Drought Monitoring Using Data Mining Techniques: A Case Study for Nebraska, USA

Abstract

Drought has an impact on many aspects of society. To help decision makers reduce the impacts of drought, it is important to improve our understanding of the characteristics and relationships of atmospheric and oceanic parameters that cause drought. In this study, the use of data mining techniques is introduced to find associations between drought and several oceanic and climatic indices that could help users in making knowledgeable decisions about drought responses before the drought actually occurs. Data mining techniques enable users to search for hidden patterns and find association rules for target data sets such as drought episodes. These techniques have been used for commercial applications, medical research, and telecommunications, but not for drought. In this study, two time-series data mining algorithms are used in Nebraska to illustrate the identification of the relationships between oceanic parameters and drought indices. The algorithms provide flexibility in time-series analyses and identify drought episodes separate from normal and wet conditions, and find relationships between drought and oceanic indices in a manner different from the traditional statistical associations. The drought episodes were determined based on the Standardized Precipitation Index (SPI) and Palmer Drought Severity Index (PDSI). Associations were observed between drought episodes and oceanic and atmospheric indices that include the Southern Oscillation Index (SOI), the Multivariate ENSO Index (MEI), the Pacific/North American (PNA) index, the North Atlantic Oscillation (NAO) Index, and the Pacific Decadal Oscillation (PDO) Index. The experimental results showed that among these indices, the SOI, MEI, and PDO have relatively stronger relationships with drought episodes over selected stations in Nebraska. Moreover, the study suggests that data mining techniques can help us to monitor drought using oceanic indices as a precursor of drought.