Abstract Rainfall variability has long posed challenges for prediction and analysis. This study explores the intersection of history and innovation by leveraging the classical work Brihat Samhita (BS), written by Varahamihira around 500 ad. Observations documented in the BS, including qualitative descriptions as well as categorical information were translated into measurable parameters which were utilised as input features for two approaches: knowledge-based forecasting, and machine learning (ML) based forecasting after data pre-processing over India and its five sub-regions namely, North-West (NW) India, North Central (NC), Central (CEN) India, North-East (NE/NEE) India, and South Peninsula (SP) India. The dataset spans from 2007 to 2017. The knowledge-based forecasting was validated using IMD observations for heavy rainfall events with an average success rate of 58.81%. Five ML models namely, random forest, XGBoost, long short-term memory (LSTM), bidirectional LSTM, and AutoML were taken into account under ML-based forecasting. To report the overall performance of ML-based forecasting, the evaluation metrics such as mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination (R2) were computed as the mean across the models and the regions. ML-based approach, validated against IMD observations, demonstrated daily rainfall prediction with average – MAE 3.33 mm, RMSE 4.49 mm, and R2 0.17. Both approaches open new pathways for advancing monsoon rainfall prediction for experimental operational forecast. Research highlghts <UnorderedList Mark="Bullet"> <ItemContent> Ancient knowledge preserved in Brihat Samhita remains relevant for rainfall prediction during monsoon even today. </ItemContent> <ItemContent> A unique attempt has been made to utilise knowledge from Brihat Samhita for rainfall prediction over the Indian region. </ItemContent> <ItemContent> Knowledge-based forecasting method achieved an average success rate of 58.8% against IMD observations for heavy rainfall events. </ItemContent> <ItemContent> Machine learning-based approach demonstrated daily rainfall prediction with average – MAE 3.33 mm, RMSE 4.49 mm, and R2 0.17. </ItemContent> <ItemContent> Incorporate more observations from Brihat Samhita to increase accuracy and robustness of this monsoon rainfall prediction method. </ItemContent> </UnorderedList>

Comparative assessment of knowledge-based and machine learning models using Brihat Samhita insights for monsoon rainfall forecasting

期刊论文
发表日期：

2026年6月23日

查看全文

Jaya Yadav,
Kailash Chandra Tiwari,
Ashim Kumar Mitra

摘要

Abstract

Rainfall variability has long posed challenges for prediction and analysis. This study explores the intersection of history and innovation by leveraging the classical work Brihat Samhita (BS), written by Varahamihira around 500 ad. Observations documented in the BS, including qualitative descriptions as well as categorical information were translated into measurable parameters which were utilised as input features for two approaches: knowledge-based forecasting, and machine learning (ML) based forecasting after data pre-processing over India and its five sub-regions namely, North-West (NW) India, North Central (NC), Central (CEN) India, North-East (NE/NEE) India, and South Peninsula (SP) India. The dataset spans from 2007 to 2017. The knowledge-based forecasting was validated using IMD observations for heavy rainfall events with an average success rate of 58.81%. Five ML models namely, random forest, XGBoost, long short-term memory (LSTM), bidirectional LSTM, and AutoML were taken into account under ML-based forecasting. To report the overall performance of ML-based forecasting, the evaluation metrics such as mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination (R²) were computed as the mean across the models and the regions. ML-based approach, validated against IMD observations, demonstrated daily rainfall prediction with average – MAE 3.33 mm, RMSE 4.49 mm, and R² 0.17. Both approaches open new pathways for advancing monsoon rainfall prediction for experimental operational forecast.

Research highlghts

Ancient knowledge preserved in Brihat Samhita remains relevant for rainfall prediction during monsoon even today.

A unique attempt has been made to utilise knowledge from Brihat Samhita for rainfall prediction over the Indian region.

Knowledge-based forecasting method achieved an average success rate of 58.8% against IMD observations for heavy rainfall events.

Machine learning-based approach demonstrated daily rainfall prediction with average – MAE 3.33 mm, RMSE 4.49 mm, and R² 0.17.

Incorporate more observations from Brihat Samhita to increase accuracy and robustness of this monsoon rainfall prediction method.