Implementing a novel deep learning technique for rainfall forecasting via climatic variables: An approach via...

Area of Science:

• Climatology and Meteorology
• Artificial Intelligence in Environmental Science
• Agricultural Science and Technology

Background:

• Rainfall variability significantly impacts crop productivity and livelihoods in developing regions, particularly those reliant on rainfed agriculture.
• Accurate rainfall forecasting is essential for agricultural planning and mitigating the effects of extreme climatic conditions, yet it remains a complex challenge due to rainfall's dynamic nature.

Purpose of the Study:

• To develop and present a deep learning forecasting model utilizing an optimized Gated Recurrent Unit (GRU) neural network.
• To predict rainfall patterns in Pakistan using historical climate data from 1991 to 2020.
• To evaluate the model's accuracy and the influence of various climatic variables on rainfall prediction.

Main Methods:

• Extraction and fine-tuning of 30 years of climate data (1991-2020) for Pakistan, including outlier and extreme value elimination.
• Application of data normalization strategies to standardize variables without information loss.
• Implementation of an optimized Gated Recurrent Unit (GRU) neural network for deep rainfall forecasting.

Main Results:

• The proposed GRU model achieved high prediction accuracy, evidenced by minimal Normalized Mean Absolute Error (NMAE) and Normalized Root Mean Squared Error (NRMSE) compared to existing models.
• Correlation and regression analyses revealed a negative association between temperature and rainfall, and a positive association for air quality variables, with significant correlations observed across all seasons.
• The second and third quarters showed a higher association with rainfall, while air quality variables exhibited a lesser or no association in the first and second quarters.

Conclusions:

• The optimized GRU model demonstrates significant feasibility and accuracy in precise rainfall forecasting, even amidst volatile climatic conditions.
• Selected climatic variables, including temperature and air quality indicators, are strongly associated with rainfall patterns throughout the year, supporting their utility in predictive models.
• The study highlights the potential of deep learning approaches for improving agricultural resilience and climate change adaptation strategies in vulnerable regions.