Specific inherent optical properties of highly turbid productive water for retrieval of water quality after optical classification

Assessments of specific inherent optical properties (SIOPs) and their variability in highly turbid and productive inland waters are essential for the accurate estimation of water quality. A new optical classification method including two classification criteria [i.e., normalized remote sensing reflectance slope (NS), and normalized remote sensing reflectance depth (ND)] was developed to divide remote sensing reflectance into four classes, i.e., class 1 (NS < −0.0017 and ND < 0.21) is low turbid and productive water; class 2 (NS < −0.0017 and ND > 0.21) is low turbid and high productive water; class 3 (NS > −0.0017 and ND < 0.09) is high turbid and low productive water; and class 4 (NS > −0.0017 and ND > 0.009) is high turbid and high productive water. The relationships between phytoplankton absorption at 440 nm [aph(440)] and chlorophyll-a concentration [Cchla] as well as between particle backscattering coefficient at 440 nm [bbp(440)] and total suspended matter concentration (CTSM) after classification were obtained from a large number of in situ data in Lake Taihu. The measured specific phytoplankton absorption $$[ {a_{\text{ph}}^{*} \left( \lambda \right)} ]$$[aph∗λ] and particle backscattering coefficient $$[ {b_{\text{bp}}^{*} \left( \lambda \right)} ]$$[bbp∗λ] show significant variations even within the same class. The mean values of $$a_{\text{ph}}^{*} \left( \lambda \right)$$aph∗λ at 440 nm $$[ {a_{\text{ph}}^{*} \left( {440} \right)} ]$$[aph∗440] for each class are 0.048 ± 0.013, 0.060 ± 0.012, 0.083 ± 0.021, and 0.056 ± 0.017 m2/mg, respectively. The mean values of $$b_{\text{bp}}^{*} \left( \lambda \right)$$bbp∗λ at 440 nm $$[ {b_{\text{bp}}^{*} \left( {440} \right)} ]$$[bbp∗440] for each class are 0.035 ± 0.01, 0.024 ± 0.004, 0.041 ± 0.009, and 0.038 ± 0.009 m2/g, respectively. The power functions of SIOPs and water constituents’ concentration indicate that $$a_{\text{ph}}^{*} \left( {440} \right)$$aph∗440 and $$b_{\text{bp}}^{*} \left( {440} \right)$$bbp∗440 vary with Cchla and CTSM. The validation results show that our proposed values for $$a_{\text{ph}}^{*} \left( {440} \right)$$aph∗440 and $$b_{\text{bp}}^{*} \left( {440} \right)$$bbp∗440 cover a very wide range of water optical properties, which are characterized from clear water to highly turbid productive water. The validation results also suggest that the retrieval accuracy of Cchla and CTSM bio-optical model was improved after classification. The root mean square error (RMSE) of Cchla was improved from 14.18 to 7.43 μg/L (mean value of all classes) and RMSE of CTSM was improved from 32.98 to 26.10 mg/L (mean value of all classes). Thus, the temporal and spatial variation of $$a_{\text{ph}}^{*} \left( {440} \right)$$aph∗440 and $$b_{\text{bp}}^{*} \left( {440} \right)$$bbp∗440 should be considered in the bio-optical retrieval model of water quality.Graphical AbstractIn complex optical properties of inland water, retrieving the water constituents with high accuracy needs to classify the water optical properties from the remote sensing spectrum by optical classification method. The figure shows the water color examples of each class.