Published In
Proceedings of the International Conference on Data Mining (DMIN'15)
Document Type
Article
Publication Date
2015
Subjects
Drought forecasting -- Simulation methods, Long-range weather forecasting -- Simulation methods
Abstract
In this study, a statistical drought early warning method is proposed using novel machine learning algorithms, with the inclusion of multiple drought-related attributes from precipitation, satellite-derived land cover vegetation indices, and surface discharge. The forecast is made for the long-term hydrological drought in the region of Central Valley, California. The wavelet transform analysis is employed in combination with support vector regression and artificial neural network algorithms for improving the drought prediction effectiveness. The performance of the drought prediction is evaluated using three statistical metrics: Coefficient of Determination (R2 ), Root-Mean-Square Error (RMSE), and Mean-Absolute-Error (MAE). The results clearly indicate that using hybrid precipitation and satellite remotelysensed data, the proposed wavelet-coupled machine learning method can effectively predict long-term drought in the area of Central Valley California, over a lead time of 3 to 6 months, which is crucial for agricultural planning, reservoir management, and authorities’ allocation of water resources.
Persistent Identifier
http://archives.pdx.edu/ds/psu/19370
Citation Details
R. Tan and M. Perkowski. (2015). Wavelet-Coupled Machine Learning Methods for Drought Forecast Utilizing Hybrid Meteorological and Remotely-Sensed Data. In Proc. Conference on Data Mining, DMIN15.
Description
Presented at the 11th International Conference on Data Mining, July 2015, Las Vegas, Nevada.