysis were applied to reduce redundancy. A random forest regression model was trained (70%) and tested (30%), and performance was assessed using the coefficient of determination (R2), mean absolute error, and root mean square error. Spatial patterns of 30 and 10 m land surface temperature were compared across representative surface types to examine mixed-pixel effects in heterogeneous urban areas. Results. Moisture-, water- and built-up related indices (NDMI, MNDWI2, NDBI) were the strongest predictors of land surface temperature, whereas the normalized difference vegetation index alone was insufficient in a heterogeneous urban setting. The 10 m land surface temperature product achieved good agreement with the Landsat reference (R²=0.84, MAE=1.14°C, RMSE=1.53°C) and revealed finer spatial detail over reservoirs, sports complexes and mixed land-cover zones. Independent validation with in-situ surface/air-temperature measurements is still needed to further constrain uncertainty and strengthen practical deployment. For citation: Nguyen Tien Thanh, Lam Van Hao. Downscaling land surface temperature to high resolution using Sentinel-2 and Landsat in Google Earth Engine: a case study in Da Nang City. Bulletin of the Tomsk Polytechnic University. Geo Assets Engineering, 2026, vol. 337, no. 10, pp. 90–102. http://doi.org/10.18799/24131830/2026/10/5254
SN - 2500-1019
SN - 2413-1830
ER -