Bayesian Combined Active/Passive (B-CAP) Soil Moisture Retrieval Algorithm

Abstract

This paper focused on exploiting remotely sensed active and passive observations over agricultural fields for soil moisture retrieval purposes. Co-polarized backscattering coefficients HH and VV and V-polarized brightness temperature TbV measurements were merged onto a Bayesian algorithm to enhance field-based retrieval estimates. The Bayesian algorithm relies on the use of active SAR to constrain passive information. It is assumed that observations are representative of an extent involving field sizes of about 800 m by 800 m, disregarding the scaling issues between the high resolution SAR pixel and the coarse resolution passive pixel. The integral equation model with multiple scattering at second order (IEM2M) and the ω-τ model were used as forward models for the backscattering coefficients and for the V-polarized brightness temperature, respectively. The Bayesian algorithm was assessed using datasets from the Soil Moisture Active Passive Validation Experiment 2012 (SMAPVEx12). Such datasets are representative of contrasting soil conditions since soil moisture spanned almost its whole feasible range from 0.10 to 0.40 cm3/cm3, at different observation geometries with incidence angles ranging from 35° to 55°. Also, the fairly large amount of measurements (97) made the dataset complete for assessment purposes. Soil moisture variability at field scale and dielectric probe error were accounted for in the comparison between retrieved estimates and in situ measurements. Performance metrics were used to quantify the agreement of the retrieval methodology to in situ information, and to assess the improvement in the combined methodology with respect to the single ones (active or passive). Overall, the root mean squared error (RMSE) showed an improvement from 0.08 to 0.11 cm3/cm3 (only active) or 0.03-0.12 cm3/cm3 (only passive, after bias correction) to 0.06-0.10 cm3/cm3 (combined), thus, demonstrating the potential of such combined soil moisture estimates. When analyzed each field separately, RMSE is less than 0.07 cm3/cm3 and correlation coefficient r is greater than 0.6 for most of the fields.