Spatial resolution enhancement of the land surface thermal field imagery based on multiple regression models on multispectral data from various space systems

The methodology has been developed for enhancing the spatial resolution of the land surface thermal field satellite imagery based on the following steps: coupling images in the visible, thermal, and radar ranges into the single multispectral data product; constructing regression models of the images’ relationship; performing the linear regression of the pseudo-thermal product with enhanced spatial resolution from the visible and radar ranges data. The methodology is implemented on the Google Earth Engine open cloud platform using the Earth Engine API and the software scripts created in the JavaScript language with the processing of multispectral image collections of various space systems at specified time intervals. The possibility of practical synthesis of the pseudo-thermal image with an enhanced spatial resolution of 10 m based on the thermal image with the resolution of 100 m and the multispectral composite with the layers’ resolution of 10 m and 30 m is shown. The technology has been developed for synthesis and calibration of the land surface temperature product with enhanced spatial resolution and daily data providing rate based on the brightness temperature product in the B10 band of Landsat 8 and linear regression on the MODIS, ASTER, and Sentinel 1 products with daily to moderate data providing rates. The software in JavaScript has been developed, and technology has been implemented in the interactive web service form with open access on the Google Earth Engine Apps cloud platform. The final data product provides the satisfactory relative root mean square error of the brightness temperature recovery of not more than 6 % according to the reference cross-calibration data of the B10 Landsat 8 band in the moderate thermal field (up to 100° C). The relative root mean square errors of the synthesized data according to the reference data on high-temperature sites (fire, hot lava) up to 28 % are due to the fact that the synthesized product contains information from high-temperature spectral bands (B07-B09 from ASTER), while the reference product (B10 from Landsat 8) does not contain such information. Technology implementation examples show that cross-calibration of the synthesized product can be performed during the year from March to October according to reference thermal images of natural or artificial objects. Objects selected for calibration must have stable thermal characteristics at the time of the satellite flight during the data acquisition period. Keywords: land surface temperature, brightness temperature, space resolution of imagery, multiply linear regression, heterogeneous multispectral data coupling, data providing rate, product cross-calibration, Google Earth Engine.