Scholarlink

<p>Rainfall forecasting is essential for mitigating the impacts of climate change and plays a vital role in hydrology, agriculture, water resource management, and disaster preparedness. However, the stochastic, non-linear, and highly variable nature of rainfall makes accurate prediction a persistent challenge. Traditional statistical models such as ARIMA, GARCH, and TAR provide foundational insights but are constrained by assumptions of linearity and stationarity. Machine learning (ML) methods, including support vector machines, random forests, and gradient boosting, improved non-linear modeling but require extensive feature engineering with limited generalizability. Deep learning (DL) approaches such as RNNs, LSTMs, GRUs, and CNNs have advanced rainfall prediction through automatic feature extraction and spatiotemporal learning, though they remain computationally demanding and prone to overfitting. Recently, attention mechanisms and transformers have emerged as promising alternatives for capturing long-range dependencies, though. their application in rainfall studies is still nascent. This review synthesizes nearly 150 studies to compare statistical, ML, and DL methods, identify challenges such as data scarcity, non-stationarity, and interpretability, and highlight future research opportunities including hybrid frameworks, transfer learning, explainable AI, data fusion, and transformer-based models for robust and climate-aware rainfall forecasting.</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

A comprehensive review of machine learning and deep learning approaches for rainfall forecasting: current progress, challenges, and future directions

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Jignesh Senjaliya,
  • Vibhisha Vaghasia,
  • Sanjay M. Shah

摘要

Rainfall forecasting is essential for mitigating the impacts of climate change and plays a vital role in hydrology, agriculture, water resource management, and disaster preparedness. However, the stochastic, non-linear, and highly variable nature of rainfall makes accurate prediction a persistent challenge. Traditional statistical models such as ARIMA, GARCH, and TAR provide foundational insights but are constrained by assumptions of linearity and stationarity. Machine learning (ML) methods, including support vector machines, random forests, and gradient boosting, improved non-linear modeling but require extensive feature engineering with limited generalizability. Deep learning (DL) approaches such as RNNs, LSTMs, GRUs, and CNNs have advanced rainfall prediction through automatic feature extraction and spatiotemporal learning, though they remain computationally demanding and prone to overfitting. Recently, attention mechanisms and transformers have emerged as promising alternatives for capturing long-range dependencies, though. their application in rainfall studies is still nascent. This review synthesizes nearly 150 studies to compare statistical, ML, and DL methods, identify challenges such as data scarcity, non-stationarity, and interpretability, and highlight future research opportunities including hybrid frameworks, transfer learning, explainable AI, data fusion, and transformer-based models for robust and climate-aware rainfall forecasting.