CONSTRUCTION OF GREEN TIDE MONITORING SYSTEM AND RESEARCH ON ITS KEY TECHNIQUES

As a kind of marine natural disaster, Green Tide has been appearing every year along the Qingdao Coast, bringing great loss to this region, since the large-scale bloom in 2008. Therefore, it is of great value to obtain the real time dynamic information about green tide distribution. In this study, methods of optical remote sensing and microwave remote sensing are employed in Green Tide Monitoring Research. A specific remote sensing data processing flow and a green tide information extraction algorithm are designed, according to the optical and microwave data of different characteristics. In the aspect of green tide spatial distribution information extraction, an automatic extraction algorithm of green tide distribution boundaries is designed based on the principle of mathematical morphology dilation/erosion. And key issues in information extraction, including the division of green tide regions, the obtaining of basic distributions, the limitation of distribution boundary, and the elimination of islands, have been solved. The automatic generation of green tide distribution boundaries from the results of remote sensing information extraction is realized. Finally, a green tide monitoring system is built based on IDL/GIS secondary development in the integrated environment of RS and GIS, achieving the integration of RS monitoring and information extraction.


INTRODUCTION
Green Tide is a kind of marine ecological anomaly occurring generally in coastal areas around the world.
It is an abnormal ecological phenomenon caused by large scale proliferation of Ulva and Enteromorpha and other kinds of algae (Hiraoka et al., 2004;Nelson et al., 2003).In recent years, the Yellow Sea, especially the coastal areas from Jiangsu Province to Yantai, Shandong Province, Chi-na, have been repeatedly disturbed by large-scale green tides (Hu et al., 2008;Liu et al., 2016).The ecological environment there is greatly damaged and the seafood industry suffers great economical losses when green tide breaks out (Liu et analysis and visualization with geodata, which makes it effective in managing huge amount of multi-period, multi-source remote sensing data in different formats and extracting the distribution information and its contour of green tide through spatial analysis.However, the present-used green tide monitoring method that combines remote sensing and GIS is complicated.
Meanwhile the level of automation is low and the method relies a lot on professional platform and manual operation, resulting in low work efficiency.There lacks research on an automatic, integrated system which is needed in green tide monitoring business.
Therefore, in this study, a green tide monitoring system where remote sensing data and algorithms are integrated into GIS environment, is designed and developed to greatly improve the efficiency in the procedure from data processing to product making.

FRAMEWORK OF THE SYSTEM
The green tide monitoring system utilizes ENVI/ IDL , which is professional in image processing, to create the remote sensing processing modules (Kong et al., 2013;Wang et al., 2011;Wang, 2011) products are made with the support of GIS for its outstanding ability of data display and map-making (Dekker et al., 1993;Wang, 2010).Basic process is shown in Fig. 1.These two main questions are focused on in this paper: (1) Integration of GIS and remote sensing In the traditional methods to extract distribution information from multi-source remote sensing data, a series of pre-processing operations are needed first in professional remote sensing software, then algorithms are designed for coverage information extraction.After that the results are transferred into GIS environment to conduct spatial analysis with to obtain the green tide distribution information and display it.
(2) Green tide distribution contour extraction The contour of the green tide is needed because there is no accurate green tide distribution contour information in the bi-value images extracted by green tide coverage information extracting algorithm and area of distribution cannot be calculated through this kind of images.Euclidean distance analysis and mathematical morphology closed operation(dilation/erosion) are combined to obtain accurate enveloping polygons of green tide coverage points.

