Rainfall Dataset of India

The India Weather Forecast built a state-level standard rainfall forecast system using a multi-model ensemble approach with model outputs from five prominent worldwide NWP centers. Pre-assigned grid point weights based on anomalous correlations (CC) between values observed and predicted are established for each element model using two seasonal datasets, and multi provision of appropriate predictions are created in real-time similar resolution. Then, forecasts are created for each state node lying within a given district by averaging the ensemble prediction fields' values. We utilize a dataset including monthly rainfall data from 1901 to 2015 that has been preprocessed to remove missing values and perform feature engineering. To forecast rainfall, we use machine learning methods such as Random Forest, Lasso, and Ridge Regression models. To examine the accuracy of rainfall predictions, the models are trained and tested for year-month and state-by-state analyses.

印度气象部门采用多模型集合预报方法，整合全球五大顶尖数值天气预报（Numerical Weather Prediction）中心的模式输出结果，构建了国家级标准降雨预报系统。研究人员基于观测值与预报值之间的距平相关系数（CC，Correlation Coefficient），利用两套季节数据集为各要素模型预设格点权重，并以相同分辨率实时生成多组适配性预报结果。随后，通过对集合预报场的数值进行平均，为指定辖区内的各州节点生成降雨预报。本研究使用了1901年至2015年的逐月降雨数据集，该数据集已完成缺失值剔除与特征工程预处理。针对降雨预报任务，我们采用了随机森林（Random Forest）、套索回归（Lasso）以及岭回归（Ridge Regression）等机器学习模型。为评估降雨预报的准确率，我们分别按年月维度与逐州维度对模型开展训练与测试。