A pattern recognition based approach to consistency analysis of geophysical datasets

Remotely sensed data from satellites are often validated by comparing them against ground-based measurements which usually are relatively sparse. Conventional consistency analysis methods provide information on each data point individually and in relation to its neighbors. In this study, a consistency analysis method based on wavelet-based feature extraction and one-class support vector machines is proposed. This method performs a consistency assessment of the entire time series in relation to others and provides a spatial distribution of consistency levels. The presented method is tested on soil moisture product from Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) on board Aqua satellite for the years 2005–2006. Time series of in-situ soil moisture measurements from the USDA Soil Climate Analysis Network (SCAN) are used as training data. Spatial distribution of consistency levels are presented as consistency maps for a region, including the states of Mississippi, Arkansas, and Louisiana in the USA. These results are correlated with the spatial distributions of averaged quality control information, mean soil moisture, and the cumulative counts of dense vegetation. Moreover, the methodology is tested for its robustness by examining its sensitivity to the spatial distribution of the network of training data sites. Finally, seasonal consistency maps for soil moisture data are developed. The degree to which the satellite estimates agree with the in-situ measurements has been represented seasonally as consistency maps which are helpful in interpreting the overall quality of the soil moisture product retrieved from satellite observations.