Characterizing drought prediction with deep learning: A literature review.

Area of Science:

• Environmental Science
• Computer Science
• Climate Science

Background:

• Drought prediction is vital for mitigating environmental and human impacts.
• Deep learning offers advanced capabilities for complex prediction tasks.
• Understanding current methodologies is key to advancing drought forecasting.

Purpose of the Study:

• To review and characterize state-of-the-art deep learning techniques for drought prediction.
• To identify commonly used climate and vegetation indices in drought forecasting.
• To map the global distribution of research on deep learning for drought prediction.

Main Methods:

• Systematic literature review of deep learning applications in drought prediction.
• Analysis of frequently used climate indices: Standardized Precipitation Index (SPI) and Standardized Precipitation Evapotranspiration Index (SPEI).
• Identification of prevalent multispectral indices, notably the Normalized Difference Vegetation Index (NDVI).
• Examination of deep learning algorithms, with a focus on Long Short-Term Memory (LSTM) networks and hybrid approaches.

Main Results:

• SPI and SPEI are the most frequently employed climate indices.
• NDVI is the most utilized multispectral index for drought prediction.
• Asia and Oceania lead in research output, while America and Africa show limited publications.
• LSTM networks, both standalone and in hybrid models, are the dominant deep learning methods.

Conclusions:

• A significant research gap exists in applying deep learning and multispectral indices for drought prediction in America and Africa.
• Developing countries present an opportunity for advancing drought prediction research.
• Further investigation is needed to enhance drought forecasting capabilities in underrepresented regions.