SENSING
The basic framework of green tide monitoring system is to extract green tide coverage information by remote sensing information extracting module and to achieve the data organization & display and product making based on GIS.Remote sensing processing modules are embedded in GIS environment and the information interaction between remote sensing processing and GIS analysis is achieved.In order to realize better fusion of the two environments and data interaction, the following are done: (1) Packaging and calling of remote sensing image processing modules: As the loose integration based on executable program(EXE) has the advantages of cross platform, flexibility, convenience and portability, image processing module sealed as EXE is more conducive to integration with GIS environment and data interaction.
Loose integration way is adopted to carry on secondary development and remote sensing image processing modules are sealed to obtain the corresponding EXEs.
Information about these image processing modules is summarized in Table 1.In the system, the remote sensing image processing module EXEs are invoked by the GIS platform through configuration files.
Table 1 Statistics for Green Tide Remote Sensing Image Processing Modules The calling process of the modules above in the system is shown (2) Unified data format: All process data in this paper are stored in a uniform data storage format and uniform naming rules.All raster data is stored in GeoTIFF format and all vector data is stored in Shapefile format.
File naming rules are unified as 'type of original data_12 bit time string_type of result_resolution_processing time', for example, 'modis_201510151605_ndvi_250_1612.tif'.
(3) Unified image property standard: The Nodata value in the image preprocessing result is set to 0 because Nodata values of images are uniformly assigned to special numbers, like -9999, in ENVI, while invalid values of images default to positive infinity, negative infinity or 0 in ArcGIS.
(4) Unified management methods: Due to the large quantities of raster and vector data generated in image processing and in the process of green tide distribution information extraction, therefore one separate folder is created as workspace where relative data are stored every day.
Unnecessary data shall not be preserved.

GREEN TIDE DISTRIBUTION CONTOUR EXTRACTION ALGORITHM
The bi-value images extracted from green tide coverage information, and the vector point data transformed from bi-value images are discrete information of green tide, as is shown in Fig. 3, where there is no division and distribution information of green tide blocks for operational green tide remote sensing monitoring system.The two methods both have their own advantages and disadvantages here.The green tide distribution obtained by the buffer analysis method is smooth and accurate, however, the detailed description of this method leads to excessive computational complexity.Mathematical morphology closed operation method is able to process massive green tide data, but the obtained contour is abrupt and does not accurately describe the boundaries of the green tide distribution.
An attempt is made in this study to explore a green tide distribution contour extraction method that not only combines the advantages of both buffer analysis method and closed operation method, but also avoid their disadvantages, in order to ensure it applicable to actual business applications.
Here, the Euclidean distance analysis is used as the dilation process of the closed operation, and the inner buffer is used as the corrosion process of it.As is shown in Fig. 7: the green tide coverage points are used as source data (as the green points shown in the figure).Firstly, Euclidean distance analysis is performed as "expansion", where the key parameter, "the maximum distance" (the yellow radiations in Fig. 7), is specified as the fusion distance, to obtain a gray-scale image like a fused overlay buffer fusing nearest neighbor related green tide coverage information in the same block.After that, condition analysis is performed on the gray-scale image to obtain the bivalue image and then it is transformed into a vector file through The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3, 2018 ISPRS TC III Mid-term Symposium "Developments, Technologies and Applications in Remote Sensing", 7-10 May, Beijing, China raster-vector conversion and the block information (as the red border area shown in Fig. 7) is extracted.Finally, the inward buffer (buffer distance is the negative value of fusion distance) is made for the resulting block to operate "corrosion" (as the purple radiations in Fig. 7) and to obtain the basic distribution of green tide (as the blue border area in Fig. 7).And the buffer for the basic distribution is made with the extrapolation distance as buffer distance to obtain the basic distribution after extrapolation.

