报告详情
Can Hybrid and Deep Learning Models Outperform Climatology? A Multivariate Benchmark for Monthly Rainfall Forecasting in Semi-Arid India
编号：104
访问权限：仅限参会人
更新：2026-11-10 16:46:29 浏览：7次
Online
报告开始：2026年11月21日 16:45（Asia/Kolkata）
报告时间：15min
所在会场：[S5] Data Science and Intelligent Analytics [S5-2] Data Science and Intelligent Analytics
暂无文件
摘要
Natural Disaster like drought or flood preparedness in semi-arid region in India heavily relies on monthly rainfall forecasts. In the era of Artificial intelligence, comparison of machine learning (ML) and deep learning (DL) models rarely done to check whether they are outperforming climatological forecasting or not. This paper is benchmarking all different 28 model configurations i.e. climatology, seasonal-naïve, ARIMA and SARIMA baselines, four ML regressors, Five DL architectures (1D CNN, LSTM, GRU, CNN-LSTM and a transformer encoder) and alongside SARIMA residual hybrids. Univariate(rainfall) and multivariate (rainfall + six climate variables input) is been used in the dataset from 1982 to 2025 at three different timescales of forecasting 1,3 and 6 months. It’s been observed that the hybrid models were performing better and the best hybrid result with effectiveness from 3.4 % to 3.9 % reduction in root mean square error (RMSE) over climatological forecast across the three regions. Though the improvement was not significant at 5% level. Multivariate inputs reduce the RMSE of standalone models by 14% compared with univariate inputs. While the multivariate input models were more successful than the stand-alone models in reducing RMSE, none of the standalone models were successful in surpassing the performance of the models using climatological forecast. These results establish climatology as the benchmark that learning-based rainfall models must improve further, and indicate that further gains will require predictors that interannual signals such as ocean-atmosphere indices.
关键词
rainfall forecasting, semi-arid regions, hybrid models, deep learning, Transformer, climatology benchmark
报告人
Sumit Singha Chowdhury
PhD Scholar Manipal Academy of Higher Education稿件作者
全部评论
重要日期
-
会议日期
11月20日
2026
至11月22日
2026
-
10月31日 2026
初稿截稿日期
-
11月07日 2026
提前注册日期
-
11月22日 2026
注册截止日期
主办单位
United Societies of Science
承办单位
Pondicherry University
协办单位
IEEE Section
IEEE Madras Section
IEEE Madras Section
联系方式
- ICon INDIA 2026
- ic******@usssociety.org
发表评论