Calamita, G.1, Perrone, A.1, Brocca, L.2, Straface, S.1,3., 2017, Soil electrical resistivity for spatial sampling design, prediction and uncertainty modelling of soil moisture, Vadose zone journal 16 (2017). doi_10.2136/vzj2017.01.0022,
URL: http://www.cnr.it/prodotto/i/374693

Geophysics constitutes a possible approach to estimate soil moisture with cost-competitive and minimally invasive measurements carrying information about various soil physical properties. In this study, a geostatistical-based methodology for the integration of easily obtainable shallow soil electrical resistivity with soil moisture measurements was developed. Simultaneous and co-located surface measurements of soil moisture and electrical resistivity were performed during two field surveys in a test site located in central Italy. After an accurate characterization of the spatial auto-and cross-semivariance of soil moisture and electrical resistivity, an approach to design a soil moisture sampling strategy was set up. Soil moisture point predictions and uncertainties were estimated by ordinary co-kriging (OCK) interpolation and co-kriging simulation (CS), respectively. Results showed that at least 17 spatially regular soil moisture samples integrated with relatively denser soil electrical resistivity data via OCK are necessary to obtain a sufficiently accurate (RMSE <= 0.05) reconstruction of the average soil moisture at the test field. A comparison with the ordinary kriging interpolator, which takes into account only soil moisture samples, showed an improvement for OCK point prediction accuracy (RMSE from 0.049 to 0.039 and explained variance from 27 to 50%). The spatial uncertainty modeling empirically confirmed that geostatistical simulations furnish spatial uncertainty patterns more realistic and less dependent on the sampling scheme than interpolation approaches. Compared with sequential Gaussian simulations, which take into account only soil moisture data, the CS provided a model of local uncertainty with narrower probability intervals and lower prediction uncertainty.

Data from https://intranet.cnr.it/people/