黄河流域多源融合实际蒸散发数据集（2000-2020）

本研究提出了基于土地覆被分类的多源蒸散发时空融合框架（CNN-LSTM-Attention），融合GLEAM、ETMonitor、SiTHv2、通量站等多源蒸散发数据及ERA5-Land数据，创新性构建14类异质下垫面子模型，通过卷积神经网络提取空间特征、长短期记忆网络建模时序依赖，并利用注意力机制增强关键特征，最终生成黄河流域2000-2020年日尺度实际蒸散发数据集，空间分辨率0.1°。结果表明，融合模型在全球268个通量站点的RMSE为0.78 mm/d，R²达0.77。在黄河流域的验证结果显示，流域年均蒸散发估算R²为0.92，在CN-Ha2等典型站点的RMSE降低至0.39-0.72 mm/d。本框架显著提升了异质下垫面蒸散发估算精度，可为黄河流域水循环研究提供有力的数据支持。

This study proposes a multi-source actual evapotranspiration (ET) spatiotemporal fusion framework (CNN-LSTM-Attention) based on land cover classification. The framework fuses multi-source ET datasets including GLEAM, ETMonitor, SiTHv2, and flux tower observations, alongside ERA5-Land data. It innovatively develops 14 heterogeneous underlying surface sub-models, where convolutional neural networks (CNN) are used to extract spatial features, long short-term memory (LSTM) networks to model temporal dependencies, and the attention mechanism to enhance key feature contributions. Finally, a daily-scale actual evapotranspiration dataset for the Yellow River Basin spanning 2000 to 2020 is generated, with a spatial resolution of 0.1°. Validation results show that the fused model achieves a root mean square error (RMSE) of 0.78 mm/d and a coefficient of determination (R²) of 0.77 across 268 global flux towers. Within the Yellow River Basin, the R² of annual average ET estimation reaches 0.92, and the RMSE at typical sites such as CN-Ha2 is reduced to 0.39–0.72 mm/d. This framework significantly improves the estimation accuracy of ET over heterogeneous underlying surfaces, providing robust data support for hydrological cycle research in the Yellow River Basin.

搜集汇总

背景与挑战

背景概述

该数据集为黄河流域2000-2020年的多源融合实际蒸散发数据，时间分辨率为日，空间分辨率为0.1°，数据大小为1.17 GB，采用开放获取方式共享。数据集通过融合GLEAM、ETMonitor等多源数据，利用CNN-LSTM-Attention框架构建14类异质下垫面子模型，显著提升了蒸散发估算精度，适用于黄河流域水循环研究。