GNSS-R Soil Moisture Retrieval with a Deep Learning Approach
GNSS-R Soil Moisture Retrieval with a Deep Learning Approach
复制标题
使用深度学习方法反演 GNSS-R 土壤湿度
DOI：
10.1109/igarss47720.2021.9555022
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
C. Chew
中科院分区：
文献类型：
--
作者：
T. Roberts;Ian Colwell;R. Shah;S. Lowe;C. Chew
可利用SMAP的数据对全球导航卫星系统反射测量值进行校准，以产生对某些水文/气象研究有用的具有更高时空分辨率的土壤湿度估计值。目前的方法使用DDM（延迟多普勒地图）和土壤湿度之间的关系的简单模型，并可能在地球的某些地区失败。完整的2D DDM中包含的复杂信息可以在这些领域提供帮助，并且可以通过应用基于深度学习的技术来提取。我们的工作探索了卷积神经网络的数据驱动方法，以确定反射测量和表面参数之间的复杂关系。我们开发了一个使用CYGNSS DDA和与SMAP土壤湿度值一致的辅助数据集训练的神经网络;分析其结果并与现有的全球土壤湿度产品进行比较。
GNSS reflection measurements can be calibrated with data from SMAP to yield estimates of soil moisture with enhanced spatiotemporal resolution useful to certain hydro-logical/meteorological studies. Current approaches use simple models of the relation between the DDM (delay-Doppler map) and soil moisture and can fail in certain regions of the planet. Complex information contained in the complete 2D DDM could help in these areas, and can be extracted through the application of deep learning based techniques. Our work explores the data-driven approach of convolutional neural networks to determine complex relationships between the reflection measurement and surface parameters. We developed a neural network trained using CYGNSS DDMs and ancillary datasets aligned with SMAP soil moisture values; the results of which are analyzed and compared to existing global soil moisture products.