Meteorological drought severity forecasting utilizing blended modelling.

Area of Science:

• Environmental Science
• Data Science
• Meteorology

Background:

• Droughts are increasing in frequency and intensity due to climate variation, causing significant damage to ecosystems, agriculture, and water resources.
• Conventional drought forecasting methods struggle with complex time-space correlations in meteorological data.
• Accurate drought prediction is imperative for effective disaster management and resource planning.

Purpose of the Study:

• To develop a highly accurate drought forecasting model by integrating advanced machine learning techniques.
• To leverage the strengths of Extreme Gradient Boosting (XGBoost), Long Short Term Memory (LSTM), and Tabular Network (TabNet) for improved drought severity prediction.

Main Methods:

• An ensemble model was created by combining XGBoost for feature selection, LSTM for temporal pattern analysis, and TabNet for attention-based insights.
• The model utilized a large meteorological dataset including temperature, precipitation, humidity, and wind speed.
• Performance was evaluated using metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination (R²), Nash-Sutcliffe Efficiency (NSE), and Kling-Gupta Efficiency (KGE).

Main Results:

• The proposed ensemble model demonstrated superior performance compared to individual models.
• Achieved the lowest RMSE (0.6582) and MAE (0.5377), and the highest R² (0.5069).
• Yielded the best NSE (0.5107) and KGE (0.6039), confirming its effectiveness in drought severity forecasting.

Conclusions:

• The ensemble model significantly outperforms traditional methods in drought forecasting.
• This advanced approach aids in providing early drought warnings and supports water conservation planning.
• The integration of XGBoost, LSTM, and TabNet offers a robust solution for predicting drought severity.