高时间分辨率的积雪判识对于新疆牧区农牧业发展和雪灾预警具有重要作用,针对已有积雪产品易受复杂地形地貌,下垫面类型以及云遮蔽的影响,导致积雪判识精度降低的问题,提出一种利用深度学习方法对风云4号A星多通道辐射扫描计(AGRI)数据与地理信息数据进行多特征时序融合的积雪判识方法:以多时相FY-4A/AGRI多光谱遥感数据,以及高程、坡向、坡度和地表覆盖类型等地形地貌信息作为模型输入,以Landsat 8 OLI提取的高空间分辨率积雪覆盖图作为"真值"标签,构建并训练基于卷积神经网络的积雪判识模型,从而有效区分新疆复杂地形与下垫面地区的云、雪以及无雪地表,最终得到逐小时积雪覆盖范围产品。经数据集和2019年地面气象站实测雪盖验证,该方法精度高于国际主流MODIS逐日积雪产品MOD10A1和MYD10A1,显著降低云雪误判率。

Snow, Cover Area monitoring is an important factor in studies of global climate change, regional water balance and soil moisture. Recently, the usage of remote sensing techniques has flourished. In fact, remote sensing data provides timely adequate snow cover information for large areas. While the National Center for Remote Sensing in Lebanon (CNRS) has recently established an operational monitoring room for natural resources and natural disasters, this paper presents the implementation of a fully automated snow cover monitoring system based on Moderate Resolution Imaging Spectroradiometer (MODIS) satellite images. The system uses snow products from EOS Terra, and Aqua satellites to monitor the Snow Cover of Lebanon during the snow season (i.e. November April). The importance of this project lies in its daily and fully automated process of acquiring, processing, storing and displaying statistics of the snow covered areas in Lebanon. Applying a custom algorithm based on combining Terra and Aqua snow products will reduce cloud contamination.