Spatial Up-scaling Method Of Soil Moisture Based On Bayesian Model

Soil moisture is an important parameter of energy and water cycle.Effectively obtaining soil moisture can improve regional and global climate forecasting and geological disaster prediction effects,which has important implications for meteorology,hydrology,agricultural production and national economic construction.Due to the strong heterogeneity of soil moisture,the single site observations of single site are not representative,using the average values of the single site as the "true values" will produce a large error and the valid verification of remote sensing products cannot be effectively achieved in the sites deployed by Wireless Sensor Network(WSN).Based on Bayesian model,the in-situ data of soil moisture are collected by multi-site method,and the spatial up-scaling of soil moisture is studied.Spatial up-scaling methods of soil moisture based on variational Bayesian inference and non-parametric Bayesian are proposed.The main content of this thesis:(1)Based on variational Bayesian inference,a spatial up-scaling method of soil moisture is studied.This method is based on the expectation maximization algorithm(Expectation Maximization Algorithm,EM),and the Kullback-Leibler divergence(KL-divergence for short)is used to construct the spatial up-scaling model of soil moisture.The optimal parameters and maximum posterior probability of the model are solved by alternating iteration method.Finally,the validity of the spatial up-scaling method based on variational Bayesian inference is verified.(2)A spatial up-scaling method of soil moisture is studied based on the nonparametric Bayesian.Based on the nonparametric Bayesian and dictionary learning theory,a spatial up-scaling hierarchical model is constructed.This model takes advantage of the clustering nature of the Dirichlet process to group soil moisture data with similar information into one class to share the same dictionary atom between data blocks,and a sparse prior dictionary are constructed efficiently.The hierarchical nonparametric Bayesian spatial up-scalling model is solved by Gibbs sampling method of Markov Chain Monte Carlo(MCMC),and the optimal sparse coefficients are statistically inferred.Finally,the real in-situ soil moisture data are simulated and analyzed.The results show that the non-parametric Bayesian method proposed in this thesis can effectively infer the spatial observation scale of soil moisture stations.(3)The simulation results show that the mean square error of the two methods proposed in this thesis is lower than that of the spatial up-scaling method of soil moisture based on Bayesian linear regression.Through analysis and comparison,thespatial up-scaling method based on non-parametric has the lowest mean square error and the best effect,which shows that the method is more close to the observed soil moisture data and can reflect the "ground truth" more authentically.