一种高斯过程回归的水体吸收系数反演模型

Absorption Coefficients Inversion Model Based onGaussian Process Regression

水体吸收系数是评价水体环境质量和衡量海洋对全球气候影响的重要因子。水色遥感作为目前对大范围海洋进行长时间连续监测的唯一手段,可以借助于合适的反演模型从影像上获取水体吸收系数。然而现有模型多为经验模型,反演精度和水体适应性都较差。基于自建的全球水体光学原位测量数据集SeaBASS2020,通过选用412、443、490、510、560、620、665 nm波段的遥感反射率以及620、665 nm与其余5个波段的遥感反射率比值作为模型输入特征,并以Matérn函数作为模型的核函数,提出一种基于高斯过程回归的水体吸收系数反演模型GPR-a。实验结果表明,在反演精度上GPR-a较传统的波段比值经验模型有大幅提升,其中决定系数R 2提升了24.79%,均方根误差σRMSE和平均相对误差εMRE分别降低了50%和35.17%。此外,实验还验证了GPR-a具有较强的鲁棒性和极佳的反演值不确定度估计能力。

Absorption coefficient is an important factor for evaluating the water quality and measuring the impact of the ocean on the global climate.Ocean color remote sensing is currently the only method to obtain large-scale,quasi-real-time and high continuous absorption coefficients with appropriate inversion model.However,most of the existing models are empirical models,with poor inversion accuracy and water body adaptability.Based on the self-built global in-situ water optics dataset SeaBASS2020,a new inversion model of absorption coefficients GPR-a based on Gaussian process regression is proposed with the remote sensing reflectance at 412,443,490,510,560,620,665 nm and the ratio of remote sensing reflectance at 620 and 665 nm to the other five wavelengths as inputs and Matérn as kernel function.Experimental results show that GPR-a has higher accuracy than the traditional band ratio empirical model,with the determination coefficients R 2 increased by 24.79%,and the root mean square errorσRMSE and mean relative errorεMRE decreased by 50%and 35.17%,respectively.In addition,it is also verified that GPR-a has strong robustness and excellent ability to estimate the uncertainty of inversion values.