紫外-可见光去浊度反演水体叶绿素a含量

城市景观水体的叶绿素a含量可直接反映其水质。紫外-可见光谱方法可快速低廉反演叶绿素a的含量,但城市水体水深较浅、浊度较高,容易对该波段光谱产生干扰。采用实验室培养的螺旋藻水样和浊度水样的混合水样来模拟城市景观水体环境,并使用UV—2600分光光度计获取混合水样的吸光谱数据;对吸光谱数据分别建立一元线性回归模型、偏最小二乘算法(PLS)模型和BP神经网络模型,以寻找降低水体浊度干扰的办法,为水体水质评价提供可靠参考数据。结果显示,BP神经网络预测模型可以同时预测混合水样中叶绿素a的浓度和浊度的浓度值,模型预测值与样本测量值之间的R~2为0. 997 2,并且模型的预测误差在5%以内。去浊度反演水体叶绿素a含量的能力最高;偏最小二乘算法模型测量值与预测值的R2为0. 999 4,模型的预测误差小于4%,但该模型在预测叶绿素a浓度时不能同时预测浊度值,去浊度反演叶绿素a含量的能力次之;一元线性回归模型的去浊度反演叶绿素a含量的能力最差。

Chlorophyll-a Prediction in Urban Water by UV-VIS Spectra with Consideration of Turbidity

Chlorophyll-a (Chl-a) concentration can be used as an indicator of the quality of urban landscape water. UV-VIS spectra can be used to predict Chl-a concentration quickly and cheaply, but urban water bodies are mostly shallow and have higher turbidity, which has a negative effect on their spectra in this range. In this paper, the mixed water samples with artificial cultured Spirulina and different levels of turbidity were used to simulate the urban landscape water, and the absorption spectra of those water samples were obtained by UV-2600 spectrophotometer. In order to search methods to effectively reduce the interference of turbidity on Chl-a estimation, single linear regression model, partial least squares (PLS) model and BP neural network model were built from the absorbance spectra of those mixed water samples. The results show that the BP neural network prediction model can simultaneously predict the concentration of chlorophyll a and the concentration of turbidity in the mixed water sample. The R2 between the model''s predicted value and the measured value of the sample is 0.9972, and the prediction accuracy of the model is within 5%. The turbidity had the highest ability to retrieve the chlorophyll a content of the water; the R2 of the measured and predicted values of the partial least square algorithm model was 0.9994, and the prediction error percentage of the model was less than 4%, but the model could not predict the concentration of chlorophyll a. At the same time, the turbidity value was predicted, and the ability of de-turbidity to invert chlorophyll a content was secondary; the de-turbidity of the one-dimensional linear regression model was the worst in the ability to invert chlorophyll a content.