On the parsimony, interpretability and predictive capability of a physically−based model in the optical domain for estimating soil moisture content

Soil moisture plays an important role in the transpiration, evaporation and plant growth processes at the land surface-atmosphere interface. Optical remote sensing has great potential for the retrieval of surface soil moisture content (SMC), with many empirical data-driven models or physical models developed to address this issue. Nevertheless, most data-driven models face the challenge of poor interpretability, while the application of many existing physical models is limited by complicated calibration steps. The aim of this work is to validate the potential of a physically-based approach based on the Kubelka-Munk (KM) radiative transfer theory to strike a balance between physical significance and practical applicability in the optical estimation of SMC. Specifically, an adequate and heterogeneous soil dataset in Jianghan Plain, China was used to calibrate the model wavelength by wavelength under laboratory conditions. The performance of the approach (at the optimal band) was compared with several commonly used methods. The effect of soil organic matter (SOM) on the estimation of SMC was also investigated by validating the model transferability between subsets with different SOM levels. Results showed that there were two local optimal bands at around 1460 and 1940 nm in the full band analysis of the approach, and the performance at around 1940 nm is better or comparable to linear regression, logarithmic regression, and spectral index models. Although partial least squares regression (PLSR) could achieve higher prediction accuracy with the enrichment of band information, this approach stood out for its balance of model parsimony with single-band calibration, model interpretability with the incorporation of physical mechanisms and predictive capability. More importantly, we found that the approach could enhance the spectral sensitivity in the water absorption region, avoid negative predictions at low SMCs, and reduce the interference effect of SOM on the estimation of SMC, probably due to the physical constraints inside the approach. This paper demonstrates the parsimony, interpretability, and predictive capability of the physically-based approach in the optical estimation of SMC, and provides new insights into the application of this approach in the airborne/satellite imaginary spectroscopy sensing of SMC.