基于Sentinel-2卫星的星云湖叶绿素a遥感估算研究

Remote Estimation of Chlorophyll-a in Xingyun Lake by Sentinel-2 Satellite

本研究采用了哨兵2A卫星建立针对星云湖的叶绿素a遥感估算模型。通过星地同步观测分析了各波段反射率与叶绿素a浓度的相关性,依据DVI,RVI,NDVI,TBM,MCI这5种算法采用不同敏感波段的组合建立了21个模型,再将建模数据集分为全部数据集,高浓度和低浓度数据集,将3个数据集代入21个模型建立63个回归方程,并分析这些回归方程的建模和验证效果,提出针对星云湖不同叶绿素a浓度范围的湖区采用不同的模型进行叶绿素a浓度的遥感估算。本研究结果表明:1) 哨兵2卫星4个红边波段和近红外波段反射率均与叶绿素a浓度呈强烈正相关,叶绿素a浓度高于0.1 mg/L时,最佳模型是RVI1H,其rRMSE和NMAE分别为4.01%和3.95%。叶绿素a浓度低于0.1 mg/L时,最佳模型是NDVI1L,其rRMSE和NMAE分别为25.95%,19.32%,采用TBM1L模型估算比较适合计算全湖的平均值,其MNB为?0.57%。2) 建模的回归方程决定系数高,只能说明建模数据集的线性较好,但是模型是否适用,主要依据还是验证数据的误差。综上,本研究建立的星云湖叶绿素a遥感估算模型,对于星云湖的蓝藻水华遥感监测具有一定的参考价值。

Sentinel-2A satellite was used to establish chlorophyll-a remote sensing estimation models for the Xingyun Lake in this study. The correlation between the spectral reflectivity of each band and the concentration of chlorophyll-a was analyzed by the concurrent observation. According to DVI, RVI, NDVI, TBM and MCI algorithms, a combination of different sensitive bands and algorithms is used to set up 21 models. Then the modeling dataset is divided into all datasets, high and low concentration datasets. Three data sets are substituted into 21 models to establish 63 regression equations. The modeling and validation effects of these regression equations are analyzed. The remote sensing estimation of chlorophyll-a concentration in Xingyun Lake is carried out by using different models in different chlorophyll-a concentration ranges of Xingyun Lake. The result shows that: 1) The spectral reflectivity of the 4 Red-Edge bands and Near-Infrared band of the Sentinel-2 satellite is strongly positive related to the concentration of chlorophyll-a. When the concentration of chlo-rophyll-a was higher than that of 0.1 mg/L, the best model is that the rRMSE and NMAE of RVI1H, were 4.01% and 3.95%, respectively. When the concentration of chlorophyll-a is lower than 0.1 mg/L, the best model is that the rRMSE and NMAE of NDVI1L are 25.95% and 19.32%, respectively. TBM1L model is more suitable to calculate the average value of the whole lake, and its MNB is ?0.57%. 2) The regression equation of modeling has high determination coefficient, which can only show that the linearity of the modeling data set is better, but whether the model is applicable or not is mainly based on the error of the verification data.