Modeling drought with climate, satellite, and other land data sets using data mining techniques

摘要

Drought is a natural phenomenon that influences every aspect of society. Its impacts can be reduced through better monitoring. Meteorological observations are the primary sources of information used for drought monitoring. Recent advances in satellite-based remote sensing have greatly improved our ability to measure the important characteristics and impacts of drought-related disasters. A well-integrated use of climate and satellite-derived data improves drought monitoring. Data mining techniques such as regression trees can help in identifying the relationships and integrating satellite and climate data. The data used in the study include seasonally integrated satellite vegetation metrics, land cover, and climate drought indices such as the Standardized Precipitation Index (SPI) and Palmer Drought Severity Index (PDSI) calculated at a station level. This study targets identifying the vegetation conditions to assist in verifying and predicting drought conditions by analyzing the past and present weather parameters and observations of remote sensing data. The 'Cubist' data mining software is used to generate a rule-based predictive model to identify the satellite vegetation condition using the drought indices. The results suggest that there is a strong potential to use data mining in monitoring drought conditions, particularly related to impacts that affect agricultural production.