面向GF-1 WFV数据的闽江下游叶绿素a反演模型研究

Comparison and analysis of chlorophyll-a retrieval model in the lower reaches of Minjiang River based on GF-1 WFV image

叶绿素a浓度是可直接遥感反演的重要水质参数之一,常用来评价水体的富营养化程度.为建立适合于闽江下游叶绿素a浓度的反演模型,利用地面采样数据,结合GF-1 WFV光谱响应函数,选用多元回归、BP神经网络和随机森林方法,构建了叶绿素a浓度反演模型;并根据验证数据与实测值之间的决定系数(R^2)、均方根误差(RMSE)和平均相对误差对模型反演结果进行了比较.结果发现,随机森林模型的R^2为0.895,RMSE为1.994 mg·m^-3,平均相对误差为11.502%,是3种模型中最优的.为了评估模型的性能,进一步比较了WFV影像像元反射率反演的叶绿素a浓度值与相应的实测值.结果表明,随机森林模型同样具有较高的精度,其R^2为0.709,RMSE为3.540 mg·m^-3,平均相对误差为25.616%.本研究可为闽江下游水环境的监测提供一定的理论依据和技术参考.

Chlorophyll-a(Chla) is one of the most important water quality parameters, which is directly measurable from remote sensing, it is often used to assess water eutrophication. To establish a retrieval model suitable for estimating the Chla concentration in the lower reaches of Minjiang River, in-situ data and the spectral response function of GF-1 Wild Field of View(WFV) were used by applying the Multivariate Regression, Backward Propagation Neural Network and Random Forest(RF) methods. The performance of the retrieval models was compared by measuring the coefficient of determination(R^2), root-mean-squared error(RMSE) and mean relative error between the verification data and the observed values. The RF model had an R^2 of 0.895, an RMSE of 1.994 mg·m^-3, and an average relative error of 11.502%, showing the best performance among the three models. To evaluate model performance, we further compared the Chla retrieved from the pixel reflectance of WFV image with corresponding measurements. It was found that RF model also had a high accuracy with an R^2 of 0.709, an RMSE of 3.540 mg·m^-3, and an average relative error of 25.616%. Based on these results, it can be concluded that the present study can provide a theoretical basis and technical reference for monitoring of the water environment in the lower reaches of Minjiang River.