Modeling soil moisture at the Valencia Anchor Station (VAS), Eastern Spain.

<p>The prediction of spatial and temporal variation of soil water content brings numerous benefits in the studies of soil. However, it requires a considerable number of covariates to be included in the study, complicating the analysis. Integrated nested Laplace approximations (INLA) with stochastic partial differential equation (SPDE) methodology is a possible approach that allows the inclusion of covariates in an easy way.</p><p>The current study has been conducted using INLA-SPDE to study soil moisture in the area of the Valencia Anchor Station (VAS), soil moisture validation site for the European Space Agency SMOS (Soil Moisture and Ocean Salinity). The data used were collected in a typical ecosystem of the semiarid Mediterranean conditions, subdivided into physio-hydrological units (SMOS units) which presents a certain degree of internal uniformity with respect to hydrological parameters and capture the spatial and temporal variation of soil moisture at the local fine scale. The use of the INLA-SPDE methodology presents the possibility to analyze the significance of different covariates having spatial and temporal effects and has allowed us to fit spatial and temporal hierarchical models that are too complicated to be fitted by maximum likelihood methods.</p><p>The models allow to analyze the influence of hydrodynamic properties on VAS soil moisture (texture, porosity/bulk density and soil organic matter and land use) filtering out the effect of spatial and temporal variation. With the goal of understanding the factors that affect the variability of soil moisture in the SMOS pixel (50 km x 50 km), five states of soil moisture are proposed (moisture values range from values near saturation to moisture values near wilting point). Regarding the different models for the different Soil Moisture states, a general pattern was detected where both porosity and organic matter are two significant elements in all the cases. The model with all covariates and spatial effect has the lowest DIC value. In addition, the correlation coefficient was close to 1 for the relationship between observed and predicted values.</p><p>The findings of this study demonstrate an advancement in that framework, demonstrating that it is faster than previous methodologies, provides significance of individual covariates, is reproducible, and is easy to compare with models. The use of this methodology permits to design more efficient sampling campaigns for future SMOS missions. In addition, it also allows to construct soil moisture maps in a more sensible and efficient way.</p><p>&#160;</p>