Research On Remote Sensing Image Green Tide Monitoring Based On Data Mining

Since 2007,large-scale green tide disasters have erupted in the Yellow Sea and the East China Sea for 11 consecutive years,which not only have an impact on the economic,maritime transportation,tourism and other activities of coastal cities,but also have caused serious harm to the marine ecological environment.In view of the large coverage area and the wide distribution range of green tide outburst,satellite remote sensing has become an important means of green tide detection by virtue of its wide horizon,fast data collection and continuous observation.However,remote sensing satellite data used in monitoring,due to factors such as resolution and clouds,have led to low accuracy of green tide detection methods(NDVI,FAI,EVI,etc.).Therefore,this paper takes the Yellow Sea area as research area,comprehensive using GF-1 PMS,Landsat 8 and MODIS remote sensing images to monitor green tide of the Yellow Sea in 2014-2018,and some related researches are conducted on this basis.The main research contents are listed as follows:(1)Based on clustering and classification algorithms of data mining,the common types of cloud coverage are analyzed,and then the linear relationship between the threshold(which is the value can distinguish green tide and seawater)and the spectral difference of images under different cloud coverage condition is analyzed.An automatic detection method of green tide zoning adaptive threshold for high resolution image is proposed.(2)The monitoring capability of multi-source and multi-resolution remote sensing images is analyzed,then,combined with the monitoring results of different data sources,the interannual variation characteristics of green tide coverage area of the Yellow Sea area in 2014-2018 are analyzed.(3)Dynamic monitoring and exploring the distribution of green tide outbreak whole process in the Yellow Sea from 2014 to 2018,and further analyzing the growth process of green tide from the perspective of remote sensing.The results show that:(1)Green tide adaptive threshold partition automatic detection method for high resolution remote sensing image proposed in this paper automatically selects the threshold for characteristics of different cloud coverage areas,the extraction accuracy of green tide information is obviously superior to the traditional threshold detection method.It avoids the manual participation steps of interpreters and achieves the full automatic detection of green tide.(2)Combining the adaptive threshold partition automatic detection method proposed in this paper and the normalized vegetation index method(NDVI)to analyze green tide monitoring abilities of synchronized images of GF-1 PMS,Landsat 8 and MODIS with different resolutions.The results show that the monitoring results of Landsat 8 images have a deviation of9.39%~32.70%compared with GF-1 PMS images,and the monitoring results of MODIS images have a deviation of 529.91%~1456.7%compared with GF-1 PMS images.In the past five years,the coverage area of green tide has increased first and then decreased.The maximum coverage area of green tide reached 611.96 km~2 in 2015,and then decreased year by year,the maximum coverage area of green tide in 2018 was 214.34km~2.(3)Based on the validation of monitoring results from various data sources,the dynamic monitoring of green tide outbreak process in 2014-2018 are carried out.It is found that green tide gradually drifted northeastward and landed on the coast of Shandong Peninsula since it first appeared near the North Jiangsu Shoal,but the distribution area of green tide when they were first monitored increased year by year.The first appearance time and the extinction time of green tide are different in recent five years,and the growth periods of green tide are different.It is concluded that the average daily growth rate and the average daily decline rate of green tide are not only related to the coverage area of green tide,but also related to the growth time and environment of green tide.