Scholarlink

<p>Delivering high-quality rainfall forecasts remains a challenging yet critical task for agriculture. Accurate prediction of key characteristics of the rainfall season, including the onset, cessation, and duration, is essential for effective water management, crop planning, and agricultural resilience, particularly in climate-vulnerable regions. Various artificial intelligence techniques have been applied to forecast rainfall patterns; however, a clear classification of these methods and associated rainfall parameters is necessary to support informed decision making. This critical review (CR) provides a comprehensive analysis of AI approaches in rainfall forecasting, with a focus on data validation, prediction methods over different time horizons, and forecasting of onset, cessation, and duration parameters. Following PRISMA guidelines, we systematically examined AI applications in rainfall forecasting between 2010 and 2024, screening 1862 publications, of which 183 met the inclusion criteria. Among the rainfall forecasting models, hybrid models were the most commonly used (53.22%), followed by LSTM (9.14%), CNN (4.84%), and RF (4.3%). In particular, 55.89% of the studies were conducted in Asia, reflecting the complex terrain of the region and the high vulnerability to climate-related disasters. For timescale tasks, ANN and LSTM were predominantly applied for annual forecasts (3 occurrences each). Seasonal tasks often involved LSTM, RF, CNN, and ELM methods. On the monthly scale, WANN and LSTM were the most prominent models (5 occurrences each), followed by RNN (3 occurrences). RNN, Echo State Networks, and TDNN were used to assess the date of onset, cessation, and duration of the rainy season (3 occurrences). We propose a novel classification framework that maps AI models according to forecast horizon, target rainfall parameters, and algorithmic complexity. Our analysis also highlights persistent gaps in the standardization of methodologies for seasonal rainfall parameters and the limited adaptation of advanced AI models in tropical contexts.</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

Artificial intelligence methods for rainy season forecasting: a comprehensive analysis

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Bienvenu Sonon,
  • Charlemagne D. S. J. Gbemavo,
  • Romain Glèlè Kakaï

摘要

Delivering high-quality rainfall forecasts remains a challenging yet critical task for agriculture. Accurate prediction of key characteristics of the rainfall season, including the onset, cessation, and duration, is essential for effective water management, crop planning, and agricultural resilience, particularly in climate-vulnerable regions. Various artificial intelligence techniques have been applied to forecast rainfall patterns; however, a clear classification of these methods and associated rainfall parameters is necessary to support informed decision making. This critical review (CR) provides a comprehensive analysis of AI approaches in rainfall forecasting, with a focus on data validation, prediction methods over different time horizons, and forecasting of onset, cessation, and duration parameters. Following PRISMA guidelines, we systematically examined AI applications in rainfall forecasting between 2010 and 2024, screening 1862 publications, of which 183 met the inclusion criteria. Among the rainfall forecasting models, hybrid models were the most commonly used (53.22%), followed by LSTM (9.14%), CNN (4.84%), and RF (4.3%). In particular, 55.89% of the studies were conducted in Asia, reflecting the complex terrain of the region and the high vulnerability to climate-related disasters. For timescale tasks, ANN and LSTM were predominantly applied for annual forecasts (3 occurrences each). Seasonal tasks often involved LSTM, RF, CNN, and ELM methods. On the monthly scale, WANN and LSTM were the most prominent models (5 occurrences each), followed by RNN (3 occurrences). RNN, Echo State Networks, and TDNN were used to assess the date of onset, cessation, and duration of the rainy season (3 occurrences). We propose a novel classification framework that maps AI models according to forecast horizon, target rainfall parameters, and algorithmic complexity. Our analysis also highlights persistent gaps in the standardization of methodologies for seasonal rainfall parameters and the limited adaptation of advanced AI models in tropical contexts.