2000–2021年黄河源区无云MODIS积雪覆盖面积比例数据集

A dataset of MODIS gap-filled fractional snow cover in the source area of Yellow River during 2000–2021

黄河源区是黄河流域的主要产水区和水源涵养区,积雪融水是源区的重要水源之一,高精度积雪面积数据集是源区生态水文模拟、气候变化等研究的基础。MODIS积雪产品是最广泛使用的积雪面积数据集之一。然而,MODIS积雪产品中大量的云覆盖,导致了近乎一半的信息缺失。黄河源区季节性积雪多呈现出雪层偏浅、斑块状分布且消融快等特点,使得传统统计方法很难准确捕获源区的积雪时空特征,而先进的深度学习技术能更好地深入挖掘隐藏在数据背后的时空特征。本研究利用2000–2021年逐日500 m空间分辨率的MODIS归一化积雪指数(NDSI)产品,使用基于部分卷积神经网络(PCNN)的MODIS NDSI云像元重建模型,在生成时空连续MODIS NDSI数据集的基础上,进一步采用NASA原始积雪覆盖比例(FSC)产品的标准算法,制备黄河源区2000–2021年逐日、0.005°(约500 m)的无云MODIS FSC数据集。基于源区6个地面气象台站雪深观测资料和“云假设”两方面的验证表明,数据集的总体精度可以达到94%,高估和低估均为1%,平均绝对偏差10.43%,平均相关系数为0.93,表明数据具有较高精度,与晴空状态下的MODIS积雪产品具有相当的精度。本数据集可以为黄河源区积雪动态变化监测、水资源综合管理评估、气候变化等研究工作提供数据支撑。

The source area of the Yellow River is the main water production and conservation area of the Yellow River Basin.Snowmelt water is one of the important water sources in the source area of the Yellow River.High-precision datasets of snow cover area are the foundation for the study on eco-hydrological simulation and climate change in the area.The MODIS snow cover product is one of the most widely-used snow area datasets.However,the large amount of cloud cover results in serious gap-pixels in Moderate Resolution Imaging Spectroradiometer(MODIS)snow cover products.Due to the fact that the seasonal snow cover in the source area of the Yellow River mostly presents the characteristics of shallow snow layer,patchy distribution and fast melting,it is difficult for traditional statistical methods to accurately capture the temporal and spatial characteristics of snow cover in the area.However,the advanced deep learning technology can be used to better dig into the temporal and spatial characteristics hidden behind the data.Referring to the daily MODIS normalized snow cover index(NDSI)products with the spatial resolution of 500m during 2000-2021,we used the reconstructed MODIS NDSI cloud pixel model based on partial convolution neural network(PCNN)to generate a dataset of spatiotemporal continuous MODIS NDSI.On this basis,we further adopted the standard algorithm of NASA original snow cover ratio(FSC)product to prepare the daily cloudless MODIS FSC dataset with the resolution of 0.005°(about 500 m)during 2000-2021 in the source area of the Yellow River.The validation based on the snow depth observations from six meteorological stations and“cloud assumption”shows that the overall accuracy of the dataset is 94%.Both the overestimation and underestimation are 1%;the average absolute deviation is 10.43%,and the average determination coefficient is 0.86.All these values indicate that this is a fairly high-precision dataset.The dataset can provide data support for the research on snow dynamic change monitoring,integrated water resource management and evaluation,climate change,etc.in the source area of the Yellow River.