Tornado Forecasting with Multiple Markov Boundaries

摘要: Reliable tornado forecasting with a long-lead time can greatly support emergency response and is of vital importance for the economy society. The large number meteorological variables in spatiotemporal domains complex relationships among remain top difficulties forecasting.Standard data mining approaches to tackle high dimensionality are usually designed discover single set features without alternating options domain scientists select more reliable physical interpretable variables.In this work, we provide new solution use concept multiple Markov boundaries local causal discovery identify sets precursors forecasting. Specifically, our algorithm first confines extremely feature spaces small core space, then it mines from space that may equally contribute With precursors, able report predictive but practical precursors.An extensive empirical study conducted on eight benchmark historical near Oklahoma City, OK United States. Experimental results show identified help improve reliability catastrophic