Algorithm Implementation
The whole process of green tide distribution contour extraction algorithm is shown in figure 8.In order to avoid isolated green tide point information missing in the process of inward buffering due to the absolute value of buffer distance equal to fusion distance.Firstly, the green tide blocks are divided, and then the basic distribution contours are obtained according to the types of the division results.
(1) Data input and parameter setting The bi-value images or discrete vector points obtained from green tide coverage information extraction are read, and the constraint files like land files, cloud files are imported.The key parameters involved in the algorithm are set.
(2) Green tide block partition Distance analysis: the cluster analysis tool based on Euclidean distance analysis is used to measure the straight-line distance between each pixel center and its nearest green tide pixel center.
The green tide vector point data or bi-value images were entered.
The output grid pixel size, cell_size, is set and the maximum distance is specified as the fusion distance.The result of the operation is the gray image whose gray level represents the distance value where the closer the distance from the green tide point is, the bigger the gray value is, the darker t、e color is, and vice versa; Condition analysis: in the gray-scale image obtained in distance analysis, the areas represented by gray scale are the divided blocks, whose gray values of pixels are all greater than or equal to 0, while the rest of the areas that are transparent, are not in the blocks, where the gray values of pixels are all NoData  NoData as default.The bi-value image that represents green tide blocks division is obtained;Raster-vector conversion: the rastervector tool is used to covert the bi-value image representing green tide block partition into vector polygon, so as to operate buffer analysis;Inward buffering: After distance analysis, green tide discrete points have been expanding a fusion distance (FDis).
At that time, make an inward buffer with 1/2cell_size-FDis, whose absolute value is a little smaller than fusion distance, as buffer distance to make sure that all the green tide information be covered in blocks and to avoid the problem that isolated green tide information cannot be extracted.The result is used to distinguish between green tide accumulation areas and isolated points.
(3) Basic distribution obtaining For green tide accumulation areas, the operation in the previous step is repeated on the discrete points in each block.However, the difference is that after the raster-vector conversion, an inward buffer is directly made with the opposite number of fusion distance as buffer distance, then an outward buffer is made with extrapolation distance as buffer distance and fuse all the blocks by the way of merging polygon layers to obtain the basic distribution of green tide accumulation areas.While for isolated green tide information, only an outward buffer with the extrapolation distance as buffer distance is needed to obtain the basic distribution here.
(4) Obtaining of accurate distribution contour of green tide In order to meet the spatial constraints and business needs in practical application, land files and cloud files are used as mask to operate mask processing on the obtained contour of basic green tide distribution, and the features such as "islands" and "rings" are eliminated by the way of dissolving the polygons.
Finally, accurate extraction of green tide distribution contour is accomplished and the vector files of green tide distribution that meet business need are obtained.

Algorithm Verification
As is shown in .The system sets ArcGIS10.1 as its GIS environment and realizes the extraction of green tide distribution information and monitoring results display based on ArcEngine10.1 and C#.Two kinds of optical remote sensing data (MODIS, HJ-1 CCD) and three kinds of microwave remote sensing data (COSMO, RADARSAT, Terra SAR-X) are utilized comprehensively in order to realize largescale, high-precise, all-weather green tide monitoring in this study.And remote sensing data preprocessing algorithms for different types of data and green tide coverage information extraction algorithm are designed and sealed as independent modules.These modules are called to obtain the bi-value images that contain green tide coverage information under the GIS environment (Wang et al., 2014).Then Euclidean Distance Analysis and Closing Mathematical Morphology-Dilation/Erosion are combined to extract the spatial distribution information from the green tide coverage bi-value images.At last, specific

Figure. 1
Figure.1 Flow Chart of Green Tide Monitoring System

in Figure 2 .
Figure.2The flow chart of multi-source remote sensing data preprocessing and green tide coverage information extraction

Figure 3 .
Figure 3.The Green Tide Coverage Information

Figure. 4
Figure. 4 The Effect Diagram of Direct Extraction Method

Figure. 5
Figure. 5 The Sketch Map of Block Division

Figure. 7
Figure.7 The Sketch Map Green Tide Distribution Contour Extraction Algorithm

Figure. 8
Figure. 8 Flow Chart of Green Tide Distribution Contour Extraction Figure. 9 The Verification of Green Tide Distribution Contour Extraction Algorithm

Bi-value Image Discrete Vector Points Basic Distribution Contour Obtaining Green Tide Accumulation Areas Green Tide Isolated Points Outwar d Buffer ing Distance Analysis Layer Dissolve Condition Analysis Raster-Vector Conver sion Inward Buffer ing Distance Analysis Condi ti on Analysis Raster- Vector Conver sion Inward Buffer ing Outwar d Buffer ing Mask Process Elimination of Isl ands Accurate Contour
Data Input and Parameter Setting