基于树模型机器学习方法的GNSS-R海面风速反演

GNSS-R是基于GNSS卫星反射信号的一种新技术.GNSS-R技术可以运用到海面风场反演中，传统的GNSS-R技术反演海面风场主要有波形匹配和经验函数两种方法，风速反演精度约为2m·s-1.波形匹配方法耗时多，计算量大；经验函数方法通常只使用少量物理观测量，会造成信息浪费，损失一定的反演精度.为了提高海面风速的反演精度，引入机器学习领域常用的树模型算法决策树、随机森林、GBDT等对海面风速进行预测.利用GNSS-R与ECMWF数据构成训练集和验证集，训练集用于模型学习，验证集用于检验模型的反演效果.实验结果显示，决策树和随机森林预测误差约为0.6m·s-1，GBDT等算法的预测误差约为2m·s-1，满足风速反演要求.与GNSS-R传统反演方法相比，机器学习树模型算法效果更好，在验证集上表现稳定且误差较小.因此，可以将机器学习树模型算法运用到海面风速反演中. 

GNSS-R Sea Surface Wind Speed Inversion Based on Tree Model Machine Learning Method

GNSS-R is a new technique based on GNSS satellite reflection signals, and it can be applied to the inversion of sea surface wind field. The traditional GNSS-R technology inversion of sea surface wind field mainly has waveform matching and experience function. The waveform matching method is time-consuming and computationally intensive; the empirical function method often uses only a small amount of physical observations, which causes waste of additional information and loss of certain inversion precision. The accuracy of the traditional method of wind speed inversion is about 2m·s-1. In order to improve the inversion accuracy of sea surface wind speed, the tree model algorithm decision tree, random forest and GBDT commonly used in the field of machine learning are introduced to predict the sea surface wind speed. The training set and the verification set are constructed by using GNSS-R and ECMWF data. The training set is used for model learning, and the verification set is used to test the inversion effect of the model. The prediction error of decision tree and random forest is about 0.6m·s-1, and the prediction error of GBDT and other algorithms is about 2m·s-1, which meets the requirements of wind speed inversion. Compared with the traditional GNSS-R inversion method, the machine learning tree model algorithm performs better and has stable performance and less error on the verification set. Therefore, the machine learning tree model algorithm can be applied to the sea surface wind speed inversion.