A view of the EM algorithm that justifies incremental, sparse, and other variants
Learning in graphical models
On sequential Monte Carlo sampling methods for Bayesian filtering
Statistics and Computing
Inference in Hybrid Networks: Theoretical Limits and Practical Algorithms
UAI '01 Proceedings of the 17th Conference in Uncertainty in Artificial Intelligence
Anomaly detection in streaming environmental sensor data: A data-driven modeling approach
Environmental Modelling & Software
Environmental Modelling & Software
Rao-blackwellised particle filtering for dynamic Bayesian networks
UAI'00 Proceedings of the Sixteenth conference on Uncertainty in artificial intelligence
A tutorial on particle filters for online nonlinear/non-GaussianBayesian tracking
IEEE Transactions on Signal Processing
Hi-index | 0.00 |
Radar-rainfall data are being used in an increasing number of real-time applications because of their wide spatial and temporal coverage. Because of uncertainties in radar measurements and the relationship between radar measurements and rainfall on the ground, radar-rainfall data are often combined with rain gauge data to improve their accuracy. However, while rain gauges can provide accurate estimates of rainfall, their data are sometimes corrupted with errors caused by the environment in which the gauges are deployed. This study develops a real-time method for identifying measurement errors in rain gauge data streams. This method employs a dynamic Bayesian network (DBN) model of the rain gauge data stream to sequentially forecast the next rain gauge measurement from both the rain gauge and weather radar data streams and a decision rule-based classifier to identify data errors. Because of the uncertainty in the relationship between the radar and rainfall measurements, this method uses a statistical learning method (expectation maximization) to determine the best parameters for this relationship, given an adaptively sized moving window of previous measurements. The performance of the error detector developed in this study is demonstrated using a precipitation sensor network composed of five telemetered tipping bucket rain gauges and a WSR-88D weather radar. Through an analysis using synthetic errors, the false alarm rate and false negative rate were calculated to be 0.90% and 1.5%, respectively.