A new non-parametric Bayesian based space upscaling method for in-situ soil moisture sampling

In soil moisture remote sensing products ground verification, the soil moisture in-situ sampling is heterogeneous and not consistent with the remote sensing pixel scale. So, it is very important that extensive sampling soil moisture of the surface heterogeneous and effective upscaling multi-point in situ soil moisture to remote sensing pixel scale. However, it is very difficult to collect them extensively if the soil moisture signal is complicated or dynamically changing. Considering the sparsity of soil moisture and the effects of observation noise, this paper models the linear programming problem between the soil moisture remote sensing pixel and in-situ sampling data using hierarchical non-parametric Bayesian linear regression. This model does not assume that the specific distribution of regression parameters which will be learned adaptively by the nonparametric Bayesian method. In addition, we implement the Dirichlet process to exploit the spatial similarity of the in-situ sampling data of soil moisture, thus to improve the spatial upscaling accuracy. Due to the Dirichlet process without explicit mathematical expression, the model of the posteriori probability distribution is very hard to deal with. To this end, the Gibbs sampling scheme based on MCMC (Markov Chain Monte Carlo) is adopted to infer the optimal regression weighting coefficient, and the spatial scale of the soil moisture in situ sampling is effectively upscaled. The experimental results show that the spatial upscaling method of non-parametric Bayesian linear regression is closer to the observed remote sensing pixel scale and better than the state of the art Bayesian method and Kriging method.