International audience

Understanding the vertical distribution of soil moisture is crucial for making informed decisions in various applications, ranging from precision agriculture to hydrological modeling. Four machine learning algorithms, including random forest, extreme gradient boosting, deep learning, and support vector regression were employed to estimate the soil moisture profile from collected tower-based L-band and P-band brightness temperature observations in Victoria, Australia. The results showed that random forest outperformed the other algorithms, with root mean square error (RMSE) values of 0.03, 0.04, and 0.06 m3/m3for depths of 0-5 cm, 0-30 cm, and 0-60 cm, respectively

As an emerging technique, P-band (0.3-1 GHz) may improve soil moisture remote sensing compared to L-band (1.4 GHz) SMOS (Soil Moisture and Ocean Salinity) and SMAP (Soil Moisture Active Passive) missions, because of its greater moisture retrieval depth resulting from its longer wavelength. Consequently, a number of tower-based experiments were undertaken in Victoria, Australia, to understand and quantify potential improvements. The study reported here has extended the evaluation of the tau-omega model to a scenario with a dense corn canopy whose vegetation water content reached ~20 kg/m2, and compared the soil moisture retrieval performance at P- and L-band. Based on the locally calibrated parameters, the results from both the SCA (Single Channel Algorithm) and DCA (Dual Channel Algorithm) approaches presented a clear reduction in vegetation impact at P-band compared to L-band. While the root-mean-square error (RMSE) for P-band did not achieve the 0.04-m3/m3target accuracy of SMOS and SMAP, i.e., 0.054 m3/m3for the SCA and 0.074 m3/m3for the DCA, this performance can be regarded as acceptable considering the extremely high vegetation water content. In comparison, the RMSEs at L-band were larger than 0.1 m3/m3for both the SCA and the DCA approaches. Additionally, DCA performed better in correlation coefficient and unbiased RMSE, while SCA performed better in RMSE at P-band due to the larger bias when using DCA. Moreover, the calibrated vegetation parameters at P-band were found to apply to broader conditions than those at L-band, likely due to the reduced vegetation impact.

Root zone soil moisture and its distribution throughout the profile play an important role in agricultural productivity and drought monitoring. An inversion scheme including the stratified coherent models of Njoku and Wilheit were employed to retrieve the daily soil moisture profile at 6 AM from simulated L-band and P-band radiometry observations for April 2019 in Cora Lynn, Victoria, Australia. Different levels of noise up to 4 K were imposed in this synthetic study. The average RMSE of retrieved soil moisture at the surface, middle, and bottom (60 cm) of the profile for the Njoku (Wilheit) model were 0.01 (0.04), 0.04 (0.06), and 0.05 (0.07) (all in m3/m3) when a second-order polynomial function was considered as the representative of the soil moisture profile.

The moisture retrieval depth is commonly held to be the approximately top 5 cm at L-band (~21-cm wavelength/1.41 GHz), which is seen as a limitation for hydrological applications. A widely held view is that this moisture retrieval depth increases with wavelength, ranging approximately from one-tenth to one-fourth of the wavelength. Accordingly, P-band (~40-cm wavelength/0.75 GHz) is under investigation for soil moisture observation over a deeper layer of soil. However, there is no accepted method for predicting the moisture retrieval depth, and there has been no study to confirm that the actual retrieval depth at P-band is indeed deeper than that achieved at L-band. Consequently, this research has estimated the moisture retrieval depth from theory and compared with empirical evidence from tower-based observations. Model predictions and experimental observations agreed that P-band has the potential to retrieve soil moisture over a deeper layer (~7 cm) than L-band (~5 cm) while maintaining the same correlation. However, an alternate interpretation of experimental results is that P-band has a larger correlation with soil moisture (accuracy of retrieval) than L-band but for the same 5-cm moisture retrieval depth. The results also demonstrated the increasing trend of the moisture retrieval depth for increasing wavelength, with the potential to achieving a moisture retrieval depth greater than 10 cm for P-band below 0.5 GHz. Importantly, model predictions showed that moisture retrieval depth was not only dependent on soil moisture content and observation frequency, but also the moisture gradient of the profile.

Soil Moisture is an important geophysical variable that needs reliable quantification for applications in hydrology, meteorology and agriculture. L-band radiometry has proved to be one of the best methods in soil moisture estimation using microwave signals. However, they provide measurements that correspond to a shallow depth of 5 cm and are also affected by the presence of overlaying vegetation and roughness. In contrast, P-band radiometry is expected to provide moisture information on a deeper layer of soil. Moreover, these lower frequency measurements are expected to be less affected by soil roughness and vegetation contributions. Consequently, this pilot study uses the Polarimetric P-band Multibeam Radiometer (PPMR) at 740 MHz to evaluate the response of the P-band radiometer over a realistic range of surface conditions at the field scale. A preliminary framework of P-band Microwave Emission of the Biosphere (P-MEB) has been developed as a forward model that simulates brightness temperature from soil moisture and other ancillary data collected from the field. This paper presents the model for the bare soil condition observed during June 2018 to August 2018. The results show that H-polarised PPMR data has better correlation to the soil moisture over a depth of 10 cm than the V-polarized PPMR data. A model is under improvement by incorporating a more suitable effective temperature formulation.