This code repository is an attachment for the article in Remote Sensing: Onačillová, K.; Gallay, M.; Paluba, D.; Péliová, A.; Tokarčík, O.; Laubertová, D. Combining Landsat 8 and Sentinel-2 Data in Google Earth Engine to Derive Higher Resolution Land Surface Temperature Maps in Urban Environment. Remote Sens. 2022, 14, 4076. https://doi.org/10.3390/rs14164076.
This repository contains a folder "javascript_codes" where you can find:
A JavaScript Google Earth Engine (GEE) code "LST_downscaling_GEE_APP.js" used in the LST-downscaling GEE Application to downscale (sharpen) the Land Surface Temperature (LST) derived from Landsat thermal sensing using the spectral bands of Sentinel-2
To download the downscaled LST images use the GEE Code Editor version or directly the code provided in the "LST_downscaling_GEE_APP.js" code in the "javascript_codes" folder.
There are three different output types of the algorithm: (1) the main output is the downscaled LST 10 m with residuals, (2) bivariate scatter plots of LSTL8 vs. NDVIL8, NDBIL8, NDWIL8, (3) downscaling regression model. In the GEE code, the user will obtain the following outputs:
Landsat 8/9 RGB – true color composite for Landsat 8/9
This algorithm aims to downscale the coarse spatial resolution (100/30 m) Landsat 8/9 LST to finer spatial resolution (10 m) for more accurate mapping of LST. Three different indices, namely the normalized difference vegetation index (NDVI), built-up index (NDBI), and water index (NDWI) were used for disaggregation of Landsat LST (100/30 m) to 10 m Sentinel-2 spatial resolution using linear regression. The algorithm was developed in the cloud-based platform Google Earth Engine (GEE) using Landsat-8 and Sentinel-2 open access data. We conclude that the proposed downscaling model, by addressing the linear relationship of LST at coarse and fine spatial resolutions, can be successfully applied to produce high-resolution LST maps suitable for studies of the urban thermal environment at local scales. The performance was validated by the adjusted R2 returned at the stage of model development as well as by the validation of downscaled LST results using data logger measurements at 6 sites in the Košice city, Slovakia, to represent different types of land cover.
Based on the input parameters, the respective images of Landsat Surface Reflectance (SR) Collection 2 and Sentinel-2 Level 2A (L2A) collections are used to NDVI, NDBI, and NDWI spectral indices for Landsat 8/9 (NDVIL8, NDBIL8, NDWIL8) in 30 m resolution and for Sentinel-2 in a 10 m resolution, respectively. The LST is then calculated at a 30 m resolution using the Landsat 8/9 thermal band (B10) converted to brightness temperature in degrees Celsius (LSTL8). Then, the linear regression model between LSTL8 and the spectral indices NDVIL8, NDBIL8, NDWIL8 is used to calculate regression coefficients. Finally, these regression coefficients are used to calculate the downscaled LST using Sentinel-2 NDVI, NDBI, and NDWI spectral indices in a 10 m resolution. Regression residuals are resampled using bicubic interpolation, filtered using Gaussian convolution and added back to the downscaled LST. Also, Landsat 8/9 and Sentinel-2 natural color images (RGB) are generated to compare with the final LST layers. For more detailed information see the published article in Remote Sensing journal.