Scholarlink

<p>The major milestone for modern meteorology started with the invention of weather instruments. However, ancient meteorological knowledge for weather prediction is lacking in modern techniques. This paper presents a detailed overview of rainfall prediction specified in <i>Brihat Samhita</i>, an ancient Indian text. An attempt has been made to incorporate the parameters from <i>Brihat Samhita</i> into machine learning (ML) techniques to predict monsoon rainfall across India and its five sub-regions for year 2014 and validate with India Meteorological Department (IMD) observations; North-West (NW) India, Central (CEN) India, North-East (NE) India, South Peninsular (SP) India and North-Central (NC) India. Three models, namely Linear Regression (LR), Support Vector Machine (SVM) and Random Forest (RF), have been trained and used for prediction. The model’s performance is evaluated using MAE, MSE, RMSE, and <i>R</i><sup>2</sup>. The predictions are validated against daily and monthly observations for August 2014. Overall, RF demonstrated the most robust and generalizable performance with MAE between 2.54 and 6.33 mm, MSE between 10.12 and 72.99 mm<sup>2</sup>, RMSE between 3.18 and 8.54 mm and <i>R</i><sup>2</sup> between 0.12 and –1.01. RF also offered more consistent and realistic outputs in terms of rainfall variation than LR and SVM in most regions. This study suggests that the RF model performs better and could be very useful for capturing the sequential nature of monsoon rainfall and improving long-term forecasting capabilities.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Overview of Brihat Samhita rainfall prediction methodology over the Indian region and its utilization by machine learning technique

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Jaya Yadav,
  • Ashim Kumar Mitra,
  • Kailash Chandra Tiwari

摘要

The major milestone for modern meteorology started with the invention of weather instruments. However, ancient meteorological knowledge for weather prediction is lacking in modern techniques. This paper presents a detailed overview of rainfall prediction specified in Brihat Samhita, an ancient Indian text. An attempt has been made to incorporate the parameters from Brihat Samhita into machine learning (ML) techniques to predict monsoon rainfall across India and its five sub-regions for year 2014 and validate with India Meteorological Department (IMD) observations; North-West (NW) India, Central (CEN) India, North-East (NE) India, South Peninsular (SP) India and North-Central (NC) India. Three models, namely Linear Regression (LR), Support Vector Machine (SVM) and Random Forest (RF), have been trained and used for prediction. The model’s performance is evaluated using MAE, MSE, RMSE, and R2. The predictions are validated against daily and monthly observations for August 2014. Overall, RF demonstrated the most robust and generalizable performance with MAE between 2.54 and 6.33 mm, MSE between 10.12 and 72.99 mm2, RMSE between 3.18 and 8.54 mm and R2 between 0.12 and –1.01. RF also offered more consistent and realistic outputs in terms of rainfall variation than LR and SVM in most regions. This study suggests that the RF model performs better and could be very useful for capturing the sequential nature of monsoon rainfall and improving long-term forecasting capabilities.