Development of a hierarchical Bayesian network algorithm for land surface data assimilation

Abstract

A hierarchical Bayesian network (HBN) algorithm is developed for data assimilation (DA) and tested with an instance of soil moisture assimilation from a hydrological model and ground observations. In essence, the HBN is a framework that can statistically describe Bayesian models and capture the dependencies in the models more realistically than non-hierarchical Bayesian models. In this work, DA divided into three levels – data, process, and parameter – and conditional probability models are defined for each level. The data model mainly deals with the scale differences of multi-source data in DA, while the process model is designed to take account of the non-stationary process. Moreover, both the temporal auto-correlation and the spatial correlation are considered in the process model. Soil moisture observations from the Soil Moisture Experiment in 2003 (SMEX03) and Variable Infiltration Capacity (VIC) model are sequentially assimilated with HBN. The result shows that the assimilation with HBN provides spatial and temporal distribution information of soil moisture and the assimilation result agrees well with the ground observations.

Acknowledgements

This work was supported by the Chinese National Project of Major Scientific and Technological Infrastructure, the ‘Airborne Remote Sensing System-Algorithms and Software for Science Products’, and the National Science Foundation's Arctic System Science programme (ARC 1023371).