Synergy Between Sentinel-MSI and Landsat-OLI to Support High Temporal Frequency for Soil Salinity Monitoring in an Arid Landscape

Abstract

The free-available data acquired with multispectral instruments (MSI) onboard Sentinel-2 satellites and Operational Land Imager (OLI) installed on Landsat 8 satellite significantly advances the virtual constellation paradigm for Earth observing and monitoring with medium spatial resolutions ensuring a revisiting interval time less than 5 days. Although these instruments are designed to be similar, they have different spectral response functions and different spectral and spatial resolutions, and, therefore, their data probably cannot be reliably used together. In this chapter, we analyzed exclusively the impact of dissimilarities caused by spectral response functions between these two sensors for high temporal frequency for soil salinity dynamic monitoring in an arid landscape. Knowing that the shortwave infrared (SWIR) spectral bands are the most appropriate for soil salinity discrimination, modeling, and monitoring, only the land surface reflectances in the SWIR spectral bands are considered and converted to the Soil Salinity and Sodicity Index (SSSI) and to the semiempirical predictive model (SEPM) for soil salinity mapping. These three products were compared, and the impact of the sensors’ (OLI and MSI) spectral response function differences was quantified. To achieve these, analysis was performed on two pairs of images acquired in July 2015 and August 2017 with 1-day difference between each other over the same study area, which is characterized by several soil salinity classes (i.e., extreme, very high, high, moderate, low, and nonsaline). These images were not cloudy, without shadow, and not contaminated by cirrus. They were radiometrically and atmospherically corrected, and bi-directional reflectance difference factors (BRDF) were normalized. To generate data for analysis, similarly to Landsat-OLI, Sentinel-MSI images were resampled in 30 m pixel size considering UTM projection and WGS84 datum. The comparisons of the derived products were undertaken using regression analysis (p ≤ 0.05) and root mean square difference (RMSD). In addition to the visual analysis, kappa coefficient was also used to measure the degree of similarity between the derived salinity maps using SEPM. The results obtained demonstrate that the two used pair’s dataset, acquired during 2 different years over a wide range of soil salinity degrees (2.6 ≤ EC-Lab ≤ 600 dS m−1), had very significant fits (R2 of 0.99 for the SWIR land surface reflectances and R2 ≥ 0.95 for SSSI and SEPM). Moreover, excellent agreement was observed between the two sensor products, yielding RMSD values less than 0.012 (reflectance units) for the SWIR bands and less than 0.006 for SSSI. For the SEPM, the calculated RMSD vary between 0.12 and 2.65 dS m−1, respectively, for nonsaline and extreme salinity classes, reflecting relative errors varying between 0.046 and 0.005 for the considered soil salinity classes. Statistical similarity between the derived salinity maps based on SEPM using kappa coefficient revealed an excellent agreement (0.94). Therefore, MSI and OLI sensors can be used jointly to characterize and to monitor accurately the soil salinity and its dynamic in time and space in arid landscape, provided that rigorous preprocessing issues (sensor calibration, atmospheric corrections, and BRDF normalization) must be addressed before.