Research on comprehensive drought index prediction model based on CNN-LSTM

Area of Science:

• Environmental Science
• Climate Science
• Data Science

Background:

• Ordos, a key region in Inner Mongolia, faces frequent drought disasters hindering economic growth.
• Droughts are a critical constraint on regional development due to geographical and climatic factors.

Purpose of the Study:

• To construct a Regional Comprehensive Drought Index (RSDI) for Ordos from 2001-2020.
• To evaluate the predictive performance of deep learning models (CNN, LSTM, CNN-LSTM) for RSDI.
• To enhance drought monitoring and prediction in the Ordos region.

Main Methods:

• Constructed the Regional Comprehensive Drought Index (RSDI) using a random forest model.
• Utilized temperature, precipitation, NDVI, soil moisture, land use, and DEM as input variables.
• Employed Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), and a hybrid CNN-LSTM model for RSDI prediction.

Main Results:

• The constructed RSDI showed a highly significant positive correlation with SPI (r > 0.90, P < 0.01).
• The hybrid CNN-LSTM model demonstrated superior predictive and fitting performance compared to individual CNN and LSTM models.
• The CNN-LSTM model achieved reduced prediction errors, with RMSE decreasing by 0.30 and 0.22, and MAE by 0.24 and 0.13.

Conclusions:

• The hybrid CNN-LSTM model offers improved accuracy for regional drought index prediction.
• The study provides a valuable technical reference for drought monitoring and prediction in Ordos.
• Enhanced drought prediction capabilities are crucial for mitigating the impact of drought catastrophes in the region.