Improving Temporal Resolutions of Drought Predictions Using Knowledge Discovery from Database Approach

Abstract:

Attributed to climatic change and uncertainty of weather conditions, drought has become a recurrent phenomenon in many parts of the world. Even though there had been a number of efforts for predicting drought in the past, there were limited approaches for improving the temporal resolutions. Taking this gap into account, the objective of this article was to develop an approach to improve the temporal resolutions of drought predictions using knowledge discovery from database (KDD) approach. The focus of the study was on using higher temporal resolution datasets from satellite, climate, oceanic and biophysical domain sources. A total of 24 year historical data from the years 1983 to 2006, with spatial resolution of 8 km, for selected attributes were used in constructing regression tree models. For constructing the models, 80% of the data were used for training and 20% for testing. After improving the temporal resolution from 30 days to 10 days temporal resolutions, the time lag prediction accuracy has significantly improved from 95% to 99% correlation coefficient. From these experimental analyses, it was concluded that high temporal resolution dataset has a significant impact on the overall accuracy of the prediction models. The result can be used by decision makers at different levels for their early update of drought episodes for their smart decisions in mitigating drought.

Key words: Dekadal Data, Drought Monitoring, Drought Prediction, Model, Regression Tree