Study On Scale Effect And Sub-pixel Mapping Of Cyanobacteria Bloom From Remote Sensing Observations

With the very high temporal resolution, GOCI is of great significance for continuous dynamic monitoring algae bloom. Taking Taihu lake as the research area, this study uses the GOCI image data synchronization with the MODIS LIB product and HJ-1A/1B data to analyse scale differences of different sensors for cyanobacterial bloom area information, as well as the cause of the spatial scale effect. Secondly, based on scale difference analysis, we develop scale down transformation model which is suitable for cyanobacteria blooms monitoring. Finally, with cyanobacterial blooms proportion which derived from former work, we developed sub-pixels mapping technology to derive distribution of cyanobacterial blooms at sub-pixel scale. Based on the above three parts work, we established technology framework of extraction cyanobacterial blooms extraction at sub-pixel scale. Several conclusions are drawn as follows:(1) Scale difference of cyanobacteria bloom from MODIS,GOCI and HJ-1A/1B observationsCompared with the other two sensors, cyanobacterial blooms observations from HJ-1A/1B characterized more spatial details.The differences of3sensors for cyanobacterial bloom area information mostly appear in the border region. Taking HJ-1A/1B as reference data, the area errors on two dates of MODIS extraction are0.15,0.07, and the Kappa coefficients are0.86,0.72; the area errors on2dates of GOCI extraction are0.3,0.11, and the Kappa coefficients are0.75,0.65.The spatial scale effect becomes more significant from250m to500m resolution.The nonlinear characteristics of NDVI, a common retrieving alga bloom method from satellite images, was studied, error caused by the NDVI calculation method is negligible.Then we use semi-variogram to describe the spatial heterogeneity in statistical samples. We found that y (8) has good regression relation with the cyanobacterial blooms proportion in samples and the fractal dimension. The spatial heterogeneity is the biggest impact factor of extraction difference, on the other word the proportion and fragmentation of cyanobacterial blooms affects extraction difference together. The proportion of the cyanobacterial blooms is close to50%, the fragmentation is larger,the bigger the scale difference.(2) Scaling down of the extraction cyanobacterial blooms proportion from differentsensorsDue to the result of scale down conversion method based on object does not meet the requirements of the sub-pixels mapping methods’input, this paper mainly discusses two kinds of conversion method based on pixels, respectively is the method of regression model and moving window linear spectral mixture model(MW-LSMM).Through the application in the GOCI image contrast of the two methods, we found that MW-LSMM has higher overall accuracy than the other with area errors decreasing by29.9%in MODIS validation area and by24.7%in GOCI validation area. However the regression model only decreased by18.36%and13.6%.Because the MW-LSMM does not need the support of high spatial resolution image, it has the practical application of more convenient.(3) Sub-pixel mapping of cyanobacterial blooms proportionIn this study CA was improved in three ways to overcome problems mentioned above. A cyanobacterial blooms image was employed to test running efficiency of improved CA. Although the adjusted Kappa coefficients of final Sub-pixel mapping result rose only by3.7%, number of Iterations reduced and the CA running time accounted for only25.8%of the original. The result showed that the ISPCA can significantly improve the running efficiency. In addition, a clear termination condition was made in the improved CA method, which made it more practical application value.Then we put cyanobacterial blooms proportion as the input data and derived the final result of distribution of cyanobacterial blooms at sub-pixels scale. ISPCA simulation results has abundance spatial details compared with the results of using index extraction. ISPCA derived the best result with Kappa coefficient of0.8030at100m resolution which is more close to the observations of higher resolution satellite-MODIS and HJ-1A/1B.In conclusion, the three parts of this study formed the technical framework of extraction of cyanobacterial blooms at sub-pixel scale:analysis of scale difference-scaling down of cyanobacterial blooms proportion-sub-pixel mapping.