PhD Defence Xiaolong Yu

optical properties of turbid coastal water and their remote sensing retrievals based on radiative transfer approaches

Xiaolong Yu is a PhD student in the Department of Water resources. His supervisor is W. Verhoef from the Faculty of Geo-Information Science and Earth Observation.

Radiative transfer theory describes how light, regarding its spectral radiance, travels and varies along any specific path at a specified point in the water, and ties together the water apparent optical properties (AOPs) and inherent optical properties (IOPs) through the radiative transfer equations (RTEs). In this study, the author used the analytical 2SeaColor model, which is based on the approximate solution of RTEs using two-stream approach, to retrieve IOPs, diffuse attenuation coefficient (Kd, in m-1), and the concentration of suspended particulate matter (CSPM, in mg/L) in the turbid Yangtze estuary. The application of 2SeaColor to the GOCI geostationary sensors has enabled a better understanding of the diurnal cycle of suspended sediment in the study area. The known linear relationship between CSPM and particulate backscattering coefficient (bbp, in m-1), was investigated, using 2SeaColor retrievals and field measurements. Although the validity of the linear relationship was confirmed for relatively clear waters, the author showed that in turbid waters, nonlinearity governs the relationship between CSPM and bbp. The two respective relationships between CSPM and bbp for relatively clear and turbid waters were further integrated using a sigmoid function to retrieve wide-range CSPM without the need for a switching scheme. Apart from SPM, colored dissolved organic matter (CDOM) also significantly reduces light transparency and contributes to the complexity of the optical properties in the Yangtze estuary. Thereby, the author has also investigated the seasonal variability of CDOM absorption properties. Finally, the author proposed a locally calibrated model to estimate the concentration of dissolved organic carbon (DOC). The summary of the major results obtained from this dissertation is briefly described in the following.

The thesis is structured in six chapters, of which the first chapter describes the background of ocean color remote sensing in optically complex waters and the second chapter explains the physical principals of the 2SeaColor model. In Chapter 3, the author first proposed improved IOPs parameterizations to the analytical 2SeaColor model to account for the complex optical properties in the Yangtze estuary. The improved 2SeaColor model was later employed to derive the diffuse attenuation coefficient (Kd, in m-1) from remote sensing reflectance (Rrs, in sr-1). Retrievals from the improved model were then validated with three in-situ datasets and compared with an empirical (the Zhang model) and a semi-analytical model (the Lee model). Statistics from the validation results show that the 2SeaColor model provides the best estimates of Kd for the full range of observations, with the largest determination coefficient (R2 = 0.935) and the smallest root mean squared error (RMSE = 0.078 m-1). For clear waters, where Kd (490) < 0.2 m-1, the Zhang model provides the most accurate Kd estimations, but results from the Lee model and the 2SeaColor are comparable. For turbid waters, where Kd (490) > 0.2 m-1, the 2SeaColor model is found to be more accurate, with an RMSE of 0.186m-1, compared to RMSEs of 0.279 m-1 and 0.388 m-1 for the Zhang model and the Lee model, respectively. The improved 2SeaColor model is finally applied to the GOCI (Geostationary Ocean Color Imager) level 2 product to produce Kd maps over the Yangtze estuary, resulting in a reasonable distribution and expected range of Kd for the Yangtze estuary, while the Zhang model and the Lee model show the potential of underestimation and overestimation, respectively. The consistently stable and accurate Kd estimates in both clear and turbid waters show great potentials of the 2SeaColor model for estimating Kd over optically complex waters.

In Chapter 4, the author proposed a novel model to quantitatively estimate the concentration of SPM (CSPM, in mg/L) from remotely sensed particulate backscattering (bbp, in m-1). bbp of both clear and turbid waters were first retrieved by the improved 2SeaColor model, and then the relationships between bbp and CSPM were calibrated for a wide-range of CSPM from 0.4 to 2068.8 mg/L. Calibration results show that bbp and CSPM are linearly correlated for relatively clear waters, but their relationship becomes significantly curved for turbid water. To avoid the discontinuity issue resulted from the two respective bbp-CSPM relationships for clear and turbid waters, an index of bbp (Sindex) was introduced based on a sigmoid function to retrieve CSPM. The Sindex-CSPM model was calibrated using a calibration (Cal) dataset and then validated with the validation (Val) dataset. Validation results show that the Sindex-CSPM model has overall better retrieval accuracy (rMAD = 33.4%) compared with the re-calibrated He model (rMAD = 185.2%) and the GOCI SPM model (rMAD = 50.2%). For extremely turbid waters, these three models show comparable performance with rMAD of 15.29%, 12.72%, and 17.22% for the Sindex-CSPM model, the GOCI SPM model, and the He model, respectively. The Sindex-CSPM model, without tuning the empirical constants, was also applied to an independent SeaSWIR dataset collected from the estuaries of Europe and Argentina with measured CSPM up to 1400 mg/L. Validation results show a very promising retrieval accuracy with rMAD of 30.3 % without tuning the empirical coefficients, indicating the potential good transferability of the Sindex-CSPM model among global turbid coastal waters. Further, the three models were implemented on the GOCI image acquired at 02:16 UTC, March 8th of 2013 to generate the SPM mapping products for the Yangtze estuary, where discontinues are observed for the switching-based GOCI SPM model at the front of the turbid plume, while the Sindex-CSPM model produced a smooth and realistic SPM mapping product. At last, the Sindex-CSPM model was applied to 8 GOCI images acquired on March 8th of 2013 to investigate the diurnal variation of CSPM, and the comparison with tide information shows that the predicted CSPM can be well explained by hydrologic processes, which further enhanced our confidence in the reliability and repeatability of the Sindex-CSPM model.

In Chapter 5, the author investigated the seasonal variability of CDOM absorption properties and its relationship between salinity and dissolved organic carbon (DOC). The CDOM absorption coefficient at 355 nm (ag (355)) was found to be inversely correlated with salinity, with Pearson’s coefficients r of -0.901 and -0.826 for summer and winter observations, respectively. Analysis results of the relationships between salinity and CDOM optical properties (i.e., absorption coefficient and spectral slope) suggested that terrigenous inputs dominated CDOM sources in the Yangtze estuary, but the proportion of terrigenous CDOM declined with increasing salinity. Resuspension of bottom sediments could be an important source of CDOM in the winter. We further developed an empirical model to estimate DOC concentration from ag (355) and spectral slope S275-295 using the calibration dataset by non-linear regression. The developed model was validated with the validation dataset, resulting in an acceptable error with a determination coefficient (R2) of 0.746, a root mean squared error (RMSE) of 20.99 μmol/L and a mean relative absolute error (rMAD) of 14.46%. This empirical model could be potentially adopted in retrieving DOC concentration using ocean color remote sensing over the Yangtze estuarine and coastal waters.