Predicting rainfall using machine learning, deep learning, and time series models across an altitudinal gradient in...

Area of Science:

• Meteorology
• Data Science
• Environmental Science

Background:

• Accurate rainfall prediction is critical for India, impacting agriculture and disaster preparedness.
• Limited meteorological data in the North-Western Himalayas necessitates improved forecasting models.
• Monsoon rainfall is crucial, supporting approximately 60% of India's agricultural land.

Purpose of the Study:

• To enhance rainfall prediction accuracy in the North-Western Himalayas using advanced ML and DL algorithms.
• To compare the performance of various ML, DL, and time series models for rainfall forecasting.
• To investigate the influence of altitudinal gradients on model accuracy.

Main Methods:

• Application of machine learning (ML) algorithms: random forest (RF), support vector regression (SVR), artificial neural network (ANN), k-nearest neighbour (KNN).
• Implementation of deep learning (DL) algorithms: long short-term memory (LSTM), bi-directional LSTM, deep LSTM, gated recurrent unit (GRU), simple recurrent neural network (RNN).
• Utilized time series techniques: autoregressive integrated moving average (ARIMA), trigonometric, Box-Cox transform, arma errors, trend, and seasonal components (TBATS).
• Assessed model performance using meteorological data from 1980-2021 across six altitudinal weather stations.

Main Results:

• Deep learning (DL) methods demonstrated the highest accuracy in rainfall prediction, followed by ML and time series techniques.
• Bi-directional LSTM and LSTM models showed superior performance among DL algorithms.
• Artificial neural network (ANN) was the most accurate among ML algorithms.
• Model accuracy was significantly influenced by altitude, indicating a need for more data in mountainous regions.

Conclusions:

• Advanced DL and ML algorithms offer significant improvements in rainfall prediction accuracy over traditional methods.
• Altitude is a critical factor affecting the precision of rainfall forecasting models in the North-Western Himalayas.
• Further data collection from diverse altitudinal stations is recommended to enhance forecasting reliability.