Using BMA To Fuse Multiple Satellites Soil Moisture Data

Soil moisture is an important physical effects of climate change. There are many ways to get soil moisture(SM). One of them is satellite microwave remote sensing method which is now the mainstream method since it provides real-time, high-resolution SM data. Influenced by the surface roughness, vegetation, and inversion algorithms, remote sensing satellite data are often biased. Bias correction of satellite SM data is the prerequisite of SM data fusion. Meanwhile, China Land Soil Moisture Data Assimilation System requires entire-area soil moisture data, which makes multiple satellites SM data fusion necessary.The traditional bias correction method is uniform segmentation CDF matching. We propose a method for non-uniform segmentation CDF matching for bias correction, which can provide better matching results while using less segments. Traditional soil moisture data fusion method is simple mean method, which averages each satellite SM data as the fusion result. But in fact, each satellite model is not always works well and have some uncertainty. In this paper, we use Bayesian Model Averaging(BMA) method to obtain a set of variable weights that quantifies the uncertainty of each model. Fusion results is obtained by the weighted average of multiple satellites SM data. BMA method takes posterior probability of each model as weights to describe the model uncertainty. BMA works better than mean method in SM data fusion. Finally we use shared regional BMA weights to get highresolution SM data in the whole area.