Publications

Amazirh, A; Merlin, O; Er-Raki, S (2019). Including Sentinel-1 radar data to improve the disaggregation of MODIS land surface temperature data. ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING, 150, 11-26.

Abstract
The use of land surface temperature (LST) for monitoring the consumption and water status of crops requires data at fine spatial and temporal resolutions. Unfortunately, the current spaceborne thermal sensors provide data at either high temporal (e.g. MODIS: Moderate Resolution Imaging Spectro-radiometer) or high spatial (e.g. Landsat) resolution separately. Disaggregating low spatial resolution (LR) LST data using ancillary data available at high spatio-temporal resolution could compensate for the lack of high spatial resolution (HR) LST observations. Existing LST downscaling approaches generally rely on the fractional green vegetation cover (f(gv)) derived from HR reflectances but they do not take into account the soil water availability to explain the spatial variability in LST at HR. In this context, a new method is developed to disaggregate kilometric MODIS LST at 100 m resolution by including the Sentinel-1 (S-1) backscatter, which is indirectly linked to surface soil moisture, in addition to the Landsat-7 and Landsat-8 (L-7 & L-8) reflectances. The approach is tested over two different sites-an 8 km by 8 km irrigated crop area named "R3" and a 12 km by 12 km rainfed area named "Sidi Rahal" in central Morocco (Marrakech) on the seven dates when S-1, and L-7 or L-8 acquisitions coincide with a one-day precision during the 2015-2016 growing season. The downscaling methods are applied to the 1 km resolution MODIS-Terra LST data, and their performance is assessed by comparing the 100 m disaggregated LST to Landsat LST in three cases: no disaggregation, disaggregation using Landsat f(gv) only, disaggregation using both Landsat f(gv) and S-1 backscatter. When including f(gv), only in the disaggregation procedure, the mean root mean square error in LST decreases from 4.20 to 3.60 degrees C and the mean correlation coefficient (R) increases from 0.45 to 0.69 compared to the non-disaggregated case within R3. The new methodology including the S-1 backscatter as input to the disaggregation is found to be systematically more accurate on the available dates with a disaggregation mean error decreasing to 3.35 degrees C and a mean R increasing to 0.75.

DOI:
10.1016/j.isprsjprs.2019.02.004

ISSN:
0924-2716