The Atmospheric Correction And Three-component Retrieval For Case Ⅱ Waters With Neural Network

Supported by the National 863 High Technology Projects *"Remote Sensing of Case II Waters", the Neural Network (NN) models for atmospheric correction and three-component concentrations retrieval are established for Yellow Sea and East China Sea which belong to Case II waters. The NN models are also applied to real satellite data to analyze the applicability as well as the shortcomings. Finally the suggestions for further improvements are put forward. The thesis consists of the following four aspects:(1) First of all, the bio-optical properties of Yellow Sea and East China Sea are briefly introduced. The universal model for chlorophyll absorption have not be derived presently, and the absorption of total suspended matter(TSM) has the similar result. The absorption of yellow substance and back scattering of TSM both have preliminary results.(2) The NN models for three-component retrieval based on in situ remote sensing reflectances (Rrs) have rather good results. The relative errors are respectively 44.4% for chlorophyll, 40.5% for TSM and 48.8% for gelbstoff for three-component retrieval simultaneously. However, the errors are respectively 32.5% for chlorophyll, 29.4% for TSM and 32.5% for gelbstoff for the models to derive each component individually. It's worthwhile pointing out that only 555nm and 670nm bands are sufficient to retrieve the TSM concentrations in coastal areas. Besides, the relative error for chlorophyll of low turbidity waters (color number from 6 to 9) is only 16.2%, which is much more accurate than that of all stations (color number from 6 to 21). This shows that ocean color algorithms of Case-II waters are highly related with the regional complexity of waters.(3) The Rayleigh-Corrected top of atmosphere (TOA) signals are modeled from in situ Rrs and a simplified, moderate accuracy aerosol model to set up NN models for aerosol parameters retrieval, which is then used to derive water-leaving signals. Theresults for real TOA data show that the shape of Rrs spectrum are mostly close to the in situ ones, whereas the absolute values have about 20% errors. Restricted by the matching up of in situ and satellite data, the absolute values of results couldn't be assured. Therefore, the inputs of band ratios to retrieve ocean color are suggested.(4) The NN models to retrieve three-component concentrations directly from modeled TOA signals are also set up. Results show that the TSM retrieval could have rather rational results, but is somewhat higher in low turbidity waters. Gelbstoff retrieval is applicable for Jiangshu shallow offshore and Changjiang estuary, whereas the algorithm breaks up in the vicinity of Qingdao Harbor. It shows that the applicability of the model to waters with various turbidity may need further research.Finally, the suggestions for further work are put forward.