Study Of Sediment Grain Size Remote Sensing Retrieval On Tidal Flat Surface

The mathematical relationship between spectral reflectance and tidal flat surface sediment grain size was established by using more than 200 sets of collected sediment samples and spectral reflectance which were obtained from the west and south coast of Laizhou Bay.By testing and verifying the accuracy of the mathematical method,we can determine whether the model can be used to retrieve the remote sensing data(six remote sensing images from 1995 to 2017)or not.By analyzing the retrieval results of remote sensing images,the reliability of the retrieval results is verified,and the feasibility of retrieving tidal flat surface sediment grain size using BP neural network model is analyzed.The paper is divided into six modules: The first module is the collection and laboratory analysis of the field spectral data and sediment samples.The second module is the matching of the spectral data collected in the field with the remote sensing satellite band,and the correlation coefficient between band combinations of spectral reflectance with sediment grain size data was anlalysis to construct the remote sensing retrieval factors.The third part is the establishment of BP neural network model based on the relationship between spectrum and grain size and the fourth part is the verification of the model.The fifth part is to test and analyze the precision of the model and the sixth part is to use the model to retrieve remote sensing image and analysis the retrieval results with the field data and historical geomorphological data to further demonstrate the feasibility of retrieve of tidal flat surface sediment grain size using BP neural network.The main conclusions are as follows:(1)The model validation results show that the retrieval data calculated by the model established was fitting well with measured data and the correlation coefficients were 0.83,0.89,0.87 and 0.79.The correlation coefficients of the training results are above 0.8.The mean grain size retrieval result is the best and the sand content and clay content retrieval result is worse than the other two.(2)The APE test result show that the accuracy of the model results can be directly used for retrieve the remote sensing images.The mean grain size got the best retrieval results and the average value of APE can reach 9%,but the sand content and the clay content got the worst APE.By analyse APE values deeply,the abnormal high PE value,leads to the overall calculation result of APE,which influence the instability of APE.The reason why high PE values are generate is that the BP neural network is not sensitive to predict extreme low or high value and prediction the minimum value easily lead to high PE value.By eliminating the extremely high PE value,the content of clay APE decreased from a maximum of 106% to 28%,sand content APE declined from a peak of 189% to 38%.(3)The stability of RMSD value show that difference between retrieved data and measured data is stable.PRMSD values verified the whole retrieved results is good no matter how much difference in APE.For example,the difference of APE results of sand content and clay content is very large,from less than 30% to more than 100%,but the results of the PRMSD differs in range of a few percent to about ten percent,so the PRMSD value represents the overall percentage difference between retrieved data and measured data.(4)The remote sensing data of the year of 1995,1999,2005,2009,2014 and 2017 were retrieved by using BP neural net work.It turns out that the tidal flat surface sediment grain size the size was different from different part of Laizhou Bay,it shows a trend that the grain size of south of Laizhou Bay was bigger than that of west,which is consistent with the actual law of Laizhou Bay intertidal sediment grain size distribution.