Comparative analysis of SAOCOM and Sentinel-1 data for surface soil moisture retrieval using a change detection method in a semiarid region (Douro River's basin, Spain)

Benedetta Brunelli,Francesco Mancini

DOI: https://doi.org/10.1016/j.jag.2024.103874

IF: 7.5

2024-04-29

International Journal of Applied Earth Observation and Geoinformation

Abstract:The growing interest in low-frequency SAR for soil parameter retrieval has led to the development of new active L-band satellites, that will provide novel surface soil moisture products and retrieval possibilities; however, due to data unavailability so far, limited applications have investigated the use of change detection models using L-band satellite SAR data. Since July 2020, high revisit time, high-resolution acquisitions by the Satélite Argentino de Observación COn Microondas (SAOCOM) Argentinian-Italian constellation have become accessible over Europe. Therefore, this research presents an investigation of the potential of multi-temporal L-band SAOCOM-1 for monitoring soil moisture variations underneath low and sparse agricultural vegetation. Moreover, it proposes a procedure for the mitigation of roughness contribution, by exploiting the entropy parameter derived from the dual-polarimetric decomposition. L-band sensitivity to soil moisture has been jointly evaluated in respect of Sentinel-1 C-band data by (1) comparing the temporal profiles of the backscattering coefficient, γ0 , at VV and VH polarization, with the support of decomposition parameters (entropy and ᾱ ), NDVI and precipitation data; (2) regression analysis with in situ soil moisture measurements, obtained by the REMEDHUS network in the Douro River basin (Spain); 3) evaluating the soil moisture retrievals obtained at C- and L- band using a change detection method. Finally, the effectiveness of the roughness normalization procedure for SAOCOM data has been validated using in situ data. L-band co-polarized γ0 has proved to be the best configuration for soil moisture inversion, being relatively insensitive to vegetation, as demonstrated by decomposition results and trend interpretation. Overall, regressions detected an R 2 22% higher at L-band than C-band, with values up to 0.74 for VV ( R̄2 =0.32) and up to 0.47 for the VH band ( R̄2 =0.14). Co-polarized data obtained R 2 on average 62.1% and 74.7% higher for SAOCOM and Sentinel-1. The retrieval models show an ubRMSD of 7.1% for SAOCOM data and 8.3% for Sentinel-1. The application of the proposed roughness normalization procedure to SAOCOM led to an ubRMSD of 6.7% improving the retrieved soil moisture trend by 7.9%. This exploratory analysis demonstrated SAOCOM data potential for soil moisture mapping and would serve as a foundation for more advanced retrieval procedures.

remote sensing