IDWE_CHM (NRT_L)

A near-real-time (NRT) extension of the IDWE_CHM dataset with ongoing daily updates beyond 2023. This NRT product continues to apply the IDWE framework on incoming data, thereby extending the record in near real-time. For a comprehensive description of the project, please refer to:<br><b>An Incremental Dynamic Weighting Ensemble Framework for Long-Term and NRT Precipitation Prediction</b><br>https://figshare.com/projects/An_Incremental_Dynamic_Weighting_Ensemble_Framework_for_Long-Term_and_NRT_Precipitation_Prediction/241619<br>The IDWE_CHM dataset provides <b>four precipitation variables</b>, all derived from the ensemble framework but with slightly different modeling approaches:<b>ENS_Reg</b> – A purely regression-based merged precipitation estimate. This product is generated by optimally weighting and combining the input datasets (ERA5-Land, IMERG, GSMaP, etc.) using regression, without additional classification. It serves as a baseline for the IDWE approach.<b>ENS_RegCla1</b>, <b>ENS_RegCla2</b>, <b>ENS_RegCla3</b> – Three variants of a hybrid <i>regression-plus-classification</i> approach (collectively called <b>ENS_RegCla</b>). These are produced by first applying the regression merging (as in ENS_Reg) and then using a classification step to adjust the estimates. The classification is enhanced with incremental learning, meaning the algorithm learns from errors over time. These three variants may correspond to different configurations or epochs of incremental learning, and they generally show improved skill in capturing precipitation occurrence and extremes compared to a regression-only merge.The updates of IDWE_CHM (NRT_L) are temporally coordinated with those of the five datasets integrated in the fusion process, with explicit synchronization maintained for the ERA5-Land dataset (available at: https://cds.climate.copernicus.eu/datasets/reanalysis-era5-land?tab=overview), which exhibits ~5 days latency compared to other fused datasets.

本数据集为IDWE_CHM数据集的准实时（near-real-time, NRT）扩展版本，自2023年起持续每日更新。该准实时产品沿用IDWE框架对实时接入的数据进行处理，从而实现数据集的准实时延伸。如需了解该项目的完整细节，请参阅：<br><b>面向长期与准实时降水预测的增量动态加权集成框架</b>（An Incremental Dynamic Weighting Ensemble Framework for Long-Term and NRT Precipitation Prediction）<br>https://figshare.com/projects/An_Incremental_Dynamic_Weighting_Ensemble_Framework_for_Long-Term_and_NRT_Precipitation_Prediction/241619<br><br>IDWE_CHM数据集包含**四类降水变量**，均基于上述集成框架生成，但建模路径略有差异：<br><b>ENS_Reg</b>：纯回归类融合降水估算产品。该产品通过回归方法对输入数据集（ERA5-Land、IMERG、GSMaP等）进行最优加权融合，未引入额外分类步骤，可作为IDWE方法的基准对照方案。<br><b>ENS_RegCla1</b>、<b>ENS_RegCla2</b>、<b>ENS_RegCla3</b>：混合“回归+分类”方法的三个变体（统称为<b>ENS_RegCla</b>）。此类产品首先执行与ENS_Reg一致的回归融合流程，随后通过分类步骤对估算结果进行调整；其分类模块搭载增量学习机制，可随时间推移从预测误差中持续迭代学习。这三个变体对应增量学习的不同配置或训练周期，相较于纯回归融合方案，三者在捕捉降水发生事件与极端降水事件的能力上普遍更优。<br><br>IDWE_CHM（NRT_L）的更新时序与融合流程中集成的五个数据集保持同步，其中ERA5-Land数据集（获取地址：https://cds.climate.copernicus.eu/datasets/reanalysis-era5-land?tab=overview）相较于其他融合数据集存在约5天的延迟，因此已完成明确的同步适